asl_recognizer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Artificial Intelligence Engineer Nanodegree - Probabilistic Models\n",
    "## Project: Sign Language Recognition System\n",
    "- [Introduction](#intro)\n",
    "- [Part 1 Feature Selection](#part1_tutorial)\n",
    "    - [Tutorial](#part1_tutorial)\n",
    "    - [Features Submission](#part1_submission)\n",
    "    - [Features Unittest](#part1_test)\n",
    "- [Part 2 Train the models](#part2_tutorial)\n",
    "    - [Tutorial](#part2_tutorial)\n",
    "    - [Model Selection Score Submission](#part2_submission)\n",
    "    - [Model Score Unittest](#part2_test)\n",
    "- [Part 3 Build a Recognizer](#part3_tutorial)\n",
    "    - [Tutorial](#part3_tutorial)\n",
    "    - [Recognizer Submission](#part3_submission)\n",
    "    - [Recognizer Unittest](#part3_test)\n",
    "- [Part 4 (OPTIONAL) Improve the WER with Language Models](#part4_info)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='intro'></a>\n",
    "## Introduction\n",
    "The overall goal of this project is to build a word recognizer for American Sign Language video sequences, demonstrating the power of probabalistic models.  In particular, this project employs  [hidden Markov models (HMM's)](https://en.wikipedia.org/wiki/Hidden_Markov_model) to analyze a series of measurements taken from videos of American Sign Language (ASL) collected for research (see the [RWTH-BOSTON-104 Database](http://www-i6.informatik.rwth-aachen.de/~dreuw/database-rwth-boston-104.php)).  In this video, the right-hand x and y locations are plotted as the speaker signs the sentence.\n",
    "[![ASLR demo](http://www-i6.informatik.rwth-aachen.de/~dreuw/images/demosample.png)](https://drive.google.com/open?id=0B_5qGuFe-wbhUXRuVnNZVnMtam8)\n",
    "\n",
    "The raw data, train, and test sets are pre-defined.  You will derive a variety of feature sets (explored in Part 1), as well as implement three different model selection criterion to determine the optimal number of hidden states for each word model (explored in Part 2). Finally, in Part 3 you will implement the recognizer and compare the effects the different combinations of feature sets and model selection criteria.  \n",
    "\n",
    "At the end of each Part, complete the submission cells with implementations, answer all questions, and pass the unit tests.  Then submit the completed notebook for review!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part1_tutorial'></a>\n",
    "## PART 1: Data\n",
    "\n",
    "### Features Tutorial\n",
    "##### Load the initial database\n",
    "A data handler designed for this database is provided in the student codebase as the `AslDb` class in the `asl_data` module.  This handler creates the initial [pandas](http://pandas.pydata.org/pandas-docs/stable/) dataframe from the corpus of data included in the `data` directory as well as dictionaries suitable for extracting data in a format friendly to the [hmmlearn](https://hmmlearn.readthedocs.io/en/latest/) library.  We'll use those to create models in Part 2.\n",
    "\n",
    "To start, let's set up the initial database and select an example set of features for the training set.  At the end of Part 1, you will create additional feature sets for experimentation. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>left-x</th>\n",
       "      <th>left-y</th>\n",
       "      <th>right-x</th>\n",
       "      <th>right-y</th>\n",
       "      <th>nose-x</th>\n",
       "      <th>nose-y</th>\n",
       "      <th>speaker</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>video</th>\n",
       "      <th>frame</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">98</th>\n",
       "      <th>0</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             left-x  left-y  right-x  right-y  nose-x  nose-y  speaker\n",
       "video frame                                                           \n",
       "98    0         149     181      170      175     161      62  woman-1\n",
       "      1         149     181      170      175     161      62  woman-1\n",
       "      2         149     181      170      175     161      62  woman-1\n",
       "      3         149     181      170      175     161      62  woman-1\n",
       "      4         149     181      170      175     161      62  woman-1"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from asl_data import AslDb\n",
    "\n",
    "\n",
    "asl = AslDb() # initializes the database\n",
    "asl.df.head() # displays the first five rows of the asl database, indexed by video and frame"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "left-x         149\n",
       "left-y         181\n",
       "right-x        170\n",
       "right-y        175\n",
       "nose-x         161\n",
       "nose-y          62\n",
       "speaker    woman-1\n",
       "Name: (98, 1), dtype: object"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "asl.df.ix[98,1]  # look at the data available for an individual frame"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "The frame represented by video 98, frame 1 is shown here:\n",
    "![Video 98](http://www-i6.informatik.rwth-aachen.de/~dreuw/database/rwth-boston-104/overview/images/orig/098-start.jpg)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Feature selection for training the model\n",
    "The objective of feature selection when training a model is to choose the most relevant variables while keeping the model as simple as possible, thus reducing training time.  We can use the raw features already provided or derive our own and add columns to the pandas dataframe `asl.df` for selection. As an example, in the next cell a feature named `'grnd-ry'` is added. This feature is the difference between the right-hand y value and the nose y value, which serves as the \"ground\" right y value. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>left-x</th>\n",
       "      <th>left-y</th>\n",
       "      <th>right-x</th>\n",
       "      <th>right-y</th>\n",
       "      <th>nose-x</th>\n",
       "      <th>nose-y</th>\n",
       "      <th>speaker</th>\n",
       "      <th>grnd-ry</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>video</th>\n",
       "      <th>frame</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">98</th>\n",
       "      <th>0</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             left-x  left-y  right-x  right-y  nose-x  nose-y  speaker  \\\n",
       "video frame                                                              \n",
       "98    0         149     181      170      175     161      62  woman-1   \n",
       "      1         149     181      170      175     161      62  woman-1   \n",
       "      2         149     181      170      175     161      62  woman-1   \n",
       "      3         149     181      170      175     161      62  woman-1   \n",
       "      4         149     181      170      175     161      62  woman-1   \n",
       "\n",
       "             grnd-ry  \n",
       "video frame           \n",
       "98    0          113  \n",
       "      1          113  \n",
       "      2          113  \n",
       "      3          113  \n",
       "      4          113  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "asl.df['grnd-ry'] = asl.df['right-y'] - asl.df['nose-y']\n",
    "asl.df.head()  # the new feature 'grnd-ry' is now in the frames dictionary"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Other ground features:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "asl.df sample\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>left-x</th>\n",
       "      <th>left-y</th>\n",
       "      <th>right-x</th>\n",
       "      <th>right-y</th>\n",
       "      <th>nose-x</th>\n",
       "      <th>nose-y</th>\n",
       "      <th>speaker</th>\n",
       "      <th>grnd-ry</th>\n",
       "      <th>grnd-rx</th>\n",
       "      <th>grnd-ly</th>\n",
       "      <th>grnd-lx</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>video</th>\n",
       "      <th>frame</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">98</th>\n",
       "      <th>0</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             left-x  left-y  right-x  right-y  nose-x  nose-y  speaker  \\\n",
       "video frame                                                              \n",
       "98    0         149     181      170      175     161      62  woman-1   \n",
       "      1         149     181      170      175     161      62  woman-1   \n",
       "      2         149     181      170      175     161      62  woman-1   \n",
       "      3         149     181      170      175     161      62  woman-1   \n",
       "      4         149     181      170      175     161      62  woman-1   \n",
       "\n",
       "             grnd-ry  grnd-rx  grnd-ly  grnd-lx  \n",
       "video frame                                      \n",
       "98    0          113        9      119      -12  \n",
       "      1          113        9      119      -12  \n",
       "      2          113        9      119      -12  \n",
       "      3          113        9      119      -12  \n",
       "      4          113        9      119      -12  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<font color=green>Correct!</font><br/>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from asl_utils import test_features_tryit\n",
    "asl.df['grnd-rx'] = asl.df['right-x'] - asl.df['nose-x']\n",
    "asl.df['grnd-ly'] = asl.df['left-y'] - asl.df['nose-y']\n",
    "asl.df['grnd-lx'] = asl.df['left-x'] - asl.df['nose-x']\n",
    "\n",
    "# test the code\n",
    "test_features_tryit(asl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[9, 113, -12, 119]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# collect the features into a list\n",
    "features_ground = ['grnd-rx','grnd-ry','grnd-lx','grnd-ly']\n",
    " #show a single set of features for a given (video, frame) tuple\n",
    "[asl.df.ix[98,1][v] for v in features_ground]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Build the training set\n",
    "Now that we have a feature list defined, we can pass that list to the `build_training` method to collect the features for all the words in the training set.  Each word in the training set has multiple examples from various videos.  Below we can see the unique words that have been loaded into the training set:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training words: ['JOHN', 'WRITE', 'HOMEWORK', 'IX-1P', 'SEE', 'YESTERDAY', 'IX', 'LOVE', 'MARY', 'CAN', 'GO', 'GO1', 'FUTURE', 'GO2', 'PARTY', 'FUTURE1', 'HIT', 'BLAME', 'FRED', 'FISH', 'WONT', 'EAT', 'BUT', 'CHICKEN', 'VEGETABLE', 'CHINA', 'PEOPLE', 'PREFER', 'BROCCOLI', 'LIKE', 'LEAVE', 'SAY', 'BUY', 'HOUSE', 'KNOW', 'CORN', 'CORN1', 'THINK', 'NOT', 'PAST', 'LIVE', 'CHICAGO', 'CAR', 'SHOULD', 'DECIDE', 'VISIT', 'MOVIE', 'WANT', 'SELL', 'TOMORROW', 'NEXT-WEEK', 'NEW-YORK', 'LAST-WEEK', 'WILL', 'FINISH', 'ANN', 'READ', 'BOOK', 'CHOCOLATE', 'FIND', 'SOMETHING-ONE', 'POSS', 'BROTHER', 'ARRIVE', 'HERE', 'GIVE', 'MAN', 'NEW', 'COAT', 'WOMAN', 'GIVE1', 'HAVE', 'FRANK', 'BREAK-DOWN', 'SEARCH-FOR', 'WHO', 'WHAT', 'LEG', 'FRIEND', 'CANDY', 'BLUE', 'SUE', 'BUY1', 'STOLEN', 'OLD', 'STUDENT', 'VIDEOTAPE', 'BORROW', 'MOTHER', 'POTATO', 'TELL', 'BILL', 'THROW', 'APPLE', 'NAME', 'SHOOT', 'SAY-1P', 'SELF', 'GROUP', 'JANA', 'TOY1', 'MANY', 'TOY', 'ALL', 'BOY', 'TEACHER', 'GIRL', 'BOX', 'GIVE2', 'GIVE3', 'GET', 'PUTASIDE']\n"
