-
Notifications
You must be signed in to change notification settings - Fork 0
/
kfolds.py
276 lines (243 loc) · 10.2 KB
/
kfolds.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
import random
import math
import copy
import lda
import logisticregression
import naivebayes
import csv_parser
#import sklearn.linear_model as linmod
#k is number of groups needed, data is a list of data rows
def splitdata(k, x, y):
if(len(x) != len(y)):
print "input lengths not the same... "
return
#init k groups
# x and y list
groups = []
for i in xrange(k):
groups.append([])
groups[i].append([])
groups[i].append([])
#copy data and distribute between groups!
data_copy_x = copy.deepcopy(x)
data_copy_y = copy.deepcopy(y)
for i in xrange(len(x)):
chosen_index = random.randint(0, len(data_copy_x) - 1)
groups[i%k][0].append(data_copy_x[chosen_index])
groups[i%k][1].append(data_copy_y[chosen_index])
del data_copy_x[chosen_index]
del data_copy_y[chosen_index]
return groups
#k is number of groups needed, data is a list of data rows
def kfolds_all_algos(k, x, y, isotest_x, isotest_y):
k_groups = splitdata(k, x, y)
#now we have the k groups, assign each one as test once and run tests!
print "groups split"
lda_train_results = []
lda_test_results = []
lda_iso_results = []
nb_train_results = []
nb_test_results = []
nb_iso_results = []
lr_train_results = []
lr_test_results = []
lr_iso_results = []
for i in xrange(k):
print "K Fold number " + str(i)
test = k_groups[i]
train = []
train.append([]) #x
train.append([]) #y
for j in xrange(k):
if(j != i):
train[0].extend(k_groups[j][0])
train[1].extend(k_groups[j][1])
#Now we have test and training data... what shall we do?
#train on LDA
print "Training LDA..."
(prob, mean, cov) = lda.trainLDA(copy.deepcopy(train[0]), copy.deepcopy(train[1]))
#print str(prob) + "\t" + str(mean) + "\t" + str(cov)
print "DONE training LDA."
print "Training NB..."
(py, theta) = naivebayes.trainNaiveBayesMN(copy.deepcopy(train[0]), copy.deepcopy(train[1]))
#print str(py) + "\t" + str(theta)
print "DONE training NB"
print "Training Logistic Regression..."
t_x = copy.deepcopy(train[0])
for i in xrange(len(t_x)):
temp_row = [1]
temp_row.extend(t_x[i])
t_x[i] = temp_row
(wvector, scales) = logisticregression.trainLogisticReg(0.01, 0.00001, 100, t_x, train[1])
#print str(wvector)
print "DONE training Logistic Regression.\n"
#lr_model = linmod.LogisticRegression()
#lr_model.fit(t_x, train[1])
#for model, name in ((lr_model, "LR"),):
# tp, tn, fp, fn = 0, 0, 0, 0
# for i in xrange(0, len(t_x)):
# val = model.predict(t_x[i])
# if (val == 1 and train[1][i] == 1):
# tp += 1
# elif (val == 1 and train[1][i] == 0):
# fp += 1
# elif (val == 0 and train[1][i] == 0):
# tn += 1
# elif (val == 0 and train[1][i] == 1):
# fn += 1
# print "%s - TP: %d, FP: %d, TN: %d, FN: %d" % (name, tp, fp, tn, fn)
#get Prediction Errors on left out set
lr_test_error = logisticregression.getConfusionMatrix(wvector,scales, copy.deepcopy(test[0]), copy.deepcopy(test[1]))
lr_train_error = logisticregression.getConfusionMatrix(wvector,scales, copy.deepcopy(train[0]), copy.deepcopy(train[1]))
lr_iso_error = logisticregression.getConfusionMatrix(wvector,scales, copy.deepcopy(isotest_x), copy.deepcopy(isotest_y))
lda_test_error = lda.getConfusionMatrix(prob, mean, cov, copy.deepcopy(test[0]), copy.deepcopy(test[1]))
lda_train_error = lda.getConfusionMatrix(prob, mean, cov, copy.deepcopy(train[0]), copy.deepcopy(train[1]))
lda_iso_error = lda.getConfusionMatrix(prob, mean, cov, copy.deepcopy(isotest_x), copy.deepcopy(isotest_y))
nb_test_error = naivebayes.getConfusionMatrixMN(py, theta, copy.deepcopy(test[0]), copy.deepcopy(test[1]))
nb_train_error = naivebayes.getConfusionMatrixMN(py, theta, copy.deepcopy(train[0]), copy.deepcopy(train[1]))
