-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathfinalClassifier.py
271 lines (220 loc) · 13.6 KB
/
finalClassifier.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
import csv
from utilities.readTweet import readCSV
from sklearn import svm
import numpy as np
import theano
import theano.tensor as T
import lasagne, pydot
from sklearn.metrics import f1_score
from scipy.stats import pearsonr
import matplotlib.pyplot as plt
from matplotlib.mlab import PCA
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
import lime
import lime.lime_tabular
def build_mlp(inpSize, input_var=None):
l_in = lasagne.layers.InputLayer(shape=(None,inpSize), input_var=input_var)
l_hid1 = lasagne.layers.DenseLayer(l_in, num_units=75, nonlinearity=lasagne.nonlinearities.sigmoid, W=lasagne.init.GlorotUniform())
#l_hid2 = lasagne.layers.DenseLayer(l_hid1, num_units=10, nonlinearity=lasagne.nonlinearities.sigmoid)
l_out = lasagne.layers.DenseLayer(l_hid1, num_units=1, nonlinearity=lasagne.nonlinearities.sigmoid)
return l_out
def initNetwork(numFeatures):
input_var = T.matrix('inputs')
target_var = T.ivector('targets')
network = build_mlp(numFeatures, input_var)
prediction = T.transpose(lasagne.layers.get_output(network))
loss = lasagne.objectives.squared_error(prediction, target_var)
loss = loss.mean()
params = lasagne.layers.get_all_params(network, trainable=True)
updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=0.01, momentum=0.9)
train_fn = theano.function([input_var, target_var], loss, updates=updates, allow_input_downcast=True)
test_prediction = T.transpose(lasagne.layers.get_output(network, deterministic=True))
test_acc = T.mean(T.eq(test_prediction > 0.5, target_var), dtype=theano.config.floatX)
test_loss = lasagne.objectives.squared_error(test_prediction, target_var)
val_fn = theano.function([input_var, target_var], [test_loss, test_acc, test_prediction], allow_input_downcast=True)
#theano.printing.pydotprint(train_fn, 'pic.png', var_with_name_simple = True)
return [train_fn, val_fn]
def findCutoff(pred, Y, inc):
idealTh = []; maxMatch = -1
for th in np.arange(0,1,inc):
predTh = (pred>th)[0]
matches = sum([1 if ((Y[i]==1 and predTh[i]==True) or (Y[i]==0 and predTh[i]==False)) else 0 for i in range(len(Y))])
if maxMatch <= matches:
if maxMatch == matches:
idealTh += [th]
else:
idealTh = [th]
maxMatch = matches
return (max(idealTh)+min(idealTh))/2.
def findCorrelation(XTrain):
t = np.asarray(XTrain)
t = [[abs(pearsonr(t[:,featureId1], t[:,featureId2])[0]) for featureId2 in xrange(t.shape[1])] for featureId1 in xrange(t.shape[1])]
plt.pcolor(t,cmap=plt.cm.Reds); plt.show(); plt.close()
def getFold(featureDict, userFoldDict, foldid, cond, useNgram):
#return [featureDict[user] for user in userFoldDict if cond(userFoldDict[user], foldid)]
loc2 = 'resultsDump/varun/trigram_prediction/'
loc1 = 'resultsDump/varun/unigram_prediction/'
retList = []
for user in userFoldDict:
if cond(userFoldDict[user], foldid):
if useNgram:
if cond(1,1) == True: #its test group
uni = csvReader(loc1 + str(foldid) + '_fold_test_ngram.csv', [1])
tri = csvReader(loc2 + str(foldid) + '_fold_test_ngram.csv', [1])
else: #train group
uni = csvReader(loc1 + str(foldid) + '_fold_train_ngram.csv', [1])
tri = csvReader(loc2 + str(foldid) + '_fold_train_ngram.csv', [1])
retList += [featureDict[user] + uni[user] + tri[user]]
#print len(featureDict[user] + uni[user] + tri[user])
else:
retList += [featureDict[user]]
return retList
def findAccuracyF1MLP(pred, Y, th):
predTh = (pred>th)[0]
return [sum([1 if ((Y[i]==1 and predTh[i]==True) or (Y[i]==0 and predTh[i]==False)) else 0 for i in range(len(Y))])/(len(YTest)+0.), f1_score(Y, [1 if i == True else False for i in predTh])]
def csvReader(csvLoc, colMask): #input is csv file loc and the masking array. out is a dict. key is username, value is a list of features that are not masked out.