     ]
    }
   ],
   "source": [
    "training = asl.build_training(features_ground)\n",
    "print(\"Training words: {}\".format(training.words))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "The training data in `training` is an object of class `WordsData` defined in the `asl_data` module.  in addition to the `words` list, data can be accessed with the `get_all_sequences`, `get_all_Xlengths`, `get_word_sequences`, and `get_word_Xlengths` methods. We need the `get_word_Xlengths` method to train multiple sequences with the `hmmlearn` library.  In the following example, notice that there are two lists; the first is a concatenation of all the sequences(the X portion) and the second is a list of the sequence lengths(the Lengths portion)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([[-11,  48,   7, 120],\n",
       "        [-11,  48,   8, 109],\n",
       "        [ -8,  49,  11,  98],\n",
       "        [ -7,  50,   7,  87],\n",
       "        [ -4,  54,   7,  77],\n",
       "        [ -4,  54,   6,  69],\n",
       "        [ -4,  54,   6,  69],\n",
       "        [-13,  52,   6,  69],\n",
       "        [-13,  52,   6,  69],\n",
       "        [ -8,  51,   6,  69],\n",
       "        [ -8,  51,   6,  69],\n",
       "        [ -8,  51,   6,  69],\n",
       "        [ -8,  51,   6,  69],\n",
       "        [ -8,  51,   6,  69],\n",
       "        [-10,  59,   7,  71],\n",
       "        [-15,  64,   9,  77],\n",
       "        [-17,  75,  13,  81],\n",
       "        [ -4,  48,  -4, 113],\n",
       "        [ -2,  53,  -4, 113],\n",
       "        [ -4,  55,   2,  98],\n",
       "        [ -4,  58,   2,  98],\n",
       "        [ -1,  59,   2,  89],\n",
       "        [ -1,  59,  -1,  84],\n",
       "        [ -1,  59,  -1,  84],\n",
       "        [ -7,  63,  -1,  84],\n",
       "        [ -7,  63,  -1,  84],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -7,  63,   3,  83],\n",
       "        [ -4,  70,   3,  83],\n",
       "        [ -4,  70,   3,  83],\n",
       "        [ -2,  73,   5,  90],\n",
       "        [ -3,  79,  -4,  96],\n",
       "        [-15,  98,  13, 135],\n",
       "        [ -6,  93,  12, 128],\n",
       "        [ -2,  89,  14, 118],\n",
       "        [  5,  90,  10, 108],\n",
       "        [  4,  86,   7, 105],\n",
       "        [  4,  86,   7, 105],\n",
       "        [  4,  86,  13, 100],\n",
       "        [ -3,  82,  14,  96],\n",
       "        [ -3,  82,  14,  96],\n",
       "        [  6,  89,  16, 100],\n",
       "        [  6,  89,  16, 100],\n",
       "        [  7,  85,  17, 111]], dtype=int64), [17, 20, 12])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "training.get_word_Xlengths('CHOCOLATE')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "###### More feature sets\n",
    "So far we have a simple feature set that is enough to get started modeling.  However, we might get better results if we manipulate the raw values a bit more, so we will go ahead and set up some other options now for experimentation later.  For example, we could normalize each speaker's range of motion with grouped statistics using [Pandas stats](http://pandas.pydata.org/pandas-docs/stable/api.html#api-dataframe-stats) functions and [pandas groupby](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.groupby.html).  Below is an example for finding the means of all speaker subgroups."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true,
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>left-x</th>\n",
       "      <th>left-y</th>\n",
       "      <th>right-x</th>\n",
       "      <th>right-y</th>\n",
       "      <th>nose-x</th>\n",
       "      <th>nose-y</th>\n",
       "      <th>grnd-ry</th>\n",
       "      <th>grnd-rx</th>\n",
       "      <th>grnd-ly</th>\n",
       "      <th>grnd-lx</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>speaker</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>man-1</th>\n",
       "      <td>206.248203</td>\n",
       "      <td>218.679449</td>\n",
       "      <td>155.464350</td>\n",
       "      <td>150.371031</td>\n",
       "      <td>175.031756</td>\n",
       "      <td>61.642600</td>\n",
       "      <td>88.728430</td>\n",
       "      <td>-19.567406</td>\n",
       "      <td>157.036848</td>\n",
       "      <td>31.216447</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>woman-1</th>\n",
       "      <td>164.661438</td>\n",
       "      <td>161.271242</td>\n",
       "      <td>151.017865</td>\n",
       "      <td>117.332462</td>\n",
       "      <td>162.655120</td>\n",
       "      <td>57.245098</td>\n",
       "      <td>60.087364</td>\n",
       "      <td>-11.637255</td>\n",
       "      <td>104.026144</td>\n",
       "      <td>2.006318</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>woman-2</th>\n",
       "      <td>183.214509</td>\n",
       "      <td>176.527232</td>\n",
       "      <td>156.866295</td>\n",
       "      <td>119.835714</td>\n",
       "      <td>170.318973</td>\n",
       "      <td>58.022098</td>\n",
       "      <td>61.813616</td>\n",
       "      <td>-13.452679</td>\n",
       "      <td>118.505134</td>\n",
       "      <td>12.895536</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             left-x      left-y     right-x     right-y      nose-x  \\\n",
       "speaker                                                               \n",
       "man-1    206.248203  218.679449  155.464350  150.371031  175.031756   \n",
       "woman-1  164.661438  161.271242  151.017865  117.332462  162.655120   \n",
       "woman-2  183.214509  176.527232  156.866295  119.835714  170.318973   \n",
       "\n",
       "            nose-y    grnd-ry    grnd-rx     grnd-ly    grnd-lx  \n",
       "speaker                                                          \n",
       "man-1    61.642600  88.728430 -19.567406  157.036848  31.216447  \n",
       "woman-1  57.245098  60.087364 -11.637255  104.026144   2.006318  \n",
       "woman-2  58.022098  61.813616 -13.452679  118.505134  12.895536  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_means = asl.df.groupby('speaker').mean()\n",
    "df_means"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "To select a mean that matches by speaker, use the pandas [map](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.Series.map.html) method:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>left-x</th>\n",
       "      <th>left-y</th>\n",
       "      <th>right-x</th>\n",
       "      <th>right-y</th>\n",
       "      <th>nose-x</th>\n",
       "      <th>nose-y</th>\n",
       "      <th>speaker</th>\n",
       "      <th>grnd-ry</th>\n",
       "      <th>grnd-rx</th>\n",
       "      <th>grnd-ly</th>\n",
       "      <th>grnd-lx</th>\n",
       "      <th>left-x-mean</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>video</th>\n",
       "      <th>frame</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">98</th>\n",
       "      <th>0</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "      <td>164.661438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "      <td>164.661438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "      <td>164.661438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "      <td>164.661438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>149</td>\n",
       "      <td>181</td>\n",
       "      <td>170</td>\n",
       "      <td>175</td>\n",
       "      <td>161</td>\n",
       "      <td>62</td>\n",
       "      <td>woman-1</td>\n",
       "      <td>113</td>\n",
       "      <td>9</td>\n",
       "      <td>119</td>\n",
       "      <td>-12</td>\n",
       "      <td>164.661438</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             left-x  left-y  right-x  right-y  nose-x  nose-y  speaker  \\\n",
       "video frame                                                              \n",
       "98    0         149     181      170      175     161      62  woman-1   \n",
       "      1         149     181      170      175     161      62  woman-1   \n",
       "      2         149     181      170      175     161      62  woman-1   \n",
       "      3         149     181      170      175     161      62  woman-1   \n",
       "      4         149     181      170      175     161      62  woman-1   \n",
       "\n",
       "             grnd-ry  grnd-rx  grnd-ly  grnd-lx  left-x-mean  \n",
       "video frame                                                   \n",
       "98    0          113        9      119      -12   164.661438  \n",
       "      1          113        9      119      -12   164.661438  \n",
       "      2          113        9      119      -12   164.661438  \n",
       "      3          113        9      119      -12   164.661438  \n",
       "      4          113        9      119      -12   164.661438  "
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "asl.df['left-x-mean']= asl.df['speaker'].map(df_means['left-x'])\n",
    "asl.df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Std deviation by speaker:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df_std\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>left-x</th>\n",
       "      <th>left-y</th>\n",
       "      <th>right-x</th>\n",
       "      <th>right-y</th>\n",
       "      <th>nose-x</th>\n",
       "      <th>nose-y</th>\n",
       "      <th>grnd-ry</th>\n",
       "      <th>grnd-rx</th>\n",
       "      <th>grnd-ly</th>\n",
       "      <th>grnd-lx</th>\n",
       "      <th>left-x-mean</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>speaker</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>man-1</th>\n",
       "      <td>15.154425</td>\n",
       "      <td>36.328485</td>\n",
       "      <td>18.901917</td>\n",
       "      <td>54.902340</td>\n",
       "      <td>6.654573</td>\n",
       "      <td>5.520045</td>\n",
       "      <td>53.487999</td>\n",
       "      <td>20.269032</td>\n",
       "      <td>36.572749</td>\n",
       "      <td>15.080360</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>woman-1</th>\n",
       "      <td>17.573442</td>\n",
       "      <td>26.594521</td>\n",
       "      <td>16.459943</td>\n",
       "      <td>34.667787</td>\n",
       "      <td>3.549392</td>\n",
       "      <td>3.538330</td>\n",
       "      <td>33.972660</td>\n",
       "      <td>16.764706</td>\n",
       "      <td>27.117393</td>\n",
       "      <td>17.328941</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>woman-2</th>\n",
       "      <td>15.388711</td>\n",
       "      <td>28.825025</td>\n",
       "      <td>14.890288</td>\n",
       "      <td>39.649111</td>\n",
       "      <td>4.099760</td>\n",
       "      <td>3.416167</td>\n",
       "      <td>39.128572</td>\n",
       "      <td>16.191324</td>\n",
       "      <td>29.320655</td>\n",
       "      <td>15.050938</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            left-x     left-y    right-x    right-y    nose-x    nose-y  \\\n",
       "speaker                                                                   \n",
