def __init__(self,

corpus,

tokenCount,

rng,

embeddingDim,

nodesNum=2,

nodesSize=(2, 2),

datatype=theano.config.floatX,

sentenceW=None,

sentenceB=None,

W=None,

B=None,

pooling_mode="average_exc_pad"):

self.__embeddingDim = embeddingDim

self.__nodesNum = nodesNum

self.__nodesSize = nodesSize

self.__WBound = 0.2

self.__MAXDIM = 10000

self.__datatype = datatype

self.__pooling_mode = pooling_mode

# For DomEmbeddingNN optimizer.

# self.shareRandge = T.arange(maxRandge)

# Get sentence layer W

if W is None:

W = theano.shared(

numpy.asarray(

rng.uniform(low=-self.__WBound, high=self.__WBound, size=(nodesNum, nodesSize[0], nodesSize[1])),

dtype=datatype

),

borrow=True

)

# Get sentence layer b

if B is None:

B0 = numpy.zeros((nodesNum,), dtype=datatype)

B = theano.shared(value=B0, borrow=True)

self.output, _ = theano.scan(fn=self.__dealWithSentence,

non_sequences=[corpus, W, B],

sequences=[dict(input=tokenCount, taps=[-1, -0])],

strict=True)

self.W = W

self.B = B

self.params = [self.W, self.B]

self.outputDimension = nodesNum * (embeddingDim - nodesSize[1] + 1)

def __dealWithSentence(self, wc0, wc1, docs, W, B):

sentence = docs[wc0:wc1]

sentence_out = conv.conv2d(input=sentence, filters=W)

sentence_pool = downsample.max_pool_2d(sentence_out, (self.__MAXDIM, 1), mode=self.__pooling_mode, ignore_border=False)

sentence_output = T.tanh(sentence_pool + B.dimshuffle([0, 'x', 'x']))

sentence_embedding = sentence_output.flatten(1)