main_stanford.py
3.18 KB
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
import numpy
import time
import sys
import subprocess
import os
import random
from modules.data import load
from modules.rnn.nnet_for_dependency_trees import model
#from modules.metrics.accuracy import conlleval
from modules.utils.tools import shuffle, words_in_from_down_to_top_order, load_conll_data, load_stanford_data
import theano.tensor as T
import theano
import itertools
if __name__ == '__main__':
s = {'lr':0.0627142536696559,
'verbose':1,
'decay':False, # decay on the learning rate if improvement stops
'nepochs':1,
'seed':345,
'nh':300, # dimension of hidden state
'nc':5 , # number of y classes
'ds':30} # dimension of sentiment state
# instanciate the model
numpy.random.seed(s['seed'])
random.seed(s['seed'])
rnn = model( nh = s['nh'],
nc = s['nc'],
ds = s['ds'],
w2v_model_path = "/home/norbert/Doktorat/clarni2sent/treelstm/data/glove/glove.840B.300d.txt", #sciezka do pliku z embeddingami
max_phrase_length = 60 # przydaloby sie to uzaleznic od danych, ale nie jest to konieczne.
# Wazne, ze to jest wartosc nie mniejsza niz dlugosc najdluzszego zdania w danych
)
train_data = load_stanford_data("/home/norbert/Doktorat/clarni2sent/treelstm/data/sst/train/dlabels.txt", "/home/norbert/Doktorat/clarni2sent/treelstm/data/sst/train/dparents.txt","/home/norbert/Doktorat/clarni2sent/treelstm/data/sst/train/sents.toks",rnn.words2ids)[0:100]
test_data = load_stanford_data("/home/norbert/Doktorat/clarni2sent/treelstm/data/sst/test/dlabels.txt", "/home/norbert/Doktorat/clarni2sent/treelstm/data/sst/test/dparents.txt","/home/norbert/Doktorat/clarni2sent/treelstm/data/sst/test/sents.toks",rnn.words2ids)
n_train = len(train_data)
n_test = len(test_data)
#to do: training with early stopping on validation set
print("Number of training phrases: ", n_train)
s['clr'] = s['lr']
for e in xrange(s['nepochs']):
print("epoch: ", e)
# shuffle
#shuffle([data], s['seed'])
#s['ce'] = e
tic = time.time()
for i in range(n_train):
#print(data[i][0],data[i][1], data[i][2], train_data[i][3])
print(i)
#print(train_data[i][0][range(k)],train_data[i][1][range(k)], train_data[i][2][range(k)], train_data[i][3][k-1])
rnn.train(train_data[i][0],train_data[i][1], train_data[i][2], train_data[i][3], s['clr'])
#rnn.normalize()
#if s['verbose']:
# print ('[learning] epoch %i >> %2.2f%%'%(e,(i+1)*100./n_train_phrases),'completed in %.2f (sec) <<\r'%(time.time()-tic))
# sys.stdout.flush()
print(time.time()-tic)
# Test:
accuracy = 0.0
for i in range(n_test):
accuracy += int(test_data[i][3] == rnn.classify(test_data[i][0],test_data[i][1], test_data[i][2]))
#print(test_data[i][3],rnn.classify(test_data[i][0],test_data[i][1], test_data[i][2]),accuracy)
print("accuracy on test data:")
print(accuracy/n_test)