       "man-1    15.154425  36.328485  18.901917  54.902340  6.654573  5.520045   \n",
       "woman-1  17.573442  26.594521  16.459943  34.667787  3.549392  3.538330   \n",
       "woman-2  15.388711  28.825025  14.890288  39.649111  4.099760  3.416167   \n",
       "\n",
       "           grnd-ry    grnd-rx    grnd-ly    grnd-lx  left-x-mean  \n",
       "speaker                                                           \n",
       "man-1    53.487999  20.269032  36.572749  15.080360          0.0  \n",
       "woman-1  33.972660  16.764706  27.117393  17.328941          0.0  \n",
       "woman-2  39.128572  16.191324  29.320655  15.050938          0.0  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<font color=green>Correct!</font><br/>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from asl_utils import test_std_tryit\n",
    "df_std = asl.df.groupby('speaker').std()\n",
    "\n",
    "# test the code\n",
    "test_std_tryit(df_std)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part1_submission'></a>\n",
    "### Features Implementation Submission\n",
    "Implement four feature sets and answer the question that follows.\n",
    "- normalized Cartesian coordinates\n",
    "    - use *mean* and *standard deviation* statistics and the [standard score](https://en.wikipedia.org/wiki/Standard_score) equation to account for speakers with different heights and arm length\n",
    "    \n",
    "- polar coordinates\n",
    "    - calculate polar coordinates with [Cartesian to polar equations](https://en.wikipedia.org/wiki/Polar_coordinate_system#Converting_between_polar_and_Cartesian_coordinates)\n",
    "    - use the [np.arctan2](https://docs.scipy.org/doc/numpy-1.10.0/reference/generated/numpy.arctan2.html) function and *swap the x and y axes* to move the $0$ to $2\\pi$ discontinuity to 12 o'clock instead of 3 o'clock;  in other words, the normal break in radians value from $0$ to $2\\pi$ occurs directly to the left of the speaker's nose, which may be in the signing area and interfere with results.  By swapping the x and y axes, that discontinuity move to directly above the speaker's head, an area not generally used in signing.\n",
    "\n",
    "- delta difference\n",
    "    - as described in Thad's lecture, use the difference in values between one frame and the next frames as features\n",
    "    - pandas [diff method](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.diff.html) and [fillna method](http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.fillna.html) will be helpful for this one\n",
    "\n",
    "- custom features\n",
    "    - These are your own design; combine techniques used above or come up with something else entirely. We look forward to seeing what you come up with! \n",
    "    Some ideas to get you started:\n",
    "        - normalize using a [feature scaling equation](https://en.wikipedia.org/wiki/Feature_scaling)\n",
    "        - normalize the polar coordinates\n",
    "        - adding additional deltas\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Added features for normalized by speaker values of left, right, x, y\n",
    "# Name these 'norm-rx', 'norm-ry', 'norm-lx', and 'norm-ly'\n",
    "# using Z-score scaling (X-Xmean)/Xstd\n",
    "\n",
    "originalFeatures = ['right-x', 'right-y','left-x', 'left-y',  'nose-x', 'nose-y']\n",
    "features_norm = ['norm-rx', 'norm-ry', 'norm-lx','norm-ly']\n",
    "for i in range(4):\n",
    "    asl.df[features_norm[i]] = (asl.df[originalFeatures[i]] - asl.df['speaker'].map(df_means[originalFeatures[i]]))/asl.df['speaker'].map(df_std[originalFeatures[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Added features for polar coordinate values where the nose is the origin\n",
    "# Name these 'polar-rr', 'polar-rtheta', 'polar-lr', and 'polar-ltheta'\n",
    "# Note that 'polar-rr' and 'polar-rtheta' refer to the radius and angle\n",
    "features_polar = ['polar-rr', 'polar-rtheta', 'polar-lr', 'polar-ltheta']\n",
    "\n",
    "asl.df['polar-rtheta'] = np.arctan2(asl.df['grnd-rx'],asl.df['grnd-ry'])\n",
    "asl.df['polar-rr'] =np.sqrt(np.square(asl.df['grnd-ry'])+ np.square(asl.df['grnd-rx']))\n",
    "asl.df['polar-ltheta'] = np.arctan2(asl.df['grnd-lx'],asl.df['grnd-ly'])\n",
    "asl.df['polar-lr'] =np.sqrt(np.square(asl.df['grnd-ly'])+ np.square(asl.df['grnd-lx']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Added features for left, right, x, y differences by one time step, i.e. the \"delta\" values discussed in the lecture\n",
    "# Name these 'delta-rx', 'delta-ry', 'delta-lx', and 'delta-ly'\n",
    "\n",
    "features_delta = ['delta-rx', 'delta-ry', 'delta-lx', 'delta-ly']\n",
    "videos = list(set([i for i in zip(*asl.df.index.values)][0]))\n",
    "\n",
    "delta_frame = asl.df[features_ground].diff(1)\n",
    "delta_frame.loc[[tp for tp in zip(videos,np.zeros(len(videos)))]] = 0\n",
    "asl.df[features_delta]=delta_frame"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Added custom features that are the relative x and y positions of the hands.\n",
    "\n",
    "features_custom = ['diff-x', 'diff-y', 'norm-diff-x', 'norm-diff-y']\n",
    "\n",
    "asl.df['diff-x'] = asl.df['left-x'] - asl.df['right-x']\n",
    "asl.df['diff-y'] = asl.df['left-y'] - asl.df['right-y']\n",
    "\n",
    "#Update the df_means and df_std data frames\n",
    "df_means = asl.df.groupby('speaker').mean()\n",
    "df_std = asl.df.groupby('speaker').std()\n",
    "\n",
    "asl.df['norm-diff-x'] = (asl.df['diff-x'] - asl.df['speaker'].map(df_means['diff-x']))/asl.df['speaker'].map(df_std['diff-x'])\n",
    "asl.df['norm-diff-y'] = (asl.df['diff-y'] - asl.df['speaker'].map(df_means['diff-y']))/asl.df['speaker'].map(df_std['diff-y'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Question 1:**  What custom features did you choose for the features_custom set and why?\n",
    "\n",
    "**Answer 1:** Implemented the difference between the x and y positions of the two hands as the custom features. Also was included the normalized values for these features. These features should help us get a sense of the symmetry of a gesture, and the distance between hands. Both can be valuable information that may be directly lost when working with hand position data relative to nose."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part1_test'></a>\n",
    "### Features Unit Testing\n",
    "Run the following unit tests as a sanity check on the defined \"ground\", \"norm\", \"polar\", and 'delta\"\n",
    "feature sets.  The test simply looks for some valid values but is not exhaustive.  However, the project should not be submitted if these tests don't pass."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "....\n",
      "----------------------------------------------------------------------\n",
      "Ran 4 tests in 0.021s\n",
      "\n",
      "OK\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<unittest.runner.TextTestResult run=4 errors=0 failures=0>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import unittest\n",
    "\n",
    "class TestFeatures(unittest.TestCase):\n",
    "\n",
    "    def test_features_ground(self):\n",
    "        sample = (asl.df.ix[98, 1][features_ground]).tolist()\n",
    "        self.assertEqual(sample, [9, 113, -12, 119])\n",
    "\n",
    "    def test_features_norm(self):\n",
    "        sample = (asl.df.ix[98, 1][features_norm]).tolist()\n",
    "        np.testing.assert_almost_equal(sample, [ 1.153,  1.663, -0.891,  0.742], 3)\n",
    "\n",
    "    def test_features_polar(self):\n",
    "        sample = (asl.df.ix[98,1][features_polar]).tolist()\n",
    "        np.testing.assert_almost_equal(sample, [113.3578, 0.0794, 119.603, -0.1005], 3)\n",
    "\n",
    "    def test_features_delta(self):\n",
    "        sample = (asl.df.ix[98, 0][features_delta]).tolist()\n",
    "        self.assertEqual(sample, [0, 0, 0, 0])\n",
    "        sample = (asl.df.ix[98, 18][features_delta]).tolist()\n",
    "        self.assertTrue(sample in [[-16, -5, -2, 4], [-14, -9, 0, 0]], \"Sample value found was {}\".format(sample))\n",
    "                         \n",
    "suite = unittest.TestLoader().loadTestsFromModule(TestFeatures())\n",
    "unittest.TextTestRunner().run(suite)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part2_tutorial'></a>\n",
    "## PART 2: Model Selection\n",
    "### Model Selection Tutorial\n",
    "The objective of Model Selection is to tune the number of states for each word HMM prior to testing on unseen data.  In this section you will explore three methods: \n",
    "- Log likelihood using cross-validation folds (CV)\n",
    "- Bayesian Information Criterion (BIC)\n",
    "- Discriminative Information Criterion (DIC) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Train a single word\n",
    "Now that we have built a training set with sequence data, we can \"train\" models for each word.  As a simple starting example, we train a single word using Gaussian hidden Markov models (HMM).   By using the `fit` method during training, the [Baum-Welch Expectation-Maximization](https://en.wikipedia.org/wiki/Baum%E2%80%93Welch_algorithm) (EM) algorithm is invoked iteratively to find the best estimate for the model *for the number of hidden states specified* from a group of sample seequences. For this example, we *assume* the correct number of hidden states is 3, but that is just a guess.  How do we know what the \"best\" number of states for training is?  We will need to find some model selection technique to choose the best parameter."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of states trained in model for BOOK is 3\n",
      "logL = -2331.113812743319\n"
     ]
    }
   ],
   "source": [
    "import warnings\n",
    "from hmmlearn.hmm import GaussianHMM\n",
    "\n",
    "def train_a_word(word, num_hidden_states, features):\n",
    "    \n",
    "    warnings.filterwarnings(\"ignore\", category=DeprecationWarning)\n",
    "    training = asl.build_training(features)  \n",
    "    X, lengths = training.get_word_Xlengths(word)\n",
    "    model = GaussianHMM(n_components=num_hidden_states, n_iter=1000).fit(X, lengths)\n",
    "    logL = model.score(X, lengths)\n",
    "    return model, logL\n",
    "\n",
    "demoword = 'BOOK'\n",
    "model, logL = train_a_word(demoword, 3, features_ground)\n",
    "print(\"Number of states trained in model for {} is {}\".format(demoword, model.n_components))\n",
    "print(\"logL = {}\".format(logL))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "The HMM model has been trained and information can be pulled from the model, including means and variances for each feature and hidden state.  The [log likelihood](http://math.stackexchange.com/questions/892832/why-we-consider-log-likelihood-instead-of-likelihood-in-gaussian-distribution) for any individual sample or group of samples can also be calculated with the `score` method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of states trained in model for BOOK is 3\n",
      "hidden state #0\n",
      "mean =  [ -3.46504869  50.66686933  14.02391587  52.04731066]\n",
      "variance =  [ 49.12346305  43.04799144  39.35109609  47.24195772]\n",
      "\n",
      "hidden state #1\n",
      "mean =  [ -11.45300909   94.109178     19.03512475  102.2030162 ]\n",
      "variance =  [  77.403668    203.35441965   26.68898447  156.12444034]\n",
