blob: 28abc950880dd40c2de5fd738f85cf87dc668f67 (
plain)
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
|
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Author: Rico Sennrich
"""Average embeddings of special null words for RDLM.
Usage:
average_null_embedding.py NPLM_PATH INPUT_MODEL TRAINING_FILE OUTPUT_MODEL
"""
import sys
import os
import numpy
def load_model(model_file):
return nplm.NeuralLM.from_file(model_file)
def get_weights(path, vocab, len_context):
d = [[0] * vocab for i in range(len_context)]
for line in open(path):
for i, word in enumerate(line.split()[:-1]):
d[i][int(word)] += 1
return d
if __name__ == "__main__":
nplm_path = sys.argv[1]
model_input = sys.argv[2]
training_instances = sys.argv[3]
model_output = sys.argv[4]
sys.path.append(os.path.join(nplm_path, 'python'))
import nplm
model = load_model(model_input)
len_context = len(open(training_instances).readline().split()) - 1
sys.stderr.write('reading ngrams...')
weights = numpy.array(
get_weights(
training_instances, len(model.input_embeddings), len_context))
sys.stderr.write('done\n')
for i in range(len_context):
index = model.word_to_index_input['<null_{0}>'.format(i)]
model.input_embeddings[index] = numpy.average(
numpy.array(model.input_embeddings), weights=weights[i], axis=0)
sys.stderr.write('writing model...')
model.to_file(open(model_output, 'w'))
sys.stderr.write('done\n')
|