nb_iso_error = naivebayes.getConfusionMatrixMN(py, theta, copy.deepcopy(isotest_x), copy.deepcopy(isotest_y))
#add to sets the false positives (for now)
lr_train_results.append(lr_train_error)
lr_test_results.append(lr_test_error)
lr_iso_results.append(lr_iso_error)
lda_train_results.append(lda_train_error)
lda_test_results.append(lda_test_error)
lda_iso_results.append(lda_iso_error)
nb_train_results.append(nb_train_error)
nb_test_results.append(nb_test_error)
nb_iso_results.append(nb_iso_error)
#calc average training and test error for each algorithm
avr_lda_train = averageconfusionmatrix(lda_train_results)
avr_lda_test = averageconfusionmatrix(lda_test_results)
avr_lda_iso = averageconfusionmatrix(lda_iso_results)
avr_lr_train = averageconfusionmatrix(lr_train_results)
avr_lr_test = averageconfusionmatrix(lr_test_results)
avr_lr_iso = averageconfusionmatrix(lr_iso_results)
avr_nb_train = averageconfusionmatrix(nb_train_results)
avr_nb_test = averageconfusionmatrix(nb_test_results)
avr_nb_iso = averageconfusionmatrix(nb_iso_results)
return [avr_lr_train, avr_lr_test, avr_lr_iso, avr_lda_train, avr_lda_test, avr_lda_iso, avr_nb_train, avr_nb_test, avr_nb_iso]
#return [avr_lr_train, avr_lr_test, avr_lda_train, avr_lda_test, avr_nb_train, avr_nb_test]
def averageconfusionmatrix(listofmatrices):
tp = 0
tn = 0
fp = 0
fn = 0
for matrix in xrange(len(listofmatrices)):
tp += listofmatrices[matrix][0][0]
fp += listofmatrices[matrix][0][1]
tn += listofmatrices[matrix][1][1]
fn += listofmatrices[matrix][1][0]
tp = tp / float(len(listofmatrices))
fp = fp / float(len(listofmatrices))
tn = tn / float(len(listofmatrices))
fn = fn / float(len(listofmatrices))
return [tp, fp, tn, fn]
def getfeaturesubset(indices, x):
new_set = []
for row in x:
new_row = []
for col in xrange(len(row)):
if col in indices:
new_row.append(row[col])
new_set.append(new_row)
return new_set
#subsets chosen are hard coded
def testfeaturesubsets(k, x, y, iso_x, iso_y):
if(len(x) != len(y)):
print "data not same length."
return
results = {}
#dataset with features only about sponsor
temp_x_onlyspon = getfeaturesubset([0,1,2,3,4,5,6], x)
temp_x_onlyspon_iso = getfeaturesubset([0,1,2,3,4,5,6], iso_x)
results["dataset_only_sponsor"] = kfolds_all_algos(k, temp_x_onlyspon, y, temp_x_onlyspon_iso, iso_y)
#dataset with features only about congress composition
temp_x_congress = getfeaturesubset([0,7,8,9,10,11,12,13], x)
temp_x_congress_iso = getfeaturesubset([0,7,8,9,10,11,12,13], iso_x)
results["dataset_only_congress"] = kfolds_all_algos(k, temp_x_congress, y, temp_x_congress_iso, iso_y)
#full dataset
results["dataset_full"] = kfolds_all_algos(k, x, y, iso_x, iso_y)
#full dataset - personal info of sponsor
temp_x_nopers = getfeaturesubset([0,1,2,6,7,8,9,10,11,12,13], x)
temp_x_nopers_iso = getfeaturesubset([0,1,2,6,7,8,9,10,11,12,13], iso_x)
results["dataset_full_nopersonal"] = kfolds_all_algos(k, temp_x_nopers, y, temp_x_nopers_iso, iso_y)
return (results)#, temp_x_onlyspon, temp_x_congress, temp_x_nopers)
def accuracycalc(matrix):
return (matrix[0] + matrix[2])/float(matrix[0] + matrix[3] + matrix[1] + matrix[2])
def errorcalc(matrix):
return (matrix[1] + matrix[3])/float(matrix[0] + matrix[3] + matrix[1] + matrix[2])
def precisioncalc(matrix):
if(float(matrix[0] + matrix[1]) == 0):
return 0
return (matrix[0])/float(matrix[0] + matrix[1])
def recallcalc(matrix):
if(float(matrix[0] + matrix[3]) == 0):
return 0
return (matrix[0])/float(matrix[0] + matrix[3])
def fcalc(matrix):
precision = precisioncalc(matrix)
recall = recallcalc(matrix)
if precision == 0 and recall == 0:
return -1
return 2*(precision*recall)/(precision + recall)
def kfolds_control(k, x, y):
# set aside 20% to do final testing on
if(len(x) != len(y)):
print "data not right size"
