Builds the activations of hidden layer given input words embeddings `syn0` and input word indices. Args: syn0: float tensor of shape [

Builds the activations of hidden layer given input words embeddings `syn0` and input word indices. Args: inputs: int tensor of shape [

Builds a Huffman tree for hierarchical softmax. Has the side effect of setting `max_depth`. Args: unigram_counts: list of int, holding

Builds a Huffman tree for hierarchical softmax. Has the side effect of setting `max_depth`. Args: unigram_counts: list of int, holding

Builds the graph that leads from data tensors (`inputs`, `labels`) to loss. Has the side effect of setting attribute `syn0`. Args: inpu

Builds raw vocabulary. Args: filenames: list of strings, holding names of text files. Returns: raw_vocab: a list of 2-tuples ho

Builds raw vocabulary by iterate through the corpus once and count the unique words. Args: filenames: list of strings, holding names o

Creates initial word embedding variables. Args: vocab_size: int scalar, num of words in vocabulary. scope: string scalar, scope name.

Builds the loss for hierarchical softmax. Args: inputs: int tensor of shape [batch_size] (skip_gram) or [batch_size, 2*window_size

Builds the loss for hierarchical softmax. Args: inputs: int tensor of shape [batch_size] (skip_gram) or [batch_size, 2*window_size

Builds the loss for negative sampling. Args: unigram_counts: list of int, holding word counts. Index of each entry is the same as t

Builds the loss for negative sampling. Args: inputs: int tensor of shape [batch_size] (skip_gram) or [batch_size, 2*window_size+1]

Set shape of `tensor` according to architecture and training algorithm, and split `tensor` into `inputs` and `labels`. Args: tensor: ra

Generates tensor dict mapping from tensor names to tensors. Args: filenames: list of strings, holding names of text files. Retur

Builds vocabulary. Has the side effect of setting the following attributes: - table_words: list of string, holding the list of vocabulary

Builds the vocabulary. Has the side effect of setting the following attributes: for each word `word` we have vocab[word] = index ta

Split raw text string into tokens (space-separated) and tranlate to token ids. Args: string: string scalar, the raw text string to be

Generates matrices holding word indices to be passed to Word2Vec models for each sentence. The shape and contents of output matrices depends on the

Generates matrices holding word indices to be passed to Word2Vec models for each sentence. The shape and contents of output matrices depends on the

Generates tensor dict mapping from tensor names to tensors. Args: filenames: list of strings, holding names of text files. Retur

Converts a sentence into a list of word indices. Args: sent: a scalar string tensor, a sentence where words are space-delimited. table_word

Finds the top-k words with smallest cosine distances w.r.t `word`. Args: word: string scalar, the query word. k: int scalar, num of w

Filters out-of-vocabulary words and then applies subsampling on words in a sentence. Words with high frequencies have lower keep probs. Args:

Filters out-of-vocabulary words and then applies subsampling on words in a sentence. Words with high frequencies have lower keep probs. Args:

Adds training related ops to the graph. Args: dataset: a `Word2VecDataset` instance. filenames: a list of strings, holding names of t

Constructor. Args: arch: string scalar, architecture ('skip_gram' or 'cbow'). algm: string scalar, training algorithm ('negative_samp

Constructor. Args: syn0_final: numpy array of shape [vocab_size, embed_size], final word embeddings. vocab_words: a list of s

Constructor. Args: arch: string scalar, architecture ('skip_gram' or 'cbow'). algm: string scalar: training algorithm ('negative_samp

Constructor. Args: syn0_final: numpy array of shape [vocab_size, embed_size], final word embeddings. vocab: a list of strings

Constructor. Args: max_vocab_size: int scalar, maximum vocabulary size. If > 0, only the top `max_vocab_size` most frequent words

Constructor. Args: epochs: int scalar, num times the dataset is iterated. batch_size: int scalar, the returned tensors in `get_tensor

Constructor. Args: unigram_counts: a list of ints, the counts of word tokens in the corpus. arch: string scalar, architecture ('skip

Runs the forward pass to compute loss. Args: inputs: int tensor of shape [batch_size] (skip_gram) or [batch_size, 2*window_size+1]

Get the training step signatures for `inputs`, `labels` and `progress` tensor. Args: arch: string scalar, architecture ('skip_gram' or 'cbow

Finds the top-k words with smallest cosine distances w.r.t `word`. Args: word: string scalar, the query word. k: int scalar, num of w

Generate inputs and labels for each target word. `index` is the index of the target word in `indices`.