def train_sentence_cbow(model, sentence, alpha, work=None, neu1=None):
"""
Update CBOW model by training on a single sentence.
The sentence is a list of string tokens, which are looked up in the model's
vocab dictionary. Called internally from `Word2Vec.train()`.
This is the non-optimized, Python version. If you have cython installed, gensim
will use the optimized version from word2vec_inner instead.
"""
word_vocabs = [
model.vocab[w] for w in sentence if w in model.vocab
and model.vocab[w].sample_int > model.random.rand() * 2**32
]
pos = 0
word = word_vocabs[0]
context = word_vocabs[1:]
reduced_window = model.random.randint(
model.window) # `b` in the original word2vec code
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(
word_vocabs[start:(pos + model.window + 1 - reduced_window)], start)
word2_indices = [
word2.index for pos2, word2 in enumerate(context)
if (word2 is not None and pos2 != pos)
]
l1 = word2vec.np_sum(model.syn0[word2_indices], axis=0) # 1 x vector_size
if word2_indices and model.cbow_mean:
l1 /= len(word2_indices)
word2vec.train_cbow_pair(model, word, word2_indices, l1, alpha)
return len(word_vocabs)
def train_batch_labeled_cbow(model,
sentences,
alpha,
work=None,
neu1=None):
result = 0
for sentence in sentences:
document, target = sentence
word_vocabs = [
model.wv.vocab[w] for w in document if w in model.wv.vocab
and model.wv.vocab[w].sample_int > model.random.rand() * 2**32
]
target_vocabs = [
model.lvocab[t] for t in target if t in model.lvocab
]
for target in target_vocabs:
word2_indices = [w.index for w in word_vocabs]
l1 = np_sum(model.wv.syn0[word2_indices],
axis=0) # 1 x vector_size
if word2_indices and model.cbow_mean:
l1 /= len(word2_indices)
if model.softmax:
train_cbow_pair_softmax(model, target, word2_indices, l1,
alpha)
else:
train_cbow_pair(model, target, word2_indices, l1, alpha)
result += len(word_vocabs)
return result
def train_batch_cbow(model, sentences, alpha, work=None, neu1=None):
result = 0
for sentence in sentences:
word_vocabs = [model.wv.vocab[w] for w in sentence if w in model.wv.vocab and
model.wv.vocab[w].sample_int > model.random.rand() * 2**32]
for pos, word in enumerate(word_vocabs):
reduced_window = model.random.randint(model.window)
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(word_vocabs[start:(pos + model.window + 1 - reduced_window)], start)
word2_indices = [word2.index for pos2, word2 in window_pos if (word2 is not None and pos2 != pos)]
word2_subwords = []
vocab_subwords_indices = []
ngrams_subwords_indices = []
for index in word2_indices:
vocab_subwords_indices += [index]
word2_subwords += model.wv.ngrams_word[model.wv.index2word[index]]
for subword in word2_subwords:
ngrams_subwords_indices.append(model.wv.ngrams[subword])
l1_vocab = np_sum(model.wv.syn0_vocab[vocab_subwords_indices], axis=0) # 1 x vector_size
l1_ngrams = np_sum(model.wv.syn0_ngrams[ngrams_subwords_indices], axis=0) # 1 x vector_size
l1 = np_sum([l1_vocab, l1_ngrams], axis=0)
subwords_indices = [vocab_subwords_indices] + [ngrams_subwords_indices]
if (subwords_indices[0] or subwords_indices[1]) and model.cbow_mean:
l1 /= (len(subwords_indices[0]) + len(subwords_indices[1]))
train_cbow_pair(model, word, subwords_indices, l1, alpha, is_ft=True) # train on the sliding window for target word
result += len(word_vocabs)
return result
def train_batch_cbow(model, sentences, alpha, work=None, neu1=None):
"""Update CBOW model by training on a sequence of sentences.
Called internally from :meth:`~gensim.models.fasttext.FastText.train`.
Notes
-----
This is the non-optimized, Python version. If you have cython installed, gensim will use the optimized version
from :mod:`gensim.models.fasttext_inner` instead.
Parameters
----------
model : :class:`~gensim.models.fasttext.FastText`
Model instance.
sentences : iterable of list of str
Iterable of the sentences.
alpha : float
Learning rate.
work : :class:`numpy.ndarray`, optional
UNUSED.
neu1 : :class:`numpy.ndarray`, optional
UNUSED.
Returns
-------
int
Effective number of words trained.
