main.py
2.63 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
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
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':5,
'seed':345,
'nh':5, # dimension of hidden state
'nc':2 , # number of y classes
'ds':5} # 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 = "embeddings/embeddings_tmp.pkl", #sciezka do pliku z embeddingami - te podane tutaj sa tylko przeykladowym plikiem
max_phrase_length = 50 # przydaloby sie to uzaleznic od danych, ale nie jest to konieczne.
# Wazne, ze to jest wartosc nie mniejsza niz dlugosc najdluzszego zdania w danych
)
conll_format_data = 'data/slowa.conll'
data = load_conll_data(conll_format_data, rnn.words2ids)
# UWAGA: w pliku ekstowym zawierajacym frazy w conll na koncu musza byc dwie puste linie
train_data = data[0:int(0.9*len(data))]
test_data = data[int(0.9*len(data)):len(data)]
n_train_phrases = len(train_data)
n_test_phrases = len(test_data)
#to do: training with early stopping on validation set
s['clr'] = s['lr']
for e in xrange(s['nepochs']):
# shuffle
#shuffle([data], s['seed'])
#s['ce'] = e
tic = time.time()
for i in range(n_train_phrases):
#print(rnn.f(data[i][0],data[i][1], data[i][2]))
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()
# Test:
accuracy = 0.0
for i in range(n_test_phrases):
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/n_test_phrases)