      "\n",
      "hidden state #2\n",
      "mean =  [ -1.12415027  69.44164191  17.02866283  77.7231196 ]\n",
      "variance =  [ 19.70434594  16.83041492  30.51552305  11.03678246]\n",
      "\n"
     ]
    }
   ],
   "source": [
    "def show_model_stats(word, model):\n",
    "    print(\"Number of states trained in model for {} is {}\".format(word, model.n_components))    \n",
    "    variance=np.array([np.diag(model.covars_[i]) for i in range(model.n_components)])    \n",
    "    for i in range(model.n_components):  # for each hidden state\n",
    "        print(\"hidden state #{}\".format(i))\n",
    "        print(\"mean = \", model.means_[i])\n",
    "        print(\"variance = \", variance[i])\n",
    "        print()\n",
    "    \n",
    "show_model_stats(demoword, model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Tried a few features and words!\n",
    "Experiment by changing the feature set, word, and/or num_hidden_states values in the next cell to see changes in values.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of states trained in model for ALL is 3\n",
      "hidden state #0\n",
      "mean =  [-1.70000156  0.80000072  0.          0.        ]\n",
      "variance =  [  7.01100592e+00   1.96100202e+00   1.00000101e-03   1.00000101e-03]\n",
      "\n",
      "hidden state #1\n",
      "mean =  [ 9.79999004 -3.99999591  0.6999993  -1.1999988 ]\n",
      "variance =  [ 36.56106078  14.60100148   7.61099285   8.16099325]\n",
      "\n",
      "hidden state #2\n",
      "mean =  [-12.99999984  -4.33333323  -1.99999996   0.66666665]\n",
      "variance =  [ 18.67000106   2.8922227    0.67000006   1.55888887]\n",
      "\n",
      "logL = -130.9454744807456\n"
     ]
    }
   ],
   "source": [
    "my_testword = 'ALL'\n",
    "model, logL = train_a_word(my_testword, 3, features_delta)\n",
    "show_model_stats(my_testword, model)\n",
    "print(\"logL = {}\".format(logL))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Visualize the hidden states\n",
    "We can plot the means and variances for each state and feature.  Try varying the number of states trained for the HMM model and examine the variances.  Are there some models that are \"better\" than others?  How can you tell?  We would like to hear what you think in the classroom online."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYQAAAEICAYAAABfz4NwAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ns4RgjDEGsIRgjDHG+/8O6vncBBEPzgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x23790985080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX4AAAEICAYAAABYoZ8gAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n9UfRHwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x23790a57e48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEICAYAAACktLTqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nOB45ugAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x23790b9e6a0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEICAYAAACktLTqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n3A3DMFohZu6GYRitEDN3wzCMVsj/B9fmipMLOTNGAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x23790d6d748>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import math\n",
    "from matplotlib import (cm, pyplot as plt, mlab)\n",
    "\n",
    "def visualize(word, model):\n",
    "    \"\"\" visualize the input model for a particular word \"\"\"\n",
    "    variance=np.array([np.diag(model.covars_[i]) for i in range(model.n_components)])\n",
    "    figures = []\n",
    "    for parm_idx in range(len(model.means_[0])):\n",
    "        xmin = int(min(model.means_[:,parm_idx]) - max(variance[:,parm_idx]))\n",
    "        xmax = int(max(model.means_[:,parm_idx]) + max(variance[:,parm_idx]))\n",
    "        fig, axs = plt.subplots(model.n_components, sharex=True, sharey=False)\n",
    "        colours = cm.rainbow(np.linspace(0, 1, model.n_components))\n",
    "        for i, (ax, colour) in enumerate(zip(axs, colours)):\n",
    "            x = np.linspace(xmin, xmax, 100)\n",
    "            mu = model.means_[i,parm_idx]\n",
    "            sigma = math.sqrt(np.diag(model.covars_[i])[parm_idx])\n",
    "            ax.plot(x, mlab.normpdf(x, mu, sigma), c=colour)\n",
    "            ax.set_title(\"{} feature {} hidden state #{}\".format(word, parm_idx, i))\n",
    "\n",
    "            ax.grid(True)\n",
    "        figures.append(plt)\n",
    "    for p in figures:\n",
    "        p.show()\n",
    "        \n",
    "visualize(my_testword, model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "#####  ModelSelector class\n",
    "Review the `ModelSelector` class from the codebase found in the `my_model_selectors.py` module.  It is designed to be a strategy pattern for choosing different model selectors.  For the project submission in this section, subclass `SelectorModel` to implement the following model selectors.  In other words, you will write your own classes/functions in the `my_model_selectors.py` module and run them from this notebook:\n",
    "\n",
    "- `SelectorCV `:  Log likelihood with CV\n",
    "- `SelectorBIC`: BIC \n",
    "- `SelectorDIC`: DIC\n",
    "\n",
    "You will train each word in the training set with a range of values for the number of hidden states, and then score these alternatives with the model selector, choosing the \"best\" according to each strategy. The simple case of training with a constant value for `n_components` can be called using the provided `SelectorConstant` subclass as follow:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of states trained in model for JOHN is 3\n"
     ]
    }
   ],
   "source": [
    "from my_model_selectors import SelectorConstant\n",
    "\n",
    "training = asl.build_training(features_ground) \n",
    "word = 'JOHN'\n",
    "model = SelectorConstant(training.get_all_sequences(), training.get_all_Xlengths(), word, n_constant=3).select()\n",
    "print(\"Number of states trained in model for {} is {}\".format(word, model.n_components))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Cross-validation folds\n",
    "If we simply score the model with the Log Likelihood calculated from the feature sequences it has been trained on, we should expect that more complex models will have higher likelihoods. However, that doesn't tell us which would have a better likelihood score on unseen data.  The model will likely be overfit as complexity is added.  To estimate which topology model is better using only the training data, we can compare scores using cross-validation.  One technique for cross-validation is to break the training set into \"folds\" and rotate which fold is left out of training.  The \"left out\" fold scored.  This gives us a proxy method of finding the best model to use on \"unseen data\". In the following example, a set of word sequences is broken into three folds using the [scikit-learn Kfold](http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.KFold.html) class object. When you implement `SelectorCV`, you will use this technique."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train fold indices:[2 3 4 5] Test fold indices:[0 1]\n",
      "Train fold indices:[0 1 4 5] Test fold indices:[2 3]\n",
      "Train fold indices:[0 1 2 3] Test fold indices:[4 5]\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import KFold\n",
    "\n",
    "training = asl.build_training(features_polar)\n",
    "word = 'VEGETABLE' \n",
    "word_sequences = training.get_word_sequences(word)\n",
    "split_method = KFold()\n",
    "for cv_train_idx, cv_test_idx in split_method.split(word_sequences):\n",
    "    print(\"Train fold indices:{} Test fold indices:{}\".format(cv_train_idx, cv_test_idx))  # view indices of the folds"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Tip:** In order to run `hmmlearn` training using the X,lengths tuples on the new folds, subsets must be combined based on the indices given for the folds.  A helper utility has been provided in the `asl_utils` module named `combine_sequences` for this purpose."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Scoring models with other criterion\n",
    "Scoring model topologies with **BIC** balances fit and complexity within the training set for each word.  In the BIC equation, a penalty term penalizes complexity to avoid overfitting, so that it is not necessary to also use cross-validation in the selection process.  There are a number of references on the internet for this criterion.  These [slides](http://www2.imm.dtu.dk/courses/02433/doc/ch6_slides.pdf) include a formula you may find helpful for your implementation.\n",
    "\n",
    "The advantages of scoring model topologies with **DIC** over BIC are presented by Alain Biem in this [reference](http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.58.6208&rep=rep1&type=pdf) (also found [here](https://pdfs.semanticscholar.org/ed3d/7c4a5f607201f3848d4c02dd9ba17c791fc2.pdf)).  DIC scores the discriminant ability of a training set for one word against competing words.  Instead of a penalty term for complexity, it provides a penalty if model liklihoods for non-matching words are too similar to model likelihoods for the correct word in the word set."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part2_submission'></a>\n",
    "### Model Selection Implementation Submission\n",
    "Implement `SelectorCV`, `SelectorBIC`, and `SelectorDIC` classes in the `my_model_selectors.py` module.  Run the selectors on the following five words. Then answer the questions about your results.\n",
    "\n",
    "**Tip:** The `hmmlearn` library may not be able to train or score all models.  Implement try/except contructs as necessary to eliminate non-viable models from consideration."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "words_to_train = ['FISH', 'BOOK', 'VEGETABLE', 'FUTURE', 'JOHN']\n",
    "import timeit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# autoreload for automatically reloading changes made in my_model_selectors and my_recognizer\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training complete for FISH with 5 states with time 0.3062538743794822 seconds\n",
      "Training complete for BOOK with 6 states with time 7.112399142717507 seconds\n",
      "Training complete for VEGETABLE with 2 states with time 2.5221136710323 seconds\n",
      "Training complete for FUTURE with 2 states with time 6.273336370541246 seconds\n",
      "Training complete for JOHN with 12 states with time 47.98720740721412 seconds\n"
     ]
    }
   ],
   "source": [
    "# Implemented SelectorCV in my_model_selector.py\n",
    "from my_model_selectors import SelectorCV\n",
    "\n",
    "training = asl.build_training(features_ground)  # Experiment here with different feature sets defined in part 1\n",
    "sequences = training.get_all_sequences()\n",
    "Xlengths = training.get_all_Xlengths()\n",
    "for word in words_to_train:\n",
    "    start = timeit.default_timer()\n",
    "    model = SelectorCV(sequences, Xlengths, word, \n",
    "                    min_n_components=2, max_n_components=15, random_state = 14).select()\n",
    "    end = timeit.default_timer()-start\n",
    "    if model is not None:\n",
    "        print(\"Training complete for {} with {} states with time {} seconds\".format(word, model.n_components, end))\n",
