-
Notifications
You must be signed in to change notification settings - Fork 0
/
lr.py
364 lines (306 loc) · 13.1 KB
/
lr.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
# @Author: Atul Sahay <atul>
# @Date: 2018-08-07T18:09:32+05:30
# @Email: atulsahay01@gmail.com
# @Filename: model1.py
# @Last modified by: atul
# @Last modified time: 2018-08-19T20:37:14+05:30
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
import sys
import os
import math
"""
You are allowed to change the name of function arguments as per your convinience,
but it should be meaningful.
Like : y, y_train, y_test, output_var, target, output_label, ... are accepted
but do not keep it like : abc, a, b, etc.
Same applies to variable naming also
"""
# Provide Map for non int data (post_day,basetime_day)
def to_map(data_set):
data_set = pd.concat([data_set,pd.get_dummies(data_set['post_day'], prefix='post_day')],axis=1)
# now drop the original
data_set.drop(['post_day'],axis=1, inplace=True)
data_set = pd.concat([data_set,pd.get_dummies(data_set['basetime_day'], prefix='basetime_day')],axis=1)
# now drop the original
data_set.drop(['basetime_day'],axis=1, inplace=True)
return data_set
# Feature scaling , here I have used min_max
def to_normalize(data_set):
global train_mean, train_std
train_mean = data_set.mean()
train_std = data_set.std()
data_set = (data_set - data_set.min())/(data_set.max() - data_set.min())
return data_set
# Split in x and y
def split(data):
x_train = data.iloc[:,:-1]
y_train = data.iloc[:,-1]
return x_train, y_train
# Provide features set and target set
def get_features(file_path):
# Given a file path , return feature matrix and target labels
data = pd.read_csv(file_path)
return split(data)
# mean square Error
def mean_square_error(x,y_label,theta,lam):
m = x.shape[0]
hypothesis = np.dot(x, theta)
loss = hypothesis - y_label
cost = np.sum(loss ** 2)
cost_reg = cost + lam*np.sum(np.power(theta,2))
return cost_reg/m
############################################# Another Variant of Gradient##################
def gradientDescent(x, y_label, theta, alpha, Iterations,lam,x_valid,y_valid,error):
m = x.shape[0]
xTrans = x.transpose()
cost_valid_old = math.inf
train_cost_old = math.inf
best_theta = theta
for i in range(Iterations):
hypothesis = np.dot(x, theta)
loss = hypothesis - y_label
train_cost_curr = mean_square_error(x,y_label,theta,lam)
cost_valid_curr = mean_square_error(x_valid,y_valid,theta,lam)
if(i%100==0):
print("\nIteration %d | Train_cost_reg %0.18f \nCost_valid_old %0.18f \nCost_valid_curr %0.18f" % (i, train_cost_curr,cost_valid_old,cost_valid_curr))
# breaking iteration if breaking criteria met
if(cost_valid_curr>cost_valid_old):
print("\nStopped due to cost_valid criteria met")
break
if((train_cost_old-train_cost_curr)<error):
print("\nStopped due to train_error criteria met")
break
cost_valid_old = cost_valid_curr
train_cost_old = train_cost_curr
gradient = np.dot(xTrans, loss)
# print(gradient)
gradient = gradient + lam*theta
theta = theta - 2*alpha*gradient/m
best_theta = theta
print("\nIteration %d | Cost_reg %0.18f \nCost_valid_old %0.18f \nCost_valid_curr %0.18f\
\nlambda: %0.6f \tlearn:%0.6f " % (i, train_cost_curr,cost_valid_old,cost_valid_curr,lam,alpha))
print("\nDiff:%0.18f"%abs(train_cost_old-cost_valid_old))
#returning the previous iteration weight
return best_theta
#######################################################################
def generate_output(phi_test, weights):
