deep-text-classifier-mtl/input_helpers.py at master · dhwajraj/deep-text-classifier-mtl · 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
138
139
140
141
142
143
144
145
146
import numpy as np
import re
import itertools
from collections import Counter
import numpy as np
import time
import data_helpers
import gc
from tensorflow.contrib import learn
from gensim.models.word2vec import Word2Vec
import gzip


class InputHelper(object):
    pre_emb = dict()

    def loadW2V(self,emb_path, type="textgz"):
        print("Loading W2V data...")
        num_keys = 0
        if type=="textgz":
            # this seems faster than gensim non-binary load
            for line in gzip.open(emb_path):
                l = line.strip().split()
                self.pre_emb[l[0]]=np.asarray(l[1:])
            num_keys=len(self.pre_emb)
        else:
            self.pre_emb = Word2Vec.load_word2vec_format(emb_path,binary=True)
            self.pre_emb.init_sims(replace=True)
            num_keys=len(self.pre_emb.vocab)
        print("loaded word2vec len ", num_keys)
        gc.collect()

    def deletePreEmb(self):
        self.pre_emb=dict()
        gc.collect()

    def getTsvData(self, filepath):
        print("Loading training data from "+filepath)
        x=[]
        y=[]
        for line in open(filepath):
            l=line.strip().split("\t")
            if len(l)<2:
                continue
            x.append(l[1])
            v=np.array([0,1])
            if l[0]=="-1" or l[0]=="0":
                v=np.array([1,0])
            y.append(v)
        return np.asarray(x),np.asarray(y)

    def getUnlabelData(self, filepath):
        print("Loading unlabelled data from "+filepath)
        x=[]
        for line in open(filepath):
            l=line.strip()
            if len(l)<1:
                continue
            x.append(l)
        return np.asarray(x)

    def dumpValidation(self,x_text,y,shuffled_index,dev_idx,i):
        print("dumping validation "+str(i))
        x_shuffled=x_text[shuffled_index]
        y_shuffled=y[shuffled_index]
        x_dev=x_shuffled[dev_idx:]
        y_dev=y_shuffled[dev_idx:]
        del x_shuffled
        del y_shuffled
        with open('validation.txt'+str(i),'w') as f:
            for text,label in zip(x_dev,y_dev):
                f.write(str(label)+"\t"+text+"\n")
            f.close()
        del x_dev
        del y_dev

    # Data Preparatopn
    # ==================================================


    def getDataSets(self, training_paths, max_document_length, filter_h_pad, percent_dev, batch_size):
        x_list=[]
        y_list=[]
        multi_train_size = len(training_paths)
        for i in xrange(multi_train_size):
            x_temp,y_temp = self.getTsvData(training_paths[i])
            x_list.append(x_temp)
            y_list.append(y_temp)
            del x_temp
            del y_temp
        gc.collect()
        # Build vocabulary
        print("Building vocabulary")
        vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length-filter_h_pad,min_frequency=1)
        vocab_processor.fit_transform(np.concatenate(x_list,axis=0))
        print("Length of loaded vocabulary ={}".format( len(vocab_processor.vocabulary_)))
        i1=0
        train_set=[]
        dev_set=[]
        sum_no_of_batches = 0
        for x_text,y in zip(x_list, y_list):
            x = np.asarray(list(vocab_processor.transform(x_text)))
            x = np.concatenate((np.zeros((len(x),filter_h_pad)),x),axis=1)
            # Randomly shuffle data
            np.random.seed(10)
            shuffle_indices = np.random.permutation(np.arange(len(y)))
            x_shuffled = x[shuffle_indices]
            y_shuffled = y[shuffle_indices]
            dev_idx = -1*len(y_shuffled)//percent_dev
            self.dumpValidation(x_text,y,shuffle_indices,dev_idx,i1)
            del x
            del x_text
            del y
            # Split train/test set
            # TODO: This is very crude, should use cross-validation
            x_train, x_dev = x_shuffled[:dev_idx], x_shuffled[dev_idx:]
            y_train, y_dev = y_shuffled[:dev_idx], y_shuffled[dev_idx:]
            print("Train/Dev split for {}: {:d}/{:d}".format(training_paths[i1], len(y_train), len(y_dev)))
            sum_no_of_batches = sum_no_of_batches+(len(y_train)//batch_size)
            train_set.append((x_train,y_train))
            dev_set.append((x_dev,y_dev))
            del x_shuffled
            del y_shuffled
            del x_train
            del x_dev
            i1=i1+1
        del x_list
        del y_list
        gc.collect()
        return train_set,dev_set,vocab_processor,sum_no_of_batches

    def getTestDataSet(self, data_path, vocab_path, max_document_length, filter_h_pad):
        x_temp,y = self.getTsvData(data_path)

        # Build vocabulary
        vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length-filter_h_pad,min_frequency=1)
        vocab_processor = vocab_processor.restore(vocab_path)
        print len(vocab_processor.vocabulary_)

        x = np.asarray(list(vocab_processor.transform(x_temp)))
        x = np.concatenate((np.zeros((len(x),filter_h_pad)),x),axis=1)
        # Randomly shuffle data
        del x_temp
        del vocab_processor
        gc.collect()
        return x, y