return {row[0]: [float(row[num+1]) for num in range(len(row[1:])) if colMask != None and colMask[num] == 1] for row in csv.reader(open(csvLoc, 'rb'), delimiter=',')} #if colMask is None, that feature is not considered (note if its None, then colMask[num] is not evaluated, hence no error happens)
def getAllFeatures(csvList, useFeatures):
featuresDicts = [csvReader('resultsDump/allCSVs/' + csvFile, useFeatures.get(csvFile, None)) for csvFile in csvList]
return {k:sum([dic[k] for dic in featuresDicts], []) for k in featuresDicts[0].keys()} #first make a list of lists of features for each user, then flatten it using sum(L,[])
def normFeatParams(X):
a = np.asarray(X)
return [np.mean(a,0), np.std(a,0)]
def normFeat(X, mn, s):
return [[(userFt[ftid]-mn[ftid])/s[ftid] for ftid in range(len(userFt))] for userFt in X]
def getAccuracy(gt, pred):
return sum([1 for i in range(len(gt)) if gt[i]==pred[i]])/(len(gt)+0.)
def retainPerc(sortedList, perc = 0.99):
t = np.cumsum(sortedList)
for i in range(len(t)):
if t[i] >= perc: return i
return len(sortedList)
def randomShuffle(x, y):
idx = np.random.permutation(len(x))
return ([x[i] for i in idx], [y[i] for i in idx])
#this file is written assuming that there are 2 csvs for each feature (control and sch)
csvList = [['control_favorite_count.csv', 'control_simpleconnotation_features.csv', 'control_user_favourites_count.csv', 'control_user_followers_count.csv', 'control_user_friends_count.csv', 'control_user_statuses_count.csv', 'emoticonFeaturesCtrl.csv', 'RhymeFeaturesCtrl.csv', 'RhymeFeaturesCtrl1.csv', 'control_simplesentimentAFINN_features.csv', 'FrazierControl.csv', 'YngveControl.csv', 'control_schcount.csv', 'CPIDRScoreControl.csv', 'control_liwc_count.csv'],
['sch_favorite_count.csv', 'sch_simpleconnotation_features.csv', 'sch_user_favourites_count.csv', 'sch_user_followers_count.csv', 'sch_user_friends_count.csv', 'sch_user_statuses_count.csv', 'emoticonFeaturesSch.csv', 'RhymeFeaturesSch.csv', 'RhymeFeaturesSch1.csv', 'sch_simplesentimentAFINN_features.csv', 'FrazierSch.csv', 'YngveSch.csv', 'sch_schcount.csv', 'CPIDRScoreSchiz.csv', 'sch_liwc_count.csv']]
useFeatures = {'control_favorite_count.csv':[0,0,1,0,1], 'sch_favorite_count.csv':[0,0,1,0,1],
'control_simpleconnotation_features.csv':[1,1,1,1,1,1,0], 'sch_simpleconnotation_features.csv':[1,1,1,1,1,1,0],
'control_user_favourites_count.csv':[1,0,1,1,0], 'sch_user_favourites_count.csv':[1,0,1,1,0],
'control_user_followers_count.csv':[0,0,0,0,0], 'sch_user_followers_count.csv':[0,0,0,0,0],
'control_user_friends_count.csv':[0,0,0,0,0], 'sch_user_friends_count.csv':[0,0,0,0,0],
'control_user_statuses_count.csv':[1,0,1,1,0], 'sch_user_statuses_count.csv':[1,0,1,1,0],
'emoticonFeaturesCtrl.csv':[1,0,1,1,0], 'emoticonFeaturesSch.csv':[1,0,1,1,0],
'RhymeFeaturesCtrl.csv':[1]*4, 'RhymeFeaturesSch.csv':[1]*4,
'RhymeFeaturesCtrl1.csv':[1]*4, 'RhymeFeaturesSch1.csv':[1]*4,
'control_simplesentimentAFINN_features.csv':[0]*11+[1,0,1,0], 'sch_simplesentimentAFINN_features.csv':[0]*11+[1,0,1,0],
