teReX/SSL.py at master · Tilana/teReX · GitHub

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
import numpy as np
import pandas as pd
from sklearn.semi_supervised import LabelPropagation
from sklearn.semi_supervised import LabelSpreading
from sklearn.metrics import accuracy_score
from sklearn.metrics.pairwise import cosine_similarity
from scipy import sparse
from plotFunctions import surface
from helper import createConfigName, generateVocabulary
from itertools import product

def SSL():
	# PARAMETERS
	RBF = 1
	gammaArray = [0.5, 1, 5, 10, 20, 50, 100]
	conversion = 'tfidf'	# one of None, 'tfidf', 'MM', 'raw_tfidf'
	nrLabeledData = 250

	# Load Data
	name = 'NG_guns_motorcycles'
	filename = 'processedDocuments/' + name + '.pkl'
	resultFilename = 'results/' + name + '_' + str(conversion) + '_'+ str(nrLabeledData) + '.txt'

	results = pd.DataFrame(data = {'gamma': gammaArray})

	for params in product((0,1), repeat=3):
		cosSim = params[0]
		avgFF = params[1]
		useContextSim = params[2]

		configName = createConfigName(cosSim, avgFF, useContextSim)
		print configName

		data = pd.read_pickle(filename)
		vocabulary = generateVocabulary(data.sentences.tolist())
		X = np.load('matrices/' + name + '.npy')
		X = X[:-2,:-2]
		contextSim = np.load('matrices/' + name + '_contextSim.npy')
		nrDocs = len(data)

		labels = np.ones([X.shape[0]])*-1
		trueLabelIndex = range(0,nrLabeledData)
		labels[trueLabelIndex] = data.loc[trueLabelIndex, 'category'].tolist()

		# remove DD and FD
		#X = X[:,nrDocs:]
		# remove $Start$ and $End$

		if useContextSim:
			X[nrDocs:, nrDocs:] = contextSim

		if avgFF:
			FF = np.add(X[nrDocs:, nrDocs:], contextSim)/2
			X[nrDocs:, nrDocs:] = FF

		# Renormalize
		#if renormalize:
		#	DF = X[nrDocs:, :nrDocs]
		#	rowsums = DF.sum(axis=1)
		#	for i in range(len(rowsums)):
		#		DF[i] = DF[i]/rowsums[i]
		#	X[nrDocs:,:nrDocs] = DF
		#	X[:nrDocs,nrDocs:] = np.transpose(DF)
		#FF = X[nrDocs:,:]
		#X[nrDocs:,:] = np.transpose(FF)

		# Remove posts with no features
		#DF = X[:nrDocs,:]
		#indZeroFeatures = np.where(DF.sum(axis=1)==0)[0]
		#for ind in indZeroFeatures:
		#	X = np.delete(X,ind,0)
		#data.drop(data.index[indZeroFeatures], inplace=True)
		#data.index = range(len(data))
		#nrDocs = len(data)

		# Normalize
		#DF = X[:nrDocs,:]
		#FF = X[nrDocs:,:]
		#rowsum = DF.sum(axis=1)
		#X[nrDocs:, nrDocs:] = np.transpose(X[nrDocs:, nrDocs:])
		#X[-1,-1] = 1
		#FF = X[nrDocs:, nrDocs:]
		#FF_rowsum = FF.sum(axis=1)

		if conversion=='tfidf':
			DF = np.array(data.tfIdf.tolist())
			X[:nrDocs, nrDocs:] = DF
			#X[nrDocs:, :nrDocs] = np.transpose(DF)
			#X = X[:,:]

		if conversion=='raw_tfidf':
			DF = np.array(data.tfIdf.tolist())
			X = DF

		if conversion=='MM':
			DF = X[:nrDocs, nrDocs:]
			FF = X[nrDocs:, nrDocs:]
			X = np.dot(DF,FF)

		if cosSim:
			X = cosine_similarity(X,X)

		print 'Label Propagation'
		if RBF:
			labelProp_accuracy = []
			labelSpread_accuracy = []
			for gamma in gammaArray:
				print 'Gamma: %f' % gamma
				labelPropagation = LabelPropagation('rbf', gamma=gamma, alpha=1, useInputMatrix=0, max_iter=100)
				labelPropagation.fit(X, labels)
				predictLabels = labelPropagation.transduction_
				curr_acc = accuracy_score(data.category.tolist()[nrLabeledData:], predictLabels.tolist()[nrLabeledData:len(data)])
				labelProp_accuracy.append(curr_acc)
				print 'Label Prop. Test Accuracy: %f' % curr_acc
				labelSpread = LabelSpreading('rbf', gamma=gamma, alpha=1)
				#Test
				labelSpread.fit(X,labels)
				predictLabels = labelSpread.transduction_
				curr_acc = accuracy_score(data.category.tolist()[nrLabeledData:], predictLabels.tolist()[nrLabeledData:len(data)])
				labelSpread_accuracy.append(curr_acc)
				print 'Label Spread. Test Accuracy: %f' % curr_acc

			results[configName+'_LP'] = labelProp_accuracy
			results[configName+'_LS'] = labelSpread_accuracy

		else:
			labelPropagation = LabelPropagation(alpha=1, useInputMatrix=1, max_iter=200)
			print labelPropagation
			labelPropagation.fit(X, labels)
			predictLabels = labelPropagation.transduction_
			print 'Test Accuracy: %f' % accuracy_score(data.category.tolist()[nrLabeledData+1:], predictLabels.tolist()[nrLabeledData+1:len(data)])

	results.to_csv(resultFilename, sep='\t', encoding='utf-8')


if __name__ =='__main__':
	SSL()