    "    else:\n",
    "        print(\"Training failed for {}\".format(word))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training complete for FISH with 5 states with time 0.37151905295532117 seconds\n",
      "Training complete for BOOK with 8 states with time 2.4381848263255392 seconds\n",
      "Training complete for VEGETABLE with 9 states with time 0.8401907940159532 seconds\n",
      "Training complete for FUTURE with 9 states with time 2.284627872682691 seconds\n",
      "Training complete for JOHN with 13 states with time 19.590519256767962 seconds\n"
     ]
    }
   ],
   "source": [
    "# Implemented SelectorBIC in module my_model_selectors.py\n",
    "from my_model_selectors import SelectorBIC\n",
    "\n",
    "training = asl.build_training(features_ground)  # Experiment here with different feature sets defined in part 1\n",
    "sequences = training.get_all_sequences()\n",
    "Xlengths = training.get_all_Xlengths()\n",
    "for word in words_to_train:\n",
    "    start = timeit.default_timer()\n",
    "    model = SelectorBIC(sequences, Xlengths, word, \n",
    "                    min_n_components=2, max_n_components=15, random_state = 14).select()\n",
    "    end = timeit.default_timer()-start\n",
    "    if model is not None:\n",
    "        print(\"Training complete for {} with {} states with time {} seconds\".format(word, model.n_components, end))\n",
    "    else:\n",
    "        print(\"Training failed for {}\".format(word))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training complete for FISH with 3 states with time 1.5058167703243157 seconds\n",
      "Training complete for BOOK with 15 states with time 4.572510753055951 seconds\n",
      "Training complete for VEGETABLE with 15 states with time 3.0521982497501625 seconds\n",
      "Training complete for FUTURE with 15 states with time 4.508048836046498 seconds\n",
      "Training complete for JOHN with 15 states with time 18.885892427831635 seconds\n"
     ]
    }
   ],
   "source": [
    "# Implemented SelectorDIC in module my_model_selectors.py\n",
    "from my_model_selectors import SelectorDIC\n",
    "\n",
    "training = asl.build_training(features_ground)  # Experiment here with different feature sets defined in part 1\n",
    "sequences = training.get_all_sequences()\n",
    "Xlengths = training.get_all_Xlengths()\n",
    "for word in words_to_train:\n",
    "    start = timeit.default_timer()\n",
    "    model = SelectorDIC(sequences, Xlengths, word, \n",
    "                    min_n_components=2, max_n_components=15, random_state = 14).select()\n",
    "    end = timeit.default_timer()-start\n",
    "    if model is not None:\n",
    "        print(\"Training complete for {} with {} states with time {} seconds\".format(word, model.n_components, end))\n",
    "    else:\n",
    "        print(\"Training failed for {}\".format(word))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Question 2:**  Compare and contrast the possible advantages and disadvantages of the various model selectors implemented.\n",
    "\n",
    "**Answer 2:** BIC criterian tries to penalize complex models by adding a term that is the product of the number of free parameters and the sample size. This has the desirable impact of avoiding overfitting on the training data. BIC however, does not explicitly compare the likelihood of \"getting it wrong\" against other words. DIC on the other hand, while does not explicitly control of model size, does compare the likelohood with average likelihood of misclassifying the dataset. We do note in the dataset that the models trained using DIC are larger than those trained using BIC - howevver, they do appear to be less complex than the models trained by CV - thus suggesting that anti-likelihood adjustment leads to simpler models to some extent. We note that CV models are large, specially when sufficient data is available (consider the word \"JOHN\" with 113 sample points for training). This suggests that the selector does not sufficiently control of complex models, despite us tying to avoid overfitting using folds."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part2_test'></a>\n",
    "### Model Selector Unit Testing\n",
    "Run the following unit tests as a sanity check on the implemented model selectors.  The test simply looks for valid interfaces  but is not exhaustive. However, the project should not be submitted if these tests don't pass."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "....\n",
      "----------------------------------------------------------------------\n",
      "Ran 4 tests in 53.056s\n",
      "\n",
      "OK\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<unittest.runner.TextTestResult run=4 errors=0 failures=0>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from asl_test_model_selectors import TestSelectors\n",
    "suite = unittest.TestLoader().loadTestsFromModule(TestSelectors())\n",
    "unittest.TextTestRunner().run(suite)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part3_tutorial'></a>\n",
    "## PART 3: Recognizer\n",
    "The objective of this section is to \"put it all together\".  Using the four feature sets created and the three model selectors, you will experiment with the models and present your results.  Instead of training only five specific words as in the previous section, train the entire set with a feature set and model selector strategy.  \n",
    "### Recognizer Tutorial\n",
    "##### Train the full training set\n",
    "The following example trains the entire set with the example `features_ground` and `SelectorConstant` features and model selector.  Use this pattern for you experimentation and final submission cells.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of word models returned = 112\n"
     ]
    }
   ],
   "source": [
    "from my_model_selectors import SelectorConstant\n",
    "\n",
    "def train_all_words(features, model_selector):\n",
    "    training = asl.build_training(features)\n",
    "    sequences = training.get_all_sequences()\n",
    "    Xlengths = training.get_all_Xlengths()\n",
    "    model_dict = {}\n",
    "    for word in training.words:\n",
    "        model = model_selector(sequences, Xlengths, word, \n",
    "                        n_constant=3).select()\n",
    "        model_dict[word]=model\n",
    "    return model_dict\n",
    "\n",
    "models = train_all_words(features_polar, SelectorConstant)\n",
    "print(\"Number of word models returned = {}\".format(len(models)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "##### Load the test set\n",
    "The `build_test` method in `ASLdb` is similar to the `build_training` method already presented, but there are a few differences:\n",
    "- the object is type `SinglesData` \n",
    "- the internal dictionary keys are the index of the test word rather than the word itself\n",
    "- the getter methods are `get_all_sequences`, `get_all_Xlengths`, `get_item_sequences` and `get_item_Xlengths`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of test set items: 178\n",
      "Number of test set sentences: 40\n"
     ]
    }
   ],
   "source": [
    "test_set = asl.build_test(features_ground)\n",
    "print(\"Number of test set items: {}\".format(test_set.num_items))\n",
    "print(\"Number of test set sentences: {}\".format(len(test_set.sentences_index)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part3_submission'></a>\n",
    "### Recognizer Implementation Submission\n",
    "For the final project submission, students must implement a recognizer following guidance in the `my_recognizer.py` module.  Experiment with the four feature sets and the three model selection methods (that's 12 possible combinations). You can add and remove cells for experimentation or run the recognizers locally in some other way during your experiments, but retain the results for your discussion.  For submission, you will provide code cells of **only three** interesting combinations for your discussion (see questions below). At least one of these should produce a word error rate of less than 60%, i.e. WER < 0.60 . \n",
    "\n",
    "**Tip:** The hmmlearn library may not be able to train or score all models.  Implement try/except contructs as necessary to eliminate non-viable models from consideration."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from my_recognizer import recognize\n",
    "from asl_utils import show_errors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>Selector</th>\n",
       "      <th>BIC</th>\n",
       "      <th>CV</th>\n",
       "      <th>DIC</th>\n",
       "      <th>Mean</th>\n",
       "      <th>Min</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Features</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Polar+Delta+C2</th>\n",
       "      <td>0.449438</td>\n",
       "      <td>0.477528</td>\n",
       "      <td>0.505618</td>\n",
       "      <td>0.477528</td>\n",
       "      <td>0.449438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Delta+C2</th>\n",
       "      <td>0.449438</td>\n",
       "      <td>0.511236</td>\n",
       "      <td>0.539326</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.449438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Delta+C1+C2</th>\n",
       "      <td>0.455056</td>\n",
       "      <td>0.488764</td>\n",
       "      <td>0.516854</td>\n",
       "      <td>0.486891</td>\n",
       "      <td>0.455056</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Polar+Delta+C1</th>\n",
       "      <td>0.466292</td>\n",
       "      <td>0.505618</td>\n",
       "      <td>0.516854</td>\n",
       "      <td>0.496255</td>\n",
       "      <td>0.466292</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Polar+Delta</th>\n",
       "      <td>0.477528</td>\n",
       "      <td>0.595506</td>\n",
       "      <td>0.511236</td>\n",
       "      <td>0.528090</td>\n",
       "      <td>0.477528</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Delta</th>\n",
       "      <td>0.483146</td>\n",
       "      <td>0.505618</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.496255</td>\n",
       "      <td>0.483146</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Polar+Delta+C2</th>\n",
       "      <td>0.483146</td>\n",
       "      <td>0.550562</td>\n",
       "      <td>0.522472</td>\n",
       "      <td>0.518727</td>\n",
       "      <td>0.483146</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Delta+C1</th>\n",
       "      <td>0.488764</td>\n",
       "      <td>0.578652</td>\n",
       "      <td>0.522472</td>\n",
       "      <td>0.529963</td>\n",
       "      <td>0.488764</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Polar+Delta+C1</th>\n",
       "      <td>0.505618</td>\n",
       "      <td>0.522472</td>\n",
       "      <td>0.567416</td>\n",
       "      <td>0.531835</td>\n",
       "      <td>0.505618</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Polar+C2</th>\n",
       "      <td>0.528090</td>\n",
       "      <td>0.522472</td>\n",
       "      <td>0.505618</td>\n",
       "      <td>0.518727</td>\n",
       "      <td>0.505618</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+C2</th>\n",
       "      <td>0.556180</td>\n",
       "      <td>0.539326</td>\n",
       "      <td>0.550562</td>\n",