# # writes a file (output.csv) containing target variables in required format for Kaggle Submission.
print("Generating the output file:--")
df = pd.DataFrame(columns=['target'])
# y_P_list = []
# idList = [ i for i in range(int(len(phi_test)))]
for i in range(int(len(phi_test))):
# print(phi_test[i])
y_pred = weights.dot(phi_test[i])
if(y_pred<0):
y_pred = 0
# y_pred = int(round(y_pred))
df.loc[i] = np.array([y_pred])
df.to_csv('output1.csv')
print("Done")
#################################################### Task -3 #############################
# mean square Error - norm P
def mean_square_errorP(x,y_label,theta,lam,p):
m = x.shape[0]
hypothesis = np.dot(x, theta)
loss = hypothesis - y_label
cost = np.sum(loss ** 2)
cost_reg = cost + lam*(np.sum(np.power(theta,p)))**(1/p)
return cost_reg/m
# Gradient descent for p norm
def gradientDescentP(x, y_label, theta, alpha, Iterations,lam,x_valid,y_valid,error,p):
m = x.shape[0]
xTrans = x.transpose()
cost_valid_old = math.inf
train_cost_old = math.inf
# print(x_valid.shape[0])
best_theta = theta
for i in range(Iterations):
hypothesis = np.dot(x, theta)
loss = hypothesis - y_label
train_cost_curr = mean_square_errorP(x,y_label,theta,lam,p)
cost_valid_curr = mean_square_errorP(x_valid,y_valid,theta,lam,p)
if(i%100==0):
print("\nIteration %d | Train_cost_reg %0.18f \nCost_valid_old %0.18f \nCost_valid_curr %0.18f" % (i, train_cost_curr,cost_valid_old,cost_valid_curr))
# breaking iteration if halting criteria met
if(cost_valid_curr>cost_valid_old):
print("\nStopped due to cost_valid criteria met")
break
if((train_cost_old-train_cost_curr)<error):
print("\nStopped due to train_error criteria met")
break
cost_valid_old = cost_valid_curr
train_cost_old = train_cost_curr
gradient = np.dot(xTrans, loss)
# print(gradient)
theta_P_norm = (np.sum(np.power(theta,p)))**((1-p)/p)
gradient = gradient + lam*theta_P_norm*theta
theta = theta - 2*alpha*gradient/m
best_theta = theta
print("\nIteration %d | Cost_reg %0.18f \nCost_valid_old %0.18f \nCost_valid_curr %0.18f\
\nlambda: %0.6f \tlearn:%0.6f\t P-Norm = %d" % (i, train_cost_curr,cost_valid_old,cost_valid_curr,lam,alpha,p))
print("\nDiff:%0.18f"%abs(train_cost_old-cost_valid_old))
#returning the previous iteration weight
return best_theta
################################################### Task 3 ends here ######################
################################################## Task 4 starts(To report mse on basis of x) ##########################
###### sigmoidal function ############
def sigmoidal(value):
# print(value)
a = math.exp(value) + 1
return 1/a
##### gaussian function #############
def gaussian(x, mu, sig):
# print(x,mu,sig)
return np.exp(-np.power(x - mu, 2) / (2 * np.power(sig, 2)))
################################################# Task 4 Ends ##########################################################
def generatePlots(x, y, theta):
# """
# generates and saves plots of top three features with target variable.
# Note: Procedure to obtain top features is important
# """
theta = pd.Series(theta).apply(abs)
top3 = theta.nlargest(3)
indices = list(top3.index[:])
top3_features = [x.columns[i] for i in indices]
print("Top 3 Features:")
for i,j in enumerate(top3_features):
print("%d. %s"%(i+1,j))
for i in top3_features:
# plotting points as a scatter plot
plt.scatter(x[i], y, label= "likes", color= "red",
marker= "*", s=10)
# x-axis label
plt.xlabel(i)
# frequency label
plt.ylabel("Target")
# plot title
plt.title('Scatter plot!')
# showing legend
plt.legend()
# function to show the plot
plt.show()
def main():
"""
Calls functions required to do tasks in sequence
say :
train_file = first_argument
test_file = second_argument
x_train, y_train = get_features();
task1();task2();task3();.....
"""
global train_mean, train_std
# pow = 1
train_file = sys.argv[1]
# train_file = '/home/atul/college/cs725/Assignment/train.csv'
test_file = sys.argv[2]
# test_file = '/home/atul/college/cs725/Assignment/test.csv'
print("Reading Files...")