#'control_simplesentimentAFINN_features.csv':[0]*15, 'sch_simplesentimentAFINN_features.csv':[0]*15,
'FrazierControl.csv':[1,1,1,0,0,0,0,0], 'FrazierSch.csv':[1,1,1,0,0,0,0,0],
'YngveControl.csv':[1,0,1,1,0,0,0,0], 'YngveSch.csv':[1,0,1,1,0,0,0,0],
'control_schcount.csv':[1]*7, 'sch_schcount.csv':[1]*7,
#'control_schcount.csv':[0]*7, 'sch_schcount.csv':[0]*7,
'CPIDRScoreControl.csv':[1,0], 'CPIDRScoreSchiz.csv': [1,0],
# 'control_liwc_count.csv':[1,1,1,1,0,1,1,1,1,1,1,1,1,0,0,1,0,0,1,1,1,0,1,1,1,1,1,1,1,1,0,1,0,0,0,1,1,1,0,0,1,1,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,1,0],
# 'sch_liwc_count.csv':[1,1,1,1,0,1,1,1,1,1,1,1,1,0,0,1,0,0,1,1,1,0,1,1,1,1,1,1,1,1,0,1,0,0,0,1,1,1,0,0,1,1,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,1,0]
#'control_liwc_count.csv':[0]*64, 'sch_liwc_count.csv':[0]*64
}
#using only AFINN features
#csvList = [['control_simplesentimentAFINN_features.csv'], ['sch_simplesentimentAFINN_features.csv']]
#useFeatures = {'control_simplesentimentAFINN_features.csv':[0]+[1]*14, 'sch_simplesentimentAFINN_features.csv':[0]+[1]*14}
#print csvReader('resultsDump/allCSVs/' + csvList[0][0], useFeatures[csvList[0][0]])
control = getAllFeatures(csvList[0], useFeatures) #{'u1':[1,3,2,4,3...], 'u2':[...], ...}
sch = getAllFeatures(csvList[1], useFeatures)
#print len(control) #137
#print len(control['bebChK7PskxB']) #24
#print control['bebChK7PskxB'] #24 features in all
csvFileLoc = '../data/clpsych2015/schizophrenia/anonymized_user_manifest.csv'
#allControlPerFold = {foldid:[user['anonymized_name'] for user in readCSV(csvFileLoc, {'condition':'control','fold':[foldid]})] for foldid in range(0,10)} #its a dictionary with 10 keys 0 to 9 (the folds) and the values are lists of user names in that fold: {0:['u1', 'u2'], 1:[...], ...}
#allSchPerFold = {foldid:[user['anonymized_name'] for user in readCSV(csvFileLoc, {'condition':'schizophrenia','fold':[foldid]})] for foldid in range(0,10)}
#print allControlPerFold
#print allSchPerFold
controlUserFoldDict = {user['anonymized_name']:user['fold'] for user in readCSV(csvFileLoc, {'condition':'control'})} #dict of length 137, with users as keys and their folds as values
schUserFoldDict = {user['anonymized_name']:user['fold'] for user in readCSV(csvFileLoc, {'condition':'schizophrenia'})}
#print controlUserFoldDict
doPCA = False
useNgram = True
numFeatures = len(control[control.keys()[0]]) + (0,2)[useNgram]
accuracySVM = []; accuracyMLP = []; accuracyADA = []
f1SVM = []; f1MLP = []; f1ADA = []
num_epochs = 5000
print 'numFeatures', numFeatures
for foldid in range(10):
controlTest = getFold(control, controlUserFoldDict, foldid, lambda x,y:x==y, useNgram)
controlTrain = getFold(control, controlUserFoldDict, foldid, lambda x,y:x!=y, useNgram)
schTest = getFold(sch, schUserFoldDict, foldid, lambda x,y:x==y, useNgram)
schTrain = getFold(sch, schUserFoldDict, foldid, lambda x,y:x!=y, useNgram)
XTrain, YTrain = randomShuffle(controlTrain + schTrain, [1]*len(controlTrain) + [0]*len(schTrain))