return
# Shuffle the training data.
combined = zip(x, y)
random.shuffle(combined)
x, y = zip(*combined)
x = list(x)
y = list(y)
testset = [[], []] #x and y
sizeoftest = int(0.1 * len(x))
for i in xrange(sizeoftest):
chosen_index = random.randint(0, len(x) - 1)
testset[0].append(x[chosen_index])
testset[1].append(y[chosen_index])
del x[chosen_index]
del y[chosen_index]
#we have test set, now create training set with 50/50 balance
num_positives = 0
train_x = []
train_y = []
oversample_rate = 2
for i in xrange(len(y)):
if(y[i] == 1):
for _ in xrange(oversample_rate):
train_x.append(x[i])
train_y.append(y[i])
num_positives = num_positives + 1
print "number of positives: " + str(num_positives)
#now compliment with fails
# Randomize choosing the negatives so they are spread out over time.
while (num_positives > 0):
i = random.randint(0, len(x)-1)
if(y[i] == 0):
train_x.append(x[i])
train_y.append(y[i])
del x[i]
del y[i]
num_positives = num_positives - 1
print "size of x: " + str(len(train_x)) + " size of y: " + str(len(train_y)) + "size of testset: " + str(len(testset[0]))
#start off by doing feature selection
#(results, temp_x_onlyspon, temp_x_congress, temp_x_nopers) = testfeaturesubsets(k, train_x, train_y, testset[0], testset[1])
(results) = testfeaturesubsets(k, train_x, train_y, testset[0], testset[1])
for set_name, algo_data in results.iteritems():
print (set_name + "\n LR train" + "\t" + str(algo_data[0]) + "\t accuracy: " + str(accuracycalc(algo_data[0])) + "\t accuracy: " + str(fcalc(algo_data[0]))
+ "\n LR valid" + "\t" + str(algo_data[1]) + "\t A: " + str(accuracycalc(algo_data[1])) + "\t F: " + str(fcalc(algo_data[1]))
+ "\n LR isolated" + "\t" + str(algo_data[2]) + "\t Acuracy: " + str(accuracycalc(algo_data[2])) + "\t F: " + str(fcalc(algo_data[2]))
+ "\n LDA train" + "\t" + str(algo_data[3]) + "\t A: " + str(accuracycalc(algo_data[3])) + "\t F: " + str(fcalc(algo_data[3]))
+ "\n LDA valid" + "\t" + str(algo_data[4]) + "\t A: " + str(accuracycalc(algo_data[4])) + "\t F: " + str(fcalc(algo_data[4]))
+ "\n LDA isolated" + "\t" + str(algo_data[5]) + "\t A: " + str(accuracycalc(algo_data[5])) + "\t F: " + str(fcalc(algo_data[5]))
+ "\n NB train" + "\t" + str(algo_data[6]) + "\t A: " + str(accuracycalc(algo_data[6])) + "\t F: " + str(fcalc(algo_data[6]))
+ "\n NB valid" + "\t" + str(algo_data[7]) + "\t A: " + str(accuracycalc(algo_data[7])) + "\t F: " + str(fcalc(algo_data[7]))
+ "\n NB isolated" + "\t" + str(algo_data[8]) + "\t A: " + str(accuracycalc(algo_data[8])) + "\t F: " + str(fcalc(algo_data[8]))
+ "\n")
print "parsing..."
(x,y) = csv_parser.parse_data("final_data.csv")
print "parsed"
kfolds_control(4, x, y)