explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-07-21 21:49:49 +03:00
Code Issues Packages Projects Releases Wiki Activity

Fix GPU usage in chainer example

Browse Source
This commit is contained in:
Matthew Honnibal 2016-11-19 10:58:00 -06:00
parent 4c84aae571
commit 8a2de46fcb
1 changed files with 30 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
examples/chainer_sentiment.py
View File

@ -1,6 +1,7 @@
'''WIP --- Doesn't work well yet''' '''WIP --- Doesn't work well yet'''
import plac import plac
import random import random
import six
import pathlib import pathlib
import cPickle as pickle import cPickle as pickle
@ -9,7 +10,10 @@ from itertools import izip
import spacy import spacy
import cytoolz import cytoolz
import numpy as np import cupy as xp
import cupy.cuda
import chainer.cuda
import chainer.links as L import chainer.links as L
import chainer.functions as F import chainer.functions as F
from chainer import Chain, Variable, report from chainer import Chain, Variable, report
@ -17,7 +21,7 @@ import chainer.training
import chainer.optimizers import chainer.optimizers
from chainer.training import extensions from chainer.training import extensions
from chainer.iterators import SerialIterator from chainer.iterators import SerialIterator
from chainer.datasets.tuple_dataset import TupleDataset from chainer.datasets import TupleDataset
class SentimentAnalyser(object): class SentimentAnalyser(object):
@ -79,6 +83,7 @@ class SentimentModel(Chain):
encode=_Encode(shape['nr_hidden'], shape['nr_hidden']), encode=_Encode(shape['nr_hidden'], shape['nr_hidden']),
attend=_Attend(shape['nr_hidden'], shape['nr_hidden']), attend=_Attend(shape['nr_hidden'], shape['nr_hidden']),
predict=_Predict(shape['nr_hidden'], shape['nr_class'])) predict=_Predict(shape['nr_hidden'], shape['nr_class']))
self.to_gpu(0)
def __call__(self, sentence): def __call__(self, sentence):
return self.predict( return self.predict(
@ -145,6 +150,16 @@ class SentenceDataset(TupleDataset):
get_features(sents, max_length), get_features(sents, max_length),
labels) labels)
def __getitem__(self, index):
batches = [dataset[index] for dataset in self._datasets]
if isinstance(index, slice):
length = len(batches[0])
returns = [tuple([batch[i] for batch in batches])
for i in six.moves.range(length)]
return returns
else:
return tuple(batches)
def _get_labelled_sentences(self, docs, doc_labels): def _get_labelled_sentences(self, docs, doc_labels):
labels = [] labels = []
sentences = [] sentences = []
@ -152,19 +167,17 @@ class SentenceDataset(TupleDataset):
for sent in doc.sents: for sent in doc.sents:
sentences.append(sent) sentences.append(sent)
labels.append(y) labels.append(y)
return sentences, labels return sentences, xp.asarray(labels, dtype='i')
class DocDataset(TupleDataset): class DocDataset(TupleDataset):
def __init__(self, nlp, texts, labels): def __init__(self, nlp, texts, labels):
self.max_length = max_length self.max_length = max_length
TupleDataset.__init__(self, DatasetMixin.__init__(self,
get_features( get_features(
nlp.pipe(texts, batch_size=5000, n_threads=3), self.max_length), nlp.pipe(texts, batch_size=5000, n_threads=3), self.max_length),
labels) labels)
def read_data(data_dir, limit=0): def read_data(data_dir, limit=0):
examples = [] examples = []
for subdir, label in (('pos', 1), ('neg', 0)): for subdir, label in (('pos', 1), ('neg', 0)):
@ -180,7 +193,7 @@ def read_data(data_dir, limit=0):
def get_features(docs, max_length): def get_features(docs, max_length):
docs = list(docs) docs = list(docs)
Xs = np.zeros((len(docs), max_length), dtype='int32') Xs = xp.zeros((len(docs), max_length), dtype='i')
for i, doc in enumerate(docs): for i, doc in enumerate(docs):
j = 0 j = 0
for token in doc: for token in doc:
@ -195,7 +208,7 @@ def get_features(docs, max_length):