#findCorrelation(XTrain) #plots graph of feature correlations
#[meanFt, varFt] = normFeatParams(XTrain) #both meanFt and varFt are of length = numberoffeatures
#XTrain = normFeat(XTrain, meanFt, varFt)
PCAObject = PCA(np.asarray(XTrain))
XTrain = PCAObject.center(XTrain)
if doPCA:
numFeatures = retainPerc(PCAObject.fracs, 0.99)
XTrain = PCAObject.project(XTrain)[:,0:numFeatures]
[meanFt, varFt] = normFeatParams(XTrain) #both meanFt and varFt are of length = numberoffeatures
XTrain = np.asarray(normFeat(XTrain, meanFt, varFt))
#print numFeatures, XTrain.shape
#TODO: SHUFFLE UP THE INPUT
clf = svm.SVC(kernel='rbf')
clf.fit(XTrain, YTrain)
XTest = controlTest + schTest
XTest = PCAObject.center(XTest)
if doPCA:
XTest = PCAObject.project(XTest)[:,0:numFeatures]
XTest = np.asarray(normFeat(XTest, meanFt, varFt))
YTest = [1]*len(controlTest) + [0]*len(schTest)
preds = clf.predict(XTest)
acc = getAccuracy(YTest, preds)
#print preds, acc
accuracySVM += [acc]
f1SVM += [f1_score(YTest, preds)]
#MLP
[train_fn, val_fn] = initNetwork(numFeatures)
for epoch in range(num_epochs):
#print np.asarray(XTrain).shape, np.asarray(YTrain).shape, numFeatures, epoch
train_err = train_fn(np.asarray(XTrain), np.asarray(YTrain))
#print train_err,
err, accN, pred = val_fn(np.asarray(XTrain), np.asarray(YTrain))
th = findCutoff(pred, YTrain, 0.01)
err, accN, pred = val_fn(np.asarray(XTest), np.asarray(YTest))
[accN,f1] = findAccuracyF1MLP(pred, YTest, th)
#print err,acc,pred, 'NN'
accuracyMLP += [accN]
f1MLP += [f1]
#adaboost
bdt = AdaBoostClassifier(DecisionTreeClassifier(max_depth=5), algorithm="SAMME", n_estimators=700)
#bdt = AdaBoostClassifier(DecisionTreeClassifier(max_depth=4), algorithm="SAMME", n_estimators=500)
bdt.fit(XTrain, YTrain)
predADA = bdt.predict(XTest)
accADA = getAccuracy(YTest, predADA)
accuracyADA += [accADA]
f1ADA += [f1_score(YTest, predADA)]
#explainer = lime.lime_tabular.LimeTabularExplainer(XTrain, feature_names=['f'+str(i) for i in range(numFeatures)], class_names=['Control', 'Sch'])
#exp = explainer.explain_instance(XTest[0], bdt.predict_proba, num_features=5, top_labels=1)
#exp.show_in_notebook(show_table=True, show_contributions=True, show_scaled=True, show_all=False)
#exp.as_pyplot_figure()
#exp.save_to_file('file' + str(foldid)+ '.html')
print foldid, 'SVM', acc, 'MLP', accN, 'ADA', accADA
print 'mean accuracy SVM', np.mean(accuracySVM)
print 'mean accuracy MLP', np.mean(accuracyMLP)
print 'mean accuracy ADA', np.mean(accuracyADA)
print 'mean F1 SVM', np.mean(f1SVM)
print 'mean F1 MLP', np.mean(f1MLP)
print 'mean F1 ADA', np.mean(f1ADA)
"""
0 SVM 0.633333333333 MLP 0.600000023842
1 SVM 0.642857142857 MLP 0.714285731316
2 SVM 0.785714285714 MLP 0.821428596973
3 SVM 0.642857142857 MLP 0.75
4 SVM 0.785714285714 MLP 0.678571403027
5 SVM 0.678571428571 MLP 0.75
6 SVM 0.714285714286 MLP 0.642857134342
7 SVM 0.807692307692 MLP 0.807692289352
8 SVM 0.807692307692 MLP 0.807692289352
9 SVM 0.75 MLP 0.666666686535
mean accuracy SVM 0.724871794872
mean accuracy MLP 0.723919
"""