Update code base to support per-computed math.log

MicroHMM/hmm.py

@@ -1,7 +1,9 @@
 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
+import math
 import pickle
 import pathlib
+from typing import List, Union, Tuple
 
 from MicroHMM.viterbi import Viterbi
 
@@ -23,6 +25,8 @@ def __init__(self, A=None, B=None, vocabulary=None):
         self.state_observation_pair = {}  # count of pair state and emission observation
 
     def train_one_line(self, list_of_word_tag_pair):
+        # type(List[Union[List[str, str], Tuple[str, str]]) -> None
+
         """
         train model from one line data
         :param list_of_word_tag_pair: list of tuple (word, tag)
@@ -89,18 +93,22 @@ def do_train(self):
                 if previous_state not in self.A:
                     self.A[previous_state] = {}
 
-                self.A[previous_state][state] = bigram_probability
+                self.A[previous_state][state] = math.log(bigram_probability)
 
             # compute emission probability
             # NOTE: using dict.get() to prevent start state have on emission will cause exeception
             emission_local_storage = self.state_observation_pair.get(previous_state, {})
             for word, word_count in emission_local_storage.items():
+                emmit_probability = word_count / previous_state_count
+
                 if previous_state not in self.B:
                     self.B[previous_state] = {}
 
-                self.B[previous_state][word] = word_count / previous_state_count
+                self.B[previous_state][word] = math.log(emmit_probability)
 
     def predict(self, word_list, output_graphml_file=None):
+        # type: (List[str], Union[str, None]) -> List[Tuple[str, str]]
+
         if not self.A:  # using self.A as an training-flag indicate if already trained.
             self.do_train()
 

MicroHMM/viterbi.py

@@ -1,5 +1,4 @@
 # -*- coding: utf-8 -*-
-
 import math
 
 import networkx as nx
@@ -18,7 +17,7 @@ def __init__(self, A, B, vocabulary, start_state='<start>', end_state='<end>', v
 
         # Trk part for robust model
         # TODO: why using a very small number than average_emission_probability can archive better performance?
-        self.very_small_probability = very_small_probability
+        self.very_small_probability = math.log(very_small_probability)
 
         # create networkx graph
         self.G = nx.Graph()
@@ -44,13 +43,9 @@ def _do_predict(self, word_list):
             # path_probability = transition_probability * state_observation_likelihood
 
             # using log(probability) as probability to prevent number to be too small
-            transition_probability = math.log(
-                self.A[self.start_state].get(state, self.very_small_probability)
-            )
+            transition_probability = self.A[self.start_state].get(state, self.very_small_probability)
 
-            state_observation_likelihood = math.log(
-                self.B[state].get(word, self.very_small_probability)
-            )
+            state_observation_likelihood = self.B[state].get(word, self.very_small_probability)
 
             path_probability = transition_probability + state_observation_likelihood
 
@@ -88,13 +83,9 @@ def _do_predict(self, word_list):
                     # path_probability = previous_path_probability * transition_probability * state_observation_likelihood
 
                     # using log(probability) as probability to prevent number to be too small
-                    transition_probability = math.log(
-                        self.A[i].get(state, self.very_small_probability)
-                    )
+                    transition_probability = self.A[i].get(state, self.very_small_probability)
 
-                    state_observation_likelihood = math.log(
-                        self.B[state].get(word, self.very_small_probability)
-                    )
+                    state_observation_likelihood = self.B[state].get(word, self.very_small_probability)
 
                     path_probability = previous_path_probability + transition_probability + state_observation_likelihood
 
@@ -138,9 +129,7 @@ def _do_predict(self, word_list):
             # path_probability = previous_path_probability * transition_probability
 
             # using log(probability) as probability to prevent number to be too small
-            transition_probability = math.log(
-                self.A[i].get(self.end_state, self.very_small_probability)
-            )
+            transition_probability = self.A[i].get(self.end_state, self.very_small_probability)
 
             path_probability = previous_path_probability + transition_probability
 

makefile

@@ -1,14 +1,33 @@
-.PHONY: build
-build:
-	python ./setup.py sdist bdist_wheel
+.PHONY: clean clean-test clean-pyc clean-build docs help
+.DEFAULT_GOAL := help
 
