def getContextsShape(self):
with open(self.contextsFilePath) as contextsFile:
contextsCount = binary.readi(contextsFile)
contextSize = binary.readi(contextsFile)
negative = binary.readi(contextsFile)
return contextsCount, contextSize, negative
def loadEmbeddings(embeddingsFilePath):
with open(embeddingsFilePath, 'rb') as embeddingsFile:
embeddingsCount = binary.readi(embeddingsFile)
embeddingSize = binary.readi(embeddingsFile)
embeddings = numpy.empty((embeddingsCount, embeddingSize)).astype('float32')
for embeddingIndex in range(0, embeddingsCount):
embedding = binary.readf(embeddingsFile, embeddingSize)
embeddings[embeddingIndex] = embedding
log.progress('Loading embeddings: {0:.3f}%.', embeddingIndex + 1, embeddingsCount)
log.info('Loading embeddings complete. {0} embeddings loaded.', embeddingsCount)
return embeddings
def loadMap(indexMapFilePath, inverse=False):
vocabulary = collections.OrderedDict()
with open(indexMapFilePath, 'rb') as indexMapFile:
itemsCount = binary.readi(indexMapFile)
for itemIndex in range(0, itemsCount):
wordLength = binary.readi(indexMapFile)
word = binary.reads(indexMapFile, wordLength)
index = binary.readi(indexMapFile)
if inverse:
vocabulary[index] = word
else:
vocabulary[word] = index
log.progress('Loading word map: {0:.3f}%.', itemIndex + 1, itemsCount)
log.info('Loading word map complete. {0} words loaded.', itemsCount)
return vocabulary
def getContexts(self, start, stop, step):
if step == 1:
with open(self.contextsFilePath) as contextsFile:
count = stop - start
contextSize = self.windowSize + self.negative
contextsSize = count * contextSize
contextBufferSize = contextSize * 4
# 12 for sizeof(contextsCount) + sizeof(contextSize) + sizeof(negative)
startPosition = start * contextBufferSize + 12
contextsFile.seek(startPosition, io.SEEK_SET)
contexts = binary.readi(contextsFile, contextsSize)
contexts = numpy.reshape(contexts, (count, (self.windowSize + self.negative)))
else:
contexts = []
for contextIndex in xrange(start, stop, step):
context = self[contextIndex][0]
contexts.append(context)
contexts = numpy.asarray(contexts, dtype='int32')
return contexts