       "      <td>0.548689</td>\n",
       "      <td>0.539326</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+Polar</th>\n",
       "      <td>0.550562</td>\n",
       "      <td>0.567416</td>\n",
       "      <td>0.539326</td>\n",
       "      <td>0.552434</td>\n",
       "      <td>0.539326</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground</th>\n",
       "      <td>0.550562</td>\n",
       "      <td>0.539326</td>\n",
       "      <td>0.573034</td>\n",
       "      <td>0.554307</td>\n",
       "      <td>0.539326</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+C1</th>\n",
       "      <td>0.539326</td>\n",
       "      <td>0.606742</td>\n",
       "      <td>0.539326</td>\n",
       "      <td>0.561798</td>\n",
       "      <td>0.539326</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Polar</th>\n",
       "      <td>0.544944</td>\n",
       "      <td>0.561798</td>\n",
       "      <td>0.544944</td>\n",
       "      <td>0.550562</td>\n",
       "      <td>0.544944</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ground+C1+C2</th>\n",
       "      <td>0.567416</td>\n",
       "      <td>0.544944</td>\n",
       "      <td>0.567416</td>\n",
       "      <td>0.559925</td>\n",
       "      <td>0.544944</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>C1+C2</th>\n",
       "      <td>0.584270</td>\n",
       "      <td>0.623596</td>\n",
       "      <td>0.623596</td>\n",
       "      <td>0.610487</td>\n",
       "      <td>0.584270</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>C2</th>\n",
       "      <td>0.640449</td>\n",
       "      <td>0.601124</td>\n",
       "      <td>0.662921</td>\n",
       "      <td>0.634831</td>\n",
       "      <td>0.601124</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Delta</th>\n",
       "      <td>0.612360</td>\n",
       "      <td>0.612360</td>\n",
       "      <td>0.623596</td>\n",
       "      <td>0.616105</td>\n",
       "      <td>0.612360</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>C1</th>\n",
       "      <td>0.629213</td>\n",
       "      <td>0.646067</td>\n",
       "      <td>0.674157</td>\n",
       "      <td>0.649813</td>\n",
       "      <td>0.629213</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mean</th>\n",
       "      <td>0.528090</td>\n",
       "      <td>0.555056</td>\n",
       "      <td>0.555337</td>\n",
       "      <td>0.546161</td>\n",
       "      <td>0.521910</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Selector                    BIC        CV       DIC      Mean       Min\n",
       "Features                                                               \n",
       "Polar+Delta+C2         0.449438  0.477528  0.505618  0.477528  0.449438\n",
       "Ground+Delta+C2        0.449438  0.511236  0.539326  0.500000  0.449438\n",
       "Delta+C1+C2            0.455056  0.488764  0.516854  0.486891  0.455056\n",
       "Polar+Delta+C1         0.466292  0.505618  0.516854  0.496255  0.466292\n",
       "Ground+Polar+Delta     0.477528  0.595506  0.511236  0.528090  0.477528\n",
       "Ground+Delta           0.483146  0.505618  0.500000  0.496255  0.483146\n",
       "Ground+Polar+Delta+C2  0.483146  0.550562  0.522472  0.518727  0.483146\n",
       "Ground+Delta+C1        0.488764  0.578652  0.522472  0.529963  0.488764\n",
       "Ground+Polar+Delta+C1  0.505618  0.522472  0.567416  0.531835  0.505618\n",
       "Ground+Polar+C2        0.528090  0.522472  0.505618  0.518727  0.505618\n",
       "Ground+C2              0.556180  0.539326  0.550562  0.548689  0.539326\n",
       "Ground+Polar           0.550562  0.567416  0.539326  0.552434  0.539326\n",
       "Ground                 0.550562  0.539326  0.573034  0.554307  0.539326\n",
       "Ground+C1              0.539326  0.606742  0.539326  0.561798  0.539326\n",
       "Polar                  0.544944  0.561798  0.544944  0.550562  0.544944\n",
       "Ground+C1+C2           0.567416  0.544944  0.567416  0.559925  0.544944\n",
       "C1+C2                  0.584270  0.623596  0.623596  0.610487  0.584270\n",
       "C2                     0.640449  0.601124  0.662921  0.634831  0.601124\n",
       "Delta                  0.612360  0.612360  0.623596  0.616105  0.612360\n",
       "C1                     0.629213  0.646067  0.674157  0.649813  0.629213\n",
       "Mean                   0.528090  0.555056  0.555337  0.546161  0.521910"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## We test a range of features with each of the model selectors and display results in a table.\n",
    "\n",
    "features_custom1 = ['diff-x', 'diff-y']\n",
    "features_custom2 = ['norm-diff-x', 'norm-diff-y']\n",
    "\n",
    "\n",
    "\n",
    "#Dictionary with features to test\n",
    "feature_dict = {\n",
    "                'Original': originalFeatures,\n",
    "                'Ground' : features_ground,\n",
    "                'Polar' : features_polar,\n",
    "                'Delta' : features_delta,\n",
    "                'C1': features_custom1,\n",
    "                'C2': features_custom2,\n",
    "                'C1+C2': features_custom1 + features_custom2,\n",
    "                'Ground+Polar': features_ground + features_polar,\n",
    "                'Ground+Delta': features_ground + features_delta,\n",
    "                'Ground+C1': features_ground + features_custom1,\n",
    "                'Ground+C2': features_ground + features_custom2,\n",
    "                'Ground+Polar+Delta': features_ground + features_polar + features_delta,\n",
    "                'Ground+C1+C2': features_ground + features_custom1 + features_custom2,\n",
    "                'Ground+Polar+Delta+C1': features_ground + features_polar + features_delta + features_custom1,\n",
    "                'Ground+Polar+Delta+C2': features_ground + features_polar + features_delta + features_custom2,\n",
    "                'Ground+Polar+C2': features_ground + features_custom2 + features_polar,\n",
    "                'Ground+Delta+C2': features_ground + features_custom2 + features_delta,\n",
    "                'Ground+Delta+C1': features_ground + features_custom1 + features_delta,\n",
    "                'Polar+Delta+C1': features_polar + features_custom1 + features_delta,\n",
    "                'Polar+Delta+C2': features_polar + features_custom2 + features_delta,\n",
    "                'Delta+C1+C2': features_custom1 + features_custom2 + features_delta,\n",
    "               }\n",
    "\n",
    "#Features actively being run\n",
    "active_feature_dict = {}\n",
    "\n",
    "#Dictionary with selectors\n",
    "selector_dict = {'BIC': SelectorBIC,'DIC': SelectorDIC,'CV': SelectorCV}\n",
    "\n",
    "#Load results from any previous active runs and add the results from current run\n",
    "from my_recognizer import recognize\n",
    "#results = pd.DataFrame(columns = ['Features', 'Selector', 'CorrectWords', 'Score'])\n",
    "results = pd.read_csv('results.csv')\n",
    "for fname, features in active_feature_dict.items():\n",
    "    for sname, model_selector in selector_dict.items():\n",
    "        models = train_all_words(features, model_selector)\n",
    "        test_set = asl.build_test(features)\n",
    "        probabilities, guesses = recognize(models, test_set)\n",
    "        correct_count =  sum([i==j for i,j in zip(test_set.wordlist, guesses)])\n",
    "        current_result = pd.Series([fname, sname, correct_count, 1 - (correct_count/178)], index = ['Features', 'Selector', 'CorrectWords', 'Score'])\n",
    "        results = results.append(current_result,ignore_index=True)\n",
    "        \n",
    "        \n",
    "#Save results for future. Pivot and display\n",
    "results.to_csv('results.csv', index=False)\n",
    "results_pivoted = results.pivot(index='Features', columns='Selector', values = 'Score')\n",
    "results_pivoted['Mean'] = results_pivoted.mean(axis=1)\n",
    "results_pivoted['Min'] = results_pivoted.min(axis=1)\n",
    "results_pivoted = results_pivoted.sort_values(by='Mean')\n",
    "results_pivoted = results_pivoted.sort_values(by='Min')\n",
    "selector_mean = results_pivoted.mean(axis=0)\n",
    "selector_mean.name = 'Mean'\n",
    "results_pivoted = results_pivoted.append(selector_mean)\n",
    "results_pivoted"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "**** WER = 0.4550561797752809\n",
      "Total correct: 97 out of 178\n",
      "Video  Recognized                                                    Correct\n",
      "=====================================================================================================\n",
      "    2: JOHN WRITE HOMEWORK                                           JOHN WRITE HOMEWORK\n",
      "    7: JOHN *HAVE GO *BOX                                            JOHN CAN GO CAN\n",
      "   12: JOHN CAN *GO1 CAN                                             JOHN CAN GO CAN\n",
      "   21: JOHN *CHOCOLATE *JOHN *LOVE *CAR *WHAT *MARY *IX              JOHN FISH WONT EAT BUT CAN EAT CHICKEN\n",
      "   25: JOHN *IX IX *LIKE IX                                          JOHN LIKE IX IX IX\n",
      "   28: JOHN LIKE IX IX IX                                            JOHN LIKE IX IX IX\n",
      "   30: JOHN *MARY *LOVE IX IX                                        JOHN LIKE IX IX IX\n",
      "   36: MARY *JOHN *GIVE1 *TELL *MARY *MARY                           MARY VEGETABLE KNOW IX LIKE CORN1\n",
      "   40: JOHN IX *WHO *JOHN LOVE                                       JOHN IX THINK MARY LOVE\n",
      "   43: JOHN *FUTURE BUY HOUSE                                        JOHN MUST BUY HOUSE\n",
      "   50: *JOHN *NOT BUY CAR *MARY                                      FUTURE JOHN BUY CAR SHOULD\n",
      "   54: JOHN *FUTURE *FUTURE BUY HOUSE                                JOHN SHOULD NOT BUY HOUSE\n",
      "   57: JOHN *WHO VISIT *IX                                           JOHN DECIDE VISIT MARY\n",
      "   67: JOHN FUTURE *FUTURE BUY HOUSE                                 JOHN FUTURE NOT BUY HOUSE\n",
      "   71: JOHN *FINISH VISIT MARY                                       JOHN WILL VISIT MARY\n",
      "   74: JOHN *MARY VISIT MARY                                         JOHN NOT VISIT MARY\n",
      "   77: ANN BLAME MARY                                                ANN BLAME MARY\n",
      "   84: *JOHN *GIVE1 *YESTERDAY BOOK                                  IX-1P FIND SOMETHING-ONE BOOK\n",
      "   89: *MARY *JOHN *IX *IX IX NEW COAT                               JOHN IX GIVE MAN IX NEW COAT\n",
      "   90: *WHO *IX IX *IX *IX BOOK                                      JOHN GIVE IX SOMETHING-ONE WOMAN BOOK\n",
      "   92: JOHN *IX IX *IX WOMAN BOOK                                    JOHN GIVE IX SOMETHING-ONE WOMAN BOOK\n",
      "  100: *IX NEW CAR BREAK-DOWN                                        POSS NEW CAR BREAK-DOWN\n",
      "  105: JOHN *JOHN                                                    JOHN LEG\n",
      "  107: JOHN *JOHN *ARRIVE *YESTERDAY *JOHN                           JOHN POSS FRIEND HAVE CANDY\n",
      "  108: WOMAN ARRIVE                                                  WOMAN ARRIVE\n",
      "  113: *JOHN CAR *JOHN *JOHN *BUY1                                   IX CAR BLUE SUE BUY\n",