def get_embeddings(vocab, max_rank=1000): def get_embeddings(vocab, max_rank=1000):
if max_rank is None: if max_rank is None:
max_rank = max(lex.rank+1 for lex in vocab if lex.has_vector) max_rank = max(lex.rank+1 for lex in vocab if lex.has_vector)
vectors = np.ndarray((max_rank+1, vocab.vectors_length), dtype='float32') vectors = xp.ndarray((max_rank+1, vocab.vectors_length), dtype='f')
for lex in vocab: for lex in vocab:
if lex.has_vector and lex.rank < max_rank: if lex.has_vector and lex.rank < max_rank:
lex.norm = lex.rank+1 lex.norm = lex.rank+1
@ -208,15 +221,12 @@ def get_embeddings(vocab, max_rank=1000):
def train(train_texts, train_labels, dev_texts, dev_labels, def train(train_texts, train_labels, dev_texts, dev_labels,
lstm_shape, lstm_settings, lstm_optimizer, batch_size=100, nb_epoch=5, lstm_shape, lstm_settings, lstm_optimizer, batch_size=100, nb_epoch=5,
by_sentence=True): by_sentence=True):
print("Loading spaCy")
nlp = spacy.load('en', entity=False) nlp = spacy.load('en', entity=False)
for lex in nlp.vocab: for lex in nlp.vocab:
if lex.rank >= (lstm_shape['nr_vector'] - 1): if lex.rank >= (lstm_shape['nr_vector'] - 1):
lex.norm = 0 lex.norm = 0
else: else:
lex.norm = lex.rank+1 lex.norm = lex.rank+1
#print("Get embeddings")
#embeddings = get_embeddings(nlp.vocab)
print("Make model") print("Make model")
model = Classifier(SentimentModel(lstm_shape, **lstm_settings)) model = Classifier(SentimentModel(lstm_shape, **lstm_settings))
print("Parsing texts...") print("Parsing texts...")
@ -230,13 +240,12 @@ def train(train_texts, train_labels, dev_texts, dev_labels,
shuffle=True, repeat=True) shuffle=True, repeat=True)
dev_iter = SerialIterator(dev_data, batch_size=batch_size, dev_iter = SerialIterator(dev_data, batch_size=batch_size,
shuffle=False, repeat=False) shuffle=False, repeat=False)
optimizer = chainer.optimizers.Adam() optimizer = chainer.optimizers.Adam()
optimizer.setup(model) optimizer.setup(model)
updater = chainer.training.StandardUpdater(train_iter, optimizer) updater = chainer.training.StandardUpdater(train_iter, optimizer, device=0)
trainer = chainer.training.Trainer(updater, (20, 'epoch'), out='result') trainer = chainer.training.Trainer(updater, (20, 'epoch'), out='result')
trainer.extend(extensions.Evaluator(dev_iter, model)) trainer.extend(extensions.Evaluator(dev_iter, model, device=0))
trainer.extend(extensions.LogReport()) trainer.extend(extensions.LogReport())
trainer.extend(extensions.PrintReport([ trainer.extend(extensions.PrintReport([
'epoch', 'main/accuracy', 'validation/main/accuracy'])) 'epoch', 'main/accuracy', 'validation/main/accuracy']))
@ -293,14 +302,16 @@ def main(model_dir, train_dir, dev_dir,
print("Read data") print("Read data")
train_texts, train_labels = read_data(train_dir, limit=nr_examples) train_texts, train_labels = read_data(train_dir, limit=nr_examples)
dev_texts, dev_labels = read_data(dev_dir, limit=nr_examples) dev_texts, dev_labels = read_data(dev_dir, limit=nr_examples)
train_labels = np.asarray(train_labels, dtype='int32') print("Using GPU 0")
dev_labels = np.asarray(dev_labels, dtype='int32') #chainer.cuda.get_device(0).use()
train_labels = xp.asarray(train_labels, dtype='i')
dev_labels = xp.asarray(dev_labels, dtype='i')
lstm = train(train_texts, train_labels, dev_texts, dev_labels, lstm = train(train_texts, train_labels, dev_texts, dev_labels,
{'nr_hidden': nr_hidden, 'max_length': max_length, 'nr_class': 2, {'nr_hidden': nr_hidden, 'max_length': max_length, 'nr_class': 2,
'nr_vector': 2000, 'nr_dim': 32}, 'nr_vector': 2000, 'nr_dim': 32},
{'dropout': 0.5, 'lr': learn_rate}, {'dropout': 0.5, 'lr': learn_rate},
{}, {},
nb_epoch=nb_epoch, batch_size=batch_size) nb_epoch=nb_epoch, batch_size=batch_size)
if __name__ == '__main__': if __name__ == '__main__':