"""
result = 0
for sentence in sentences:
word_vocabs = [model.wv.vocab[w] for w in sentence if w in model.wv.vocab and
model.wv.vocab[w].sample_int > model.random.rand() * 2 ** 32]
for pos, word in enumerate(word_vocabs):
reduced_window = model.random.randint(model.window)
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(word_vocabs[start:(pos + model.window + 1 - reduced_window)], start)
word2_indices = [word2.index for pos2, word2 in window_pos if (word2 is not None and pos2 != pos)]
vocab_subwords_indices = []
ngrams_subwords_indices = []
for index in word2_indices:
vocab_subwords_indices += [index]
ngrams_subwords_indices.extend(model.wv.buckets_word[index])
l1_vocab = np_sum(model.wv.syn0_vocab[vocab_subwords_indices], axis=0) # 1 x vector_size
l1_ngrams = np_sum(model.wv.syn0_ngrams[ngrams_subwords_indices], axis=0) # 1 x vector_size
l1 = np_sum([l1_vocab, l1_ngrams], axis=0)
subwords_indices = [vocab_subwords_indices] + [ngrams_subwords_indices]
if (subwords_indices[0] or subwords_indices[1]) and model.cbow_mean:
l1 /= (len(subwords_indices[0]) + len(subwords_indices[1]))
# train on the sliding window for target word
train_cbow_pair(model, word, subwords_indices, l1, alpha, is_ft=True)
result += len(word_vocabs)
return result
def train_sentence_dm(model, sentence, lbls, alpha, work=None, neu1=None, train_words=True, train_lbls=True):
"""
Update distributed memory model by training on a single sentence.
The sentence is a list of Vocab objects (or None, where the corresponding
word is not in the vocabulary. Called internally from `Doc2Vec.train()`.
This is the non-optimized, Python version. If you have a C compiler, gensim
will use the optimized version from doc2vec_inner instead.
"""
lbl_indices = [lbl.index for lbl in lbls if lbl is not None]
lbl_sum = np_sum(model.syn0[lbl_indices], axis=0)
lbl_len = len(lbl_indices)
neg_labels = []
if model.negative:
# precompute negative labels
neg_labels = zeros(model.negative + 1)
neg_labels[0] = 1.
for pos, word in enumerate(sentence):
if word is None:
continue # OOV word in the input sentence => skip
reduced_window = random.randint(model.window) # `b` in the original doc2vec code
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(sentence[start : pos + model.window + 1 - reduced_window], start)
word2_indices = [word2.index for pos2, word2 in window_pos if (word2 is not None and pos2 != pos)]
l1 = np_sum(model.syn0[word2_indices], axis=0) + lbl_sum # 1 x layer1_size
if word2_indices and model.cbow_mean:
l1 /= (len(word2_indices) + lbl_len)
neu1e = train_cbow_pair(model, word, word2_indices, l1, alpha, neg_labels, train_words, train_words)
if train_lbls:
model.syn0[lbl_indices] += neu1e
return len([word for word in sentence if word is not None])
def train_batch_cbow(model, sentences, alpha, work=None, neu1=None):
result = 0
for sentence in sentences:
word_vocabs = [
model.wv.vocab[w] for w in sentence if w in model.wv.vocab
and model.wv.vocab[w].sample_int > model.random.rand() * 2**32
]
for pos, word in enumerate(word_vocabs):
reduced_window = model.random.randint(model.window)
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(
word_vocabs[start:(pos + model.window + 1 - reduced_window)],
start)
word2_indices = [
word2.index for pos2, word2 in window_pos
if (word2 is not None and pos2 != pos)
]
word2_subwords = []
vocab_subwords_indices = []
ngrams_subwords_indices = []
for index in word2_indices:
vocab_subwords_indices += [index]
word2_subwords += model.wv.ngrams_word[
model.wv.index2word[index]]
for subword in word2_subwords:
ngrams_subwords_indices.append(model.wv.ngrams[subword])
l1_vocab = np_sum(model.wv.syn0_vocab[vocab_subwords_indices],
axis=0) # 1 x vector_size
l1_ngrams = np_sum(model.wv.syn0_ngrams[ngrams_subwords_indices],
axis=0) # 1 x vector_size
l1 = np_sum([l1_vocab, l1_ngrams], axis=0)
subwords_indices = [vocab_subwords_indices
] + [ngrams_subwords_indices]
if (subwords_indices[0]
or subwords_indices[1]) and model.cbow_mean:
l1 /= (len(subwords_indices[0]) + len(subwords_indices[1]))
train_cbow_pair(
model, word, subwords_indices, l1, alpha,
is_ft=True) # train on the sliding window for target word
result += len(word_vocabs)
return result
def train_document_dm(model, doc_words, doctag_indexes, alpha, work=None, neu1=None,
learn_doctags=True, learn_words=True, learn_hidden=True,
word_vectors=None, word_locks=None, doctag_vectors=None, doctag_locks=None):
"""
Update distributed memory model ("PV-DM") by training on a single document.