      "  119: *WHO *BUY1 *JOHN CAR *EAT                                     SUE BUY IX CAR BLUE\n",
      "  122: JOHN *GIVE1 BOOK                                              JOHN READ BOOK\n",
      "  139: JOHN *BUY1 WHAT *WHAT BOOK                                    JOHN BUY WHAT YESTERDAY BOOK\n",
      "  142: JOHN BUY YESTERDAY WHAT BOOK                                  JOHN BUY YESTERDAY WHAT BOOK\n",
      "  158: LOVE JOHN *VEGETABLE                                          LOVE JOHN WHO\n",
      "  167: JOHN *SOMETHING-ONE *IX LOVE MARY                             JOHN IX SAY LOVE MARY\n",
      "  171: JOHN *JOHN BLAME                                              JOHN MARY BLAME\n",
      "  174: *GIVE1 *HAVE GIVE1 *WHO *BOX                                  PEOPLE GROUP GIVE1 JANA TOY\n",
      "  181: *IX ARRIVE                                                    JOHN ARRIVE\n",
      "  184: ALL BOY *GIVE1 TEACHER APPLE                                  ALL BOY GIVE TEACHER APPLE\n",
      "  189: JOHN *JOHN *VISIT BOX                                         JOHN GIVE GIRL BOX\n",
      "  193: JOHN *EAT *SOMETHING-ONE *WHAT                                JOHN GIVE GIRL BOX\n",
      "  199: *JOHN CHOCOLATE WHO                                           LIKE CHOCOLATE WHO\n",
      "  201: JOHN *WHO *WHO *WOMAN BUY HOUSE                               JOHN TELL MARY IX-1P BUY HOUSE\n"
     ]
    }
   ],
   "source": [
    "features = features_custom1 + features_custom2 + features_delta\n",
    "model_selector = SelectorBIC\n",
    "\n",
    "models = train_all_words(features, model_selector)\n",
    "test_set = asl.build_test(features)\n",
    "probabilities, guesses = recognize(models, test_set)\n",
    "show_errors(guesses, test_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "**** WER = 0.449438202247191\n",
      "Total correct: 98 out of 178\n",
      "Video  Recognized                                                    Correct\n",
      "=====================================================================================================\n",
      "    2: JOHN WRITE *NEW                                               JOHN WRITE HOMEWORK\n",
      "    7: JOHN *CAR GO *CAR                                             JOHN CAN GO CAN\n",
      "   12: JOHN CAN *WHAT *HOUSE                                         JOHN CAN GO CAN\n",
      "   21: JOHN *GIVE1 WONT *MARY *CAR *CAR *FUTURE *MARY                JOHN FISH WONT EAT BUT CAN EAT CHICKEN\n",
      "   25: JOHN LIKE IX *LIKE IX                                         JOHN LIKE IX IX IX\n",
      "   28: JOHN LIKE IX *LIKE IX                                         JOHN LIKE IX IX IX\n",
      "   30: JOHN LIKE *MARY *MARY IX                                      JOHN LIKE IX IX IX\n",
      "   36: MARY *VISIT *GIVE *GO *LOVE *MARY                             MARY VEGETABLE KNOW IX LIKE CORN1\n",
      "   40: JOHN *VISIT *JOHN *VISIT *MARY                                JOHN IX THINK MARY LOVE\n",
      "   43: JOHN *JOHN BUY HOUSE                                          JOHN MUST BUY HOUSE\n",
      "   50: *JOHN *SEE BUY CAR *JOHN                                      FUTURE JOHN BUY CAR SHOULD\n",
      "   54: JOHN SHOULD *FUTURE BUY HOUSE                                 JOHN SHOULD NOT BUY HOUSE\n",
      "   57: *MARY *MARY VISIT MARY                                        JOHN DECIDE VISIT MARY\n",
      "   67: JOHN FUTURE *JOHN BUY HOUSE                                   JOHN FUTURE NOT BUY HOUSE\n",
      "   71: JOHN *VISIT VISIT MARY                                        JOHN WILL VISIT MARY\n",
      "   74: *IX *MARY VISIT MARY                                          JOHN NOT VISIT MARY\n",
      "   77: *JOHN BLAME MARY                                              ANN BLAME MARY\n",
      "   84: *JOHN *GIVE1 *NEW BOOK                                        IX-1P FIND SOMETHING-ONE BOOK\n",
      "   89: *MARY IX *IX *IX IX NEW *BREAK-DOWN                           JOHN IX GIVE MAN IX NEW COAT\n",
      "   90: JOHN *IX IX *IX WOMAN BOOK                                    JOHN GIVE IX SOMETHING-ONE WOMAN BOOK\n",
      "   92: JOHN *IX IX *IX WOMAN BOOK                                    JOHN GIVE IX SOMETHING-ONE WOMAN BOOK\n",
      "  100: POSS NEW CAR BREAK-DOWN                                       POSS NEW CAR BREAK-DOWN\n",
      "  105: JOHN *FRANK                                                   JOHN LEG\n",
      "  107: JOHN *IX *ARRIVE HAVE *JOHN                                   JOHN POSS FRIEND HAVE CANDY\n",
      "  108: *JOHN *BOOK                                                   WOMAN ARRIVE\n",
      "  113: *JOHN CAR *JOHN *JOHN *ARRIVE                                 IX CAR BLUE SUE BUY\n",
      "  119: *MARY *BUY1 IX CAR *IX                                        SUE BUY IX CAR BLUE\n",
      "  122: JOHN *GIVE1 BOOK                                              JOHN READ BOOK\n",
      "  139: JOHN *BUY1 WHAT *WHAT BOOK                                    JOHN BUY WHAT YESTERDAY BOOK\n",
      "  142: JOHN BUY YESTERDAY WHAT BOOK                                  JOHN BUY YESTERDAY WHAT BOOK\n",
      "  158: LOVE JOHN WHO                                                 LOVE JOHN WHO\n",
      "  167: JOHN IX *MARY LOVE MARY                                       JOHN IX SAY LOVE MARY\n",
      "  171: JOHN *JOHN BLAME                                              JOHN MARY BLAME\n",
      "  174: *NEW *GIVE3 GIVE1 *MARY *BLAME                                PEOPLE GROUP GIVE1 JANA TOY\n",
      "  181: JOHN ARRIVE                                                   JOHN ARRIVE\n",
      "  184: *IX BOY *GIVE1 TEACHER APPLE                                  ALL BOY GIVE TEACHER APPLE\n",
      "  189: JOHN *JOHN *VISIT BOX                                         JOHN GIVE GIRL BOX\n",
      "  193: JOHN *IX *VISIT BOX                                           JOHN GIVE GIRL BOX\n",
      "  199: *JOHN *NEW WHO                                                LIKE CHOCOLATE WHO\n",
      "  201: JOHN *FUTURE *WOMAN *WHO BUY HOUSE                            JOHN TELL MARY IX-1P BUY HOUSE\n"
     ]
    }
   ],
   "source": [
    "features = features_ground + features_custom2 + features_delta\n",
    "model_selector = SelectorBIC\n",
    "\n",
    "models = train_all_words(features, model_selector)\n",
    "test_set = asl.build_test(features)\n",
    "probabilities, guesses = recognize(models, test_set)\n",
    "show_errors(guesses, test_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "**** WER = 0.449438202247191\n",
      "Total correct: 98 out of 178\n",
      "Video  Recognized                                                    Correct\n",
      "=====================================================================================================\n",
      "    2: JOHN WRITE HOMEWORK                                           JOHN WRITE HOMEWORK\n",
      "    7: JOHN *CAR *IX *CAR                                            JOHN CAN GO CAN\n",
      "   12: JOHN CAN *GO1 CAN                                             JOHN CAN GO CAN\n",
      "   21: JOHN *BROTHER WONT *MARY *CAR *CAR *FUTURE *MARY              JOHN FISH WONT EAT BUT CAN EAT CHICKEN\n",
      "   25: JOHN *IX *LOVE IX IX                                          JOHN LIKE IX IX IX\n",
      "   28: JOHN *IX IX IX IX                                             JOHN LIKE IX IX IX\n",
      "   30: *LIKE *MARY *MARY *MARY *MARY                                 JOHN LIKE IX IX IX\n",
      "   36: MARY *JOHN *GIVE *VISIT *MARY *MARY                           MARY VEGETABLE KNOW IX LIKE CORN1\n",
      "   40: JOHN *VISIT *FUTURE1 *JOHN LOVE                               JOHN IX THINK MARY LOVE\n",
      "   43: JOHN *JOHN BUY HOUSE                                          JOHN MUST BUY HOUSE\n",
      "   50: *JOHN *SEE BUY CAR *JOHN                                      FUTURE JOHN BUY CAR SHOULD\n",
      "   54: JOHN *JOHN *FUTURE BUY HOUSE                                  JOHN SHOULD NOT BUY HOUSE\n",
      "   57: JOHN *JOHN VISIT MARY                                         JOHN DECIDE VISIT MARY\n",
      "   67: JOHN FUTURE NOT BUY HOUSE                                     JOHN FUTURE NOT BUY HOUSE\n",
      "   71: JOHN *VISIT *BLAME MARY                                       JOHN WILL VISIT MARY\n",
      "   74: JOHN *MARY VISIT MARY                                         JOHN NOT VISIT MARY\n",
      "   77: *JOHN BLAME MARY                                              ANN BLAME MARY\n",
      "   84: *JOHN *GIVE1 *JOHN BOOK                                       IX-1P FIND SOMETHING-ONE BOOK\n",
      "   89: *MARY IX *IX *IX IX NEW COAT                                  JOHN IX GIVE MAN IX NEW COAT\n",
      "   90: JOHN *IX IX *IX WOMAN BOOK                                    JOHN GIVE IX SOMETHING-ONE WOMAN BOOK\n",
      "   92: JOHN *IX IX *IX WOMAN BOOK                                    JOHN GIVE IX SOMETHING-ONE WOMAN BOOK\n",
      "  100: POSS NEW CAR BREAK-DOWN                                       POSS NEW CAR BREAK-DOWN\n",
      "  105: JOHN *SEE                                                     JOHN LEG\n",
      "  107: JOHN *IX *ARRIVE *MARY *JOHN                                  JOHN POSS FRIEND HAVE CANDY\n",
      "  108: *MARY *BOOK                                                   WOMAN ARRIVE\n",
      "  113: IX CAR *IX *JOHN *BUY1                                        IX CAR BLUE SUE BUY\n",
      "  119: *MARY *BUY1 IX CAR *GO                                        SUE BUY IX CAR BLUE\n",
      "  122: JOHN *GIVE1 BOOK                                              JOHN READ BOOK\n",
      "  139: JOHN *BUY1 WHAT *JOHN BOOK                                    JOHN BUY WHAT YESTERDAY BOOK\n",
      "  142: JOHN BUY YESTERDAY WHAT BOOK                                  JOHN BUY YESTERDAY WHAT BOOK\n",
      "  158: LOVE JOHN WHO                                                 LOVE JOHN WHO\n",
      "  167: JOHN IX *MARY LOVE MARY                                       JOHN IX SAY LOVE MARY\n",
      "  171: JOHN *JOHN BLAME                                              JOHN MARY BLAME\n",
      "  174: *GIVE1 GROUP GIVE1 *JOHN *BLAME                               PEOPLE GROUP GIVE1 JANA TOY\n",
      "  181: *IX ARRIVE                                                    JOHN ARRIVE\n",
      "  184: *IX BOY *GIVE1 TEACHER APPLE                                  ALL BOY GIVE TEACHER APPLE\n",
      "  189: JOHN *JOHN *PREFER *CAR                                       JOHN GIVE GIRL BOX\n",
      "  193: JOHN *IX *VISIT BOX                                           JOHN GIVE GIRL BOX\n",
      "  199: *JOHN CHOCOLATE WHO                                           LIKE CHOCOLATE WHO\n",
      "  201: JOHN *SHOULD *WOMAN *LIKE BUY HOUSE                           JOHN TELL MARY IX-1P BUY HOUSE\n"
     ]
    }
   ],
   "source": [
    "features = features_polar + features_custom2 + features_delta\n",
    "model_selector = SelectorBIC\n",
    "\n",
    "models = train_all_words(features, model_selector)\n",
    "test_set = asl.build_test(features)\n",
    "probabilities, guesses = recognize(models, test_set)\n",