-.PHONY: upload_test
-upload_test:
-	twine upload --repository-url https://test.pypi.org/legacy/ dist/*
+define BROWSER_PYSCRIPT
+import os, webbrowser, sys
 
-.PHONY: upload
-upload:
-	twine upload dist/*
+try:
+	from urllib import pathname2url
+except:
+	from urllib.request import pathname2url
+
+webbrowser.open("file://" + pathname2url(os.path.abspath(sys.argv[1])))
+endef
+export BROWSER_PYSCRIPT
+
+define PRINT_HELP_PYSCRIPT
+import re, sys
+
+for line in sys.stdin:
+	match = re.match(r'^([a-zA-Z_-]+):.*?## (.*)$$', line)
+	if match:
+		target, help = match.groups()
+		print("%-20s %s" % (target, help))
+endef
+export PRINT_HELP_PYSCRIPT
+
+BROWSER := python -c "$$BROWSER_PYSCRIPT"
+
+help:
+	@python -c "$$PRINT_HELP_PYSCRIPT" < $(MAKEFILE_LIST)
 
 clean: clean-build clean-pyc clean-test ## remove all build, test, coverage and Python artifacts
 
@@ -31,6 +50,32 @@ clean-test: ## remove test and coverage artifacts
 	rm -fr htmlcov/
 	rm -fr .pytest_cache
 
+lint: ## check style with flake8
+	flake8 MicroTokenizer tests
+
+test: ## run tests quickly with the default Python
+	py.test
+
+test-all: ## run tests on every Python version with tox
+	tox
+
+coverage: ## check code coverage quickly with the default Python
+	coverage run --source MicroTokenizer -m pytest
+	coverage report -m
+	coverage html
+	$(BROWSER) htmlcov/index.html
+
+docs: ## generate Sphinx HTML documentation, including API docs
+	rm -f docs/MicroTokenizer.rst
+	rm -f docs/modules.rst
+	sphinx-apidoc -o docs/ MicroTokenizer
+	$(MAKE) -C docs clean
+	$(MAKE) -C docs html
+	$(BROWSER) docs/_build/html/index.html
+
+servedocs: docs ## compile the docs watching for changes
+	watchmedo shell-command -p '*.rst' -c '$(MAKE) -C docs html' -R -D .
+
 release: dist ## package and upload a release
 	twine upload dist/*
 
@@ -39,6 +84,9 @@ dist: clean ## builds source and wheel package
 	python setup.py bdist_wheel
 	ls -l dist
 
+install: clean ## install the package to the active Python's site-packages
+	python setup.py install
+
 .PHONY: update_minor_version
 update_minor_version:
 	punch --part minor
@@ -49,4 +97,4 @@ update_patch_version:
 
 .PHONY: update_major_version
 update_major_version:
-	punch --part major
+	punch --part major

requirements.txt

@@ -1 +1,3 @@
 networkx==2.1
+pathlib;python_version<"3.4"
+pytest

requirements_dev.txt

@@ -1 +1,4 @@
-pytest==3.5.1
+pytest==3.5.1
+
+# for pytest helper
+pytest-helpers-namespace

tests/conftest.py

@@ -0,0 +1,19 @@
+# -*- coding: utf-8 -*-
+
+pytest_plugins = ['helpers_namespace']
+
+import pytest
+
+
+@pytest.helpers.register
+def train_test_cases():
+    basic_test_case = [('A', 'a'), ('B', 'b')]
+
+    test_cases = [
+    ]
+
+    for i in range(1, 4):
+        test_cases.append([basic_test_case] * i)
+
+    # return as list of list
+    return test_cases

tests/test_hmm_train.py

@@ -0,0 +1,21 @@
+#!/bin/env python
+# -*- coding: utf-8 -*-
+import pytest
+
+from MicroHMM.hmm import HMMModel
+
+
+@pytest.mark.parametrize("train_data", pytest.helpers.train_test_cases())
+def test_hmm_train(train_data):
+    first_train_data = train_data[0]
+
+    hmm_model = HMMModel()
+
+    for single_train_data in train_data:
+        hmm_model.train_one_line(single_train_data)
+
+    hmm_model.do_train()
+
+    result = hmm_model.predict([i[0] for i in first_train_data])
+
+    assert first_train_data == result