Called internally from `Doc2Vec.train()` and `Doc2Vec.infer_vector()`. This
method implements the DM model with a projection (input) layer that is
either the sum or mean of the context vectors, depending on the model's
`dm_mean` configuration field. See `train_dm_concat()` for the DM model
with a concatenated input layer.
The document is provided as `doc_words`, a list of word tokens which are looked up
in the model's vocab dictionary, and `doctag_indexes`, which provide indexes
into the doctag_vectors array.
Any of `learn_doctags', `learn_words`, and `learn_hidden` may be set False to
prevent learning-updates to those respective model weights, as if using the
(partially-)frozen model to infer other compatible vectors.
This is the non-optimized, Python version. If you have a C compiler, gensim
will use the optimized version from doc2vec_inner instead.
"""
if word_vectors is None:
word_vectors = model.syn0
if word_locks is None:
word_locks = model.syn0_lockf
if doctag_vectors is None:
doctag_vectors = model.docvecs.doctag_syn0
if doctag_locks is None:
doctag_locks = model.docvecs.doctag_syn0_lockf
word_vocabs = [model.vocab[w] for w in doc_words if w in model.vocab and
model.vocab[w].sample_int > model.random.randint(2**32)]
doctag_sum = np_sum(doctag_vectors[doctag_indexes], axis=0)
doctag_len = len(doctag_indexes)
for pos, word in enumerate(word_vocabs):
reduced_window = model.random.randint(model.window) # `b` in the original doc2vec code
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(word_vocabs[start:(pos + model.window + 1 - reduced_window)], start)
word2_indexes = [word2.index for pos2, word2 in window_pos if (word2 is not None and pos2 != pos)]
l1 = np_sum(word_vectors[word2_indexes], axis=0) + doctag_sum # 1 x layer1_size
if word2_indexes and model.cbow_mean:
l1 /= (len(word2_indexes) + doctag_len)
neu1e = train_cbow_pair(model, word, word2_indexes, l1, alpha,
learn_vectors=False, learn_hidden=learn_hidden)
if word2_indexes and not model.cbow_mean:
neu1e /= (len(word2_indexes) + doctag_len)
if learn_doctags:
doctag_vectors[doctag_indexes] += neu1e * \
np_repeat(doctag_locks[doctag_indexes], model.vector_size).reshape(-1, model.vector_size)
if learn_words:
word_vectors[word2_indexes] += neu1e * \
np_repeat(word_locks[word2_indexes], model.vector_size).reshape(-1, model.vector_size)
return len(word_vocabs)
def train_document_dm(model, doc_words, doctag_indexes, alpha, work=None, neu1=None,
learn_doctags=True, learn_words=True, learn_hidden=True,
word_vectors=None, word_locks=None, doctag_vectors=None, doctag_locks=None):
"""
Update distributed memory model ("PV-DM") by training on a single document.
Called internally from `Doc2Vec.train()` and `Doc2Vec.infer_vector()`. This
method implements the DM model with a projection (input) layer that is
either the sum or mean of the context vectors, depending on the model's
`dm_mean` configuration field. See `train_document_dm_concat()` for the DM
model with a concatenated input layer.
The document is provided as `doc_words`, a list of word tokens which are looked up
in the model's vocab dictionary, and `doctag_indexes`, which provide indexes
into the doctag_vectors array.
Any of `learn_doctags', `learn_words`, and `learn_hidden` may be set False to
prevent learning-updates to those respective model weights, as if using the
(partially-)frozen model to infer other compatible vectors.
This is the non-optimized, Python version. If you have a C compiler, gensim
will use the optimized version from doc2vec_inner instead.