    "show_errors(guesses, test_set)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Question 3:**  Summarize the error results from three combinations of features and model selectors.  What was the \"best\" combination and why?  What additional information might we use to improve our WER?  For more insight on improving WER, take a look at the introduction to Part 4.\n",
    "\n",
    "**Answer 3:** We have presented three models, all use BIC for model selection, and bring the error rate to about 45% using custom features. This suggests that feature creation can be an important aspect of model creation. We do note that one of the good performing feature sets do not contain the hand position information in any form (polar or coordinate) and relies only on the delta, and the relative hand postitions - this is an interesting finding. Moreover, while both polar and ground features combined with delta and custom features give us the best results, we observe that the polar features do better with other model selection criteria as well.\n",
    "\n",
    "We thus conclude that the best feature set of this particular sample data is **Polar + Delta + Custom** with an error rate of about **45%**. One interesting thing in our tests is that BIC appears to perform better than DIC. However, upon deeper inspection of the results and classifications, we begin to undertand what could be causing this. The frequent appearance of \"JOHN\" in the test data, along with the tendency of our models to often overclassify words as \"JOHN\" makes for an interesting mix of results. We expect that with a larger and more balanced dataset, DIC could outperform BIC in model selection.\n",
    "\n",
    "WER can further be improved by taking into account the prior probabilities of each of the words occuring. The priors can be calculated by various means including frequency of occuring in the language, grammer rules etc.. Other ways to improve WER could include improved gaussian distributions for first and last stages depending on adjacent words, since hand movements can be impacted by the positioning of the hands toward the end of the preceeding word, and similarly, by the next work gestured."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part3_test'></a>\n",
    "### Recognizer Unit Tests\n",
    "Run the following unit tests as a sanity check on the defined recognizer.  The test simply looks for some valid values but is not exhaustive. However, the project should not be submitted if these tests don't pass."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "..\n",
      "----------------------------------------------------------------------\n",
      "Ran 2 tests in 34.383s\n",
      "\n",
      "OK\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<unittest.runner.TextTestResult run=2 errors=0 failures=0>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from asl_test_recognizer import TestRecognize\n",
    "suite = unittest.TestLoader().loadTestsFromModule(TestRecognize())\n",
    "unittest.TextTestRunner().run(suite)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='part4_info'></a>\n",
    "## PART 4: (OPTIONAL)  Improve the WER with Language Models\n",
    "We've squeezed just about as much as we can out of the model and still only get about 50% of the words right! Surely we can do better than that.  Probability to the rescue again in the form of [statistical language models (SLM)](https://en.wikipedia.org/wiki/Language_model).  The basic idea is that each word has some probability of occurrence within the set, and some probability that it is adjacent to specific other words. We can use that additional information to make better choices.\n",
    "\n",
    "##### Additional reading and resources\n",
    "- [Introduction to N-grams (Stanford Jurafsky slides)](https://web.stanford.edu/class/cs124/lec/languagemodeling.pdf)\n",
    "- [Speech Recognition Techniques for a Sign Language Recognition System, Philippe Dreuw et al](https://www-i6.informatik.rwth-aachen.de/publications/download/154/Dreuw--2007.pdf) see the improved results of applying LM on *this* data!\n",
    "- [SLM data for *this* ASL dataset](ftp://wasserstoff.informatik.rwth-aachen.de/pub/rwth-boston-104/lm/)\n",
    "\n",
    "##### Optional challenge\n",
    "The recognizer you implemented in Part 3 is equivalent to a \"0-gram\" SLM.  Improve the WER with the SLM data provided with the data set in the link above using \"1-gram\", \"2-gram\", and/or \"3-gram\" statistics. The `probabilities` data you've already calculated will be useful and can be turned into a pandas DataFrame if desired (see next cell).  \n",
    "Good luck!  Share your results with the class!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# create a DataFrame of log likelihoods for the test word items\n",
    "df_probs = pd.DataFrame(data=probabilities)\n",
    "df_probs.head()"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda env:aind]",
   "language": "python",
   "name": "conda-env-aind-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.0"
  },
  "nbpresent": {
   "slides": {
    "0a2d4faf-9fb8-4cee-853b-ed68b90f3f8a": {
     "id": "0a2d4faf-9fb8-4cee-853b-ed68b90f3f8a",
     "prev": null,
     "regions": {
      "3fb9ce83-fbb2-4995-832a-f8f400734ad3": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "1dbb9346-179b-4835-b430-6369d88f1a1b",
        "part": "whole"
       },
       "id": "3fb9ce83-fbb2-4995-832a-f8f400734ad3"
      }
     }
    },
    "1519a4fa-1588-4644-98de-9c43bf0aceb5": {
     "id": "1519a4fa-1588-4644-98de-9c43bf0aceb5",
     "prev": "8a712017-49b7-449f-8264-43a032ace902",
     "regions": {
      "29546121-ed11-44b7-8144-0c44e874098f": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "365590a4-6963-4812-a1cf-688f7b6bb9ff",
        "part": "whole"
       },
       "id": "29546121-ed11-44b7-8144-0c44e874098f"
      }
     }
    },
    "176eaccb-15dd-455d-bf07-504213e7aa01": {
     "id": "176eaccb-15dd-455d-bf07-504213e7aa01",
     "prev": "de6b30f4-2463-4901-92ed-aabad78e5e0f",
     "regions": {
      "1542aa9e-dc55-4b90-adef-bf5181872b42": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "5c242050-c1f7-4b3b-8103-2ea9d71a40dc",
        "part": "whole"
       },
       "id": "1542aa9e-dc55-4b90-adef-bf5181872b42"
      }
     }
    },
    "19091b36-b0e7-49b1-b501-ec05937e0da9": {
     "id": "19091b36-b0e7-49b1-b501-ec05937e0da9",
     "prev": "1983c02e-fb99-4c05-a728-e0c0ad7c06d8",
     "regions": {
      "6529a31c-8d45-425c-b1d7-d0ac6fca6a32": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "e766909d-9421-4aaf-9fb1-bc90d27e49e3",
        "part": "whole"
       },
       "id": "6529a31c-8d45-425c-b1d7-d0ac6fca6a32"
      }
     }
    },
    "1983c02e-fb99-4c05-a728-e0c0ad7c06d8": {
     "id": "1983c02e-fb99-4c05-a728-e0c0ad7c06d8",
     "prev": "176eaccb-15dd-455d-bf07-504213e7aa01",
     "regions": {
      "1c4e605d-7f22-4f30-b3fb-74b2937e7a4a": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "4d217204-e5c0-4568-bd30-12c2e41b681d",
        "part": "whole"
       },
       "id": "1c4e605d-7f22-4f30-b3fb-74b2937e7a4a"
      }
     }
    },
    "212b111f-4527-459c-8297-1db5580ee5c9": {
     "id": "212b111f-4527-459c-8297-1db5580ee5c9",
     "prev": "76898529-e49e-4663-8d02-8261dfe1d94b",
     "regions": {
      "2e4bd280-3cd6-47d0-9c81-17737b24053b": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "0c316996-9933-4b3d-82ec-259518dc8bc9",
        "part": "whole"
       },
       "id": "2e4bd280-3cd6-47d0-9c81-17737b24053b"
      }
     }
    },
    "23a7337f-a0cf-4ed4-baa9-ec06bfdc0579": {
     "id": "23a7337f-a0cf-4ed4-baa9-ec06bfdc0579",
     "prev": "e76e9a02-54c1-4ec9-80fb-c611ed398122",
     "regions": {
      "b5721d20-d6f8-4ddb-a5aa-eb16f0cc8893": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "313015a2-b5a9-4136-a8ea-5d011e47d840",
        "part": "whole"
       },
       "id": "b5721d20-d6f8-4ddb-a5aa-eb16f0cc8893"
      }
     }
    },
    "732f1952-ee54-46fb-8067-099512824296": {
     "id": "732f1952-ee54-46fb-8067-099512824296",
     "prev": "0a2d4faf-9fb8-4cee-853b-ed68b90f3f8a",
     "regions": {
      "f31d4597-08ad-4c46-ad52-4bd2d775c624": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "aadfec52-27ca-4541-8920-fa9253d51827",
        "part": "whole"
       },
       "id": "f31d4597-08ad-4c46-ad52-4bd2d775c624"
      }
     }
    },
    "76898529-e49e-4663-8d02-8261dfe1d94b": {
     "id": "76898529-e49e-4663-8d02-8261dfe1d94b",
     "prev": "19091b36-b0e7-49b1-b501-ec05937e0da9",
     "regions": {
      "ec1746fc-aec9-4a7c-8225-9e9ac8d45889": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "b3e539be-84e2-49ce-a183-31cfc5c7ce7c",
        "part": "whole"
       },
       "id": "ec1746fc-aec9-4a7c-8225-9e9ac8d45889"
      }
     }
    },
    "8a712017-49b7-449f-8264-43a032ace902": {
     "id": "8a712017-49b7-449f-8264-43a032ace902",
     "prev": "bed9e696-630e-4747-be1c-bc3737ba992f",
     "regions": {
      "1faab517-cd16-4c63-bb01-a67246749d7a": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "3f14ddf0-4145-4687-9c33-712c3c32520f",
        "part": "whole"
       },
       "id": "1faab517-cd16-4c63-bb01-a67246749d7a"
      }
     }
    },
    "90af992d-eb6d-4496-b2d2-6aa9a95b6a61": {
     "id": "90af992d-eb6d-4496-b2d2-6aa9a95b6a61",
     "prev": "732f1952-ee54-46fb-8067-099512824296",
     "regions": {
      "4f448bec-5be9-4553-88ae-e35ed7612f25": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "c445fbfb-b8ab-4e9a-8d13-12231a1c588f",
        "part": "whole"
       },
       "id": "4f448bec-5be9-4553-88ae-e35ed7612f25"
      }
     }
    },
    "bed9e696-630e-4747-be1c-bc3737ba992f": {
     "id": "bed9e696-630e-4747-be1c-bc3737ba992f",
     "prev": "23a7337f-a0cf-4ed4-baa9-ec06bfdc0579",
     "regions": {
      "ac1513f0-404f-492b-8b42-0313e9a753b0": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "18dd2eee-8b6c-4a5e-9539-132d00a7c7e1",
        "part": "whole"
       },
       "id": "ac1513f0-404f-492b-8b42-0313e9a753b0"
      }
     }
    },
    "de6b30f4-2463-4901-92ed-aabad78e5e0f": {
     "id": "de6b30f4-2463-4901-92ed-aabad78e5e0f",
     "prev": "e36b4639-be8c-46f7-a8c9-bcfb134f9fd0",
     "regions": {
      "55ec36e0-362f-4fd3-8060-7cee056039aa": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "c3cf461e-4c9e-4dec-99d2-07bfa79cbe23",
        "part": "whole"
       },
       "id": "55ec36e0-362f-4fd3-8060-7cee056039aa"
      }
     }
    },
    "e36b4639-be8c-46f7-a8c9-bcfb134f9fd0": {
     "id": "e36b4639-be8c-46f7-a8c9-bcfb134f9fd0",
     "prev": "1519a4fa-1588-4644-98de-9c43bf0aceb5",
     "regions": {
      "4c1e9714-9ba0-45fd-8a2f-ef80a5c85c2e": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "6534d4dc-125f-47e6-a022-cf1e0d277174",
        "part": "whole"
       },
       "id": "4c1e9714-9ba0-45fd-8a2f-ef80a5c85c2e"
      }
     }
    },
    "e76e9a02-54c1-4ec9-80fb-c611ed398122": {
     "id": "e76e9a02-54c1-4ec9-80fb-c611ed398122",
     "prev": "90af992d-eb6d-4496-b2d2-6aa9a95b6a61",
     "regions": {
      "9491b84d-193b-40ff-9321-d21eb1ba88d4": {
       "attrs": {
        "height": 0.8,
        "width": 0.8,
        "x": 0.1,
        "y": 0.1
       },
       "content": {
        "cell": "b64ec10e-fa9d-4f3f-907f-6799611ed6b1",
        "part": "whole"
       },
       "id": "9491b84d-193b-40ff-9321-d21eb1ba88d4"
      }
     }
    }
   },
   "themes": {}
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}