x_test = pd.read_csv(test_file)
x_train, y_train = get_features(train_file)
print("Done")
################################## Mapping days to one hot vector################
x_train = to_map(x_train)
x_test = to_map(x_test)
##############################################################################
####################### Normalizing the data points##############################
x_train = to_normalize(x_train)
# To take validation set out in proportion of 20-80 #############################
indexes = int(0.80*x_train.shape[0])
x_train, x_valid = x_train.iloc[:indexes], x_train.iloc[indexes:]
y_train, y_valid = y_train.iloc[:indexes], y_train.iloc[indexes:]
####################### Normalizing the data points##############################
x_test = to_normalize(x_test)
x_test.promotion = x_test.promotion.fillna(0)
####################################################
#Appending a series of Ones for bias in x_train
ones = np.ones(x_train.shape[0])
x_train.insert(loc=x_train.shape[1], column='Ones', value=ones)
#Appending a series of Ones for bias in x_valid
ones = np.ones(x_valid.shape[0])
x_valid.insert(loc=x_valid.shape[1], column='Ones', value=ones)
#Appending a series of ones for bias in x_test
ones = np.ones(x_test.shape[0])
x_test.insert(loc=x_test.shape[1], column='Ones', value=ones)
################################################################
############################## Weights are intialised ####################
weights = np.ones(x_train.shape[1])
#########################################################################
#hyperparameters used
learn = 0.001
iterations = 100000
lam = 10000
error = 10**(-18)
p = 6
print("::::::::::::::LINEAR REGRESSION::::::::::::::::")
print("1. L2 regularisation")
print("2. P-Norm regularisation")
print("3. L2 regularisation with basis")
print("4. Optimised one(P-6 with sigmoidal basis)")
choice = int(input("Your choice: "))
if(choice==1):
lam = float(input("lambda value : "))
learn = float(input("Learn rate :"))
iterations = int(input("No. of iterations: "))
###### Task -1
weights = gradientDescent(x_train.values, y_train.values, weights, learn, iterations,lam,x_valid,y_valid,error)
elif(choice==2):
p = int(input("P-Norm [4,6]: "))
lam = float(input("lambda value : "))
learn = float(input("Learn rate :"))
iterations = int(input("No. of iterations: "))
###### TASK -3
#
weights = gradientDescentP(x_train.values, y_train.values, weights, learn, iterations,lam,x_valid,y_valid,error,p)
elif(choice==3):
print("1. Inverse Sigmoidal")
print("2. Gaussian Function")
basis = int(input("Your Choice: "))
lam = float(input("lambda value : "))
learn = float(input("Learn rate :"))
iterations = int(input("No. of iterations: "))
###### TASK -4
if(basis==1):
print("Inverse Sigmoidal")
x_train.iloc[:,:-1] = x_train.iloc[:,:-1].applymap(sigmoidal)
x_valid.iloc[:,:-1] = x_valid.iloc[:,:-1].applymap(sigmoidal)
x_test.iloc[:,:-1] = x_test.iloc[:,:-1].applymap(sigmoidal)
elif(basis==2):
print("Gaussian Function")
x_tr_m, x_tr_s = x_train.mean(), x_train.std()
x_v_m, x_v_s = x_valid.mean(), x_valid.std()
x_te_m, x_te_s = x_test.mean(), x_test.std()
for i in range(len(x_tr_m)):
x_train.iloc[:,i] = x_train.iloc[:,i].apply(lambda x: gaussian(x,x_tr_m[i],x_tr_s[i]))
x_valid.iloc[:,i] = x_valid.iloc[:,i].apply(lambda x: gaussian(x,x_v_m[i],x_v_s[i]))
x_test.iloc[:,i] = x_test.iloc[:,i].apply(lambda x: gaussian(x,x_te_m[i],x_te_s[i]))
x_train = x_train.fillna(0)
x_valid = x_valid.fillna(0)
x_test = x_test.fillna(0)
else:
exit("Error: Invalid Input")
weights = gradientDescent(x_train.values, y_train.values, weights, learn, iterations,lam,x_valid,y_valid,error)
elif(choice==4):
print("Basis: Inverse Sigmoid Norm: P=6")
lam = float(input("lambda value : "))
learn = float(input("Learn rate :"))
iterations = int(input("No. of iterations: "))
##### TASK -5
x_train.iloc[:,:-1] = x_train.iloc[:,:-1].applymap(sigmoidal)
x_valid.iloc[:,:-1] = x_valid.iloc[:,:-1].applymap(sigmoidal)
x_test.iloc[:,:-1] = x_test.iloc[:,:-1].applymap(sigmoidal)
p = 6
weights = gradientDescentP(x_train.values, y_train.values, weights, learn, iterations,lam,x_valid,y_valid,error,p)
else:
exit("Error: Invalid Input")
###### Generate target file
generate_output(x_test.values,weights)
choice = input("Want to see Top 3 Features: [y/n]")
if(choice=="y"):
###### TASK -2
generatePlots(x_train, y_train, weights)
#################### Driver Function
if __name__ == '__main__':
main()