"""
if word_vectors is None:
word_vectors = model.syn0
if word_locks is None:
word_locks = model.syn0_lockf
if doctag_vectors is None:
doctag_vectors = model.docvecs.doctag_syn0
if doctag_locks is None:
doctag_locks = model.docvecs.doctag_syn0_lockf
word_vocabs = [model.vocab[w] for w in doc_words if w in model.vocab and
model.vocab[w].sample_int > model.random.rand() * 2**32]
for pos, word in enumerate(word_vocabs):
reduced_window = model.random.randint(model.window) # `b` in the original doc2vec code
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(word_vocabs[start:(pos + model.window + 1 - reduced_window)], start)
word2_indexes = [word2.index for pos2, word2 in window_pos if pos2 != pos]
l1 = np_sum(word_vectors[word2_indexes], axis=0) + np_sum(doctag_vectors[doctag_indexes], axis=0)
count = len(word2_indexes) + len(doctag_indexes)
if model.cbow_mean and count > 1 :
l1 /= count
neu1e = train_cbow_pair(model, word, word2_indexes, l1, alpha,
learn_vectors=False, learn_hidden=learn_hidden)
if not model.cbow_mean and count > 1:
neu1e /= count
if learn_doctags:
for i in doctag_indexes:
doctag_vectors[i] += neu1e * doctag_locks[i]
if learn_words:
for i in word2_indexes:
word_vectors[i] += neu1e * word_locks[i]
return len(word_vocabs)
def train_batch_labeled_cbow(model, tagged_docs, alpha, work=None, neu1=None):
result = 0
for tagged_doc in tagged_docs:
document, target = tagged_doc
word_vocabs = [model.wv.vocab[w] for w in document if w in model.wv.vocab and
model.wv.vocab[w].sample_int > model.random.rand() * 2 ** 32]
target_vocabs = [model.lvocab[t] for t in target if t in model.lvocab]
for target in target_vocabs:
word2_indices = [w.index for w in word_vocabs]
l1 = np_sum(model.wv.syn0[word2_indices], axis=0) # 1 x vector_size
if word2_indices and model.cbow_mean:
l1 /= len(word2_indices)
if model.softmax:
train_cbow_pair_softmax(model, target, word2_indices, l1, alpha)
else:
train_cbow_pair(model, target, word2_indices, l1, alpha)
result += len(word_vocabs)
return result
def train_sentence_dm(model,
sentence,
lbls,
alpha,
work=None,
neu1=None,
train_words=True,
train_lbls=True):
"""
Update distributed memory model by training on a single sentence.
The sentence is a list of Vocab objects (or None, where the corresponding
word is not in the vocabulary. Called internally from `Doc2Vec.train()`.
This is the non-optimized, Python version. If you have a C compiler, gensim
will use the optimized version from doc2vec_inner instead.
"""
lbl_indices = [lbl.index for lbl in lbls if lbl is not None]
lbl_sum = np_sum(model.syn0[lbl_indices], axis=0)
lbl_len = len(lbl_indices)
neg_labels = []
if model.negative:
# precompute negative labels
neg_labels = zeros(model.negative + 1)
neg_labels[0] = 1.
for pos, word in enumerate(sentence):
if word is None:
continue # OOV word in the input sentence => skip
reduced_window = random.randint(
model.window) # `b` in the original doc2vec code
start = max(0, pos - model.window + reduced_window)
window_pos = enumerate(
sentence[start:pos + model.window + 1 - reduced_window], start)
word2_indices = [
word2.index for pos2, word2 in window_pos
if (word2 is not None and pos2 != pos)
]
l1 = np_sum(model.syn0[word2_indices],
axis=0) + lbl_sum # 1 x layer1_size
if word2_indices and model.cbow_mean:
l1 /= (len(word2_indices) + lbl_len)
neu1e = train_cbow_pair(model, word, word2_indices, l1, alpha,
neg_labels, train_words, train_words)
if train_lbls:
model.syn0[lbl_indices] += neu1e
return len([word for word in sentence if word is not None])
def train_document_dm_concat(model, doc_words, doctag_indexes, alpha, work=None, neu1=None,
learn_doctags=True, learn_words=True, learn_hidden=True,
word_vectors=None, word_locks=None, doctag_vectors=None, doctag_locks=None):
"""
Update distributed memory model ("PV-DM") by training on a single document, using a
concatenation of the context window word vectors (rather than a sum or average).
Called internally from `Doc2Vec.train()` and `Doc2Vec.infer_vector()`.
The document is provided as `doc_words`, a list of word tokens which are looked up
in the model's vocab dictionary, and `doctag_indexes`, which provide indexes
into the doctag_vectors array.
Any of `learn_doctags', `learn_words`, and `learn_hidden` may be set False to
prevent learning-updates to those respective model weights, as if using the
(partially-)frozen model to infer other compatible vectors.
This is the non-optimized, Python version. If you have a C compiler, gensim
will use the optimized version from doc2vec_inner instead.
"""
if word_vectors is None:
word_vectors = model.syn0
if word_locks is None:
word_locks = model.syn0_lockf
if doctag_vectors is None:
doctag_vectors = model.docvecs.doctag_syn0
if doctag_locks is None:
doctag_locks = model.docvecs.doctag_syn0_lockf
word_vocabs = [model.vocab[w] for w in doc_words if w in model.vocab and
model.vocab[w].sample_int > model.random.rand() * 2**32]
doctag_len = len(doctag_indexes)
if doctag_len != model.dm_tag_count:
return 0 # skip doc without expected number of doctag(s) (TODO: warn/pad?)
null_word = model.vocab['\0']
pre_pad_count = model.window
post_pad_count = model.window
padded_document_indexes = (
(pre_pad_count * [null_word.index]) # pre-padding
+ [word.index for word in word_vocabs if word is not None] # elide out-of-Vocabulary words
+ (post_pad_count * [null_word.index]) # post-padding
)
for pos in range(pre_pad_count, len(padded_document_indexes) - post_pad_count):
word_context_indexes = (
padded_document_indexes[(pos - pre_pad_count): pos] # preceding words
+ padded_document_indexes[(pos + 1):(pos + 1 + post_pad_count)] # following words
)
word_context_len = len(word_context_indexes)
predict_word = model.vocab[model.index2word[padded_document_indexes[pos]]]
# numpy advanced-indexing copies; concatenate, flatten to 1d
l1 = concatenate((doctag_vectors[doctag_indexes], word_vectors[word_context_indexes])).ravel()
neu1e = train_cbow_pair(model, predict_word, None, l1, alpha,
learn_hidden=learn_hidden, learn_vectors=False)
# filter by locks and shape for addition to source vectors
e_locks = concatenate((doctag_locks[doctag_indexes], word_locks[word_context_indexes]))
neu1e_r = (neu1e.reshape(-1, model.vector_size)
* np_repeat(e_locks, model.vector_size).reshape(-1, model.vector_size))
if learn_doctags:
np_add.at(doctag_vectors, doctag_indexes, neu1e_r[:doctag_len])
if learn_words:
np_add.at(word_vectors, word_context_indexes, neu1e_r[doctag_len:])
return len(padded_document_indexes) - pre_pad_count - post_pad_count
def train_batch(model,
sentences,
alpha,
work=None,
neu1=None,
compute_loss=False):
"""Update CBOW model by training on a sequence of sentences.
Called internally from :meth:`~gensim.models.word2vec.Word2Vec.train`.
Warnings
--------
This is the non-optimized, pure Python version. If you have a C compiler, Gensim
will use an optimized code path from :mod:`gensim.models.word2vec_inner` instead.
Parameters
----------
model : :class:`~gensim.models.word2vec.Word2Vec`
The Word2Vec model instance to train.
sentences : iterable of list of str
The corpus used to train the model.
alpha : float
The learning rate
work : object, optional
Unused.
neu1 : object, optional
Unused.
compute_loss : bool, optional
Whether or not the training loss should be computed in this batch.
Returns
-------
int
Number of words in the vocabulary actually used for training (that already existed in the vocabulary
and were not discarded by negative sampling).
"""
result = 0
for sentence in sentences:
word_vocabs = [
model.wv.vocab[w] for w in sentence if w in model.wv.vocab
and model.wv.vocab[w].sample_int > model.random.rand() * 2**32
]
word = word_vocabs[0]
start = 1
window_pos = enumerate(word_vocabs[start:], start)
word2_indices = [
word2.index for pos2, word2 in window_pos if (word2 is not None)
]
l1 = np_sum(model.wv.syn0[word2_indices], axis=0) # 1 x vector_size
if word2_indices and model.cbow_mean:
l1 /= len(word2_indices)
train_cbow_pair(model,
word,
word2_indices,
l1,
alpha,
compute_loss=compute_loss)
for word2idx in word2_indices:
train_sg_pair(model,
model.wv.index2word[word.index],
word2idx,
alpha,
compute_loss=compute_loss)
train_sg_pair(model,
model.wv.index2word[word2idx],
word.index,
alpha,
compute_loss=compute_loss)
result += len(word_vocabs)
return result
