keras: calculate the perplexity on penntreebank using LSTM keras got infinity

I am very new to KERAS, and I use the dealt dataset from the RNN Toolkit and try to use LSTM to train the language model I have problem with the calculating the perplexity though. It always get quite large negative log loss, and when using the exp function, it seems to get the infinity, I got stuck here. Below is my model code, and the github link( https://github.com/janenie/lstm_issu_keras ) is the current problematic code of mine. Can someone help me out?

the test_y data format is word index in sentences per sentence per line, so is the test_x.

class LSTMLM: def init(self, input_len, hidden_len, output_len, return_sequences=True): self.input_len = input_len self.hidden_len = hidden_len self.output_len = output_len self.seq = return_sequences self.model = Sequential()

def build(self, maxlen=50, dropout=0.2):
    self.model.add(Embedding(self.input_len, self.hidden_len, input_length=maxlen))
    self.model.add(LSTM(output_dim=self.hidden_len, return_sequences=True))
    #self.model.add(Dropout(dropout))
    self.model.add(TimeDistributedDense(self.output_len))
    self.model.add(Activation('softmax'))
    self.model.compile(loss='categorical_crossentropy', optimizer='rmsprop')

def train(self, train_x, train_y, dev_x, dev_y, batchsize=16, epoch=1):
    hist = self.model.fit(train_x, train_y, batch_size=batchsize, nb_epoch=1, \
                    show_accuracy=True, validation_data=(dev_x, dev_y))
    print hist.history

def saveModel(self, save_file):
    self.model.save_weights(save_file)

def computePPL(self, test_x, test_y, maxLen=100):
    predictions = self.model.predict(test_x)
    test_num = len(test_y)
    maxlen = predictions.shape[1]
    ppl = 0.0
    total_words = 0
    for i in xrange(test_num):
        pred_idx = test_y[i]
        #print np.max(pred_idx)
        sent_len = len(test_y[i])
        padding = maxlen - sent_len
        for j in xrange(sent_len):
            prob_ = predictions[i, j+padding, pred_idx[j]]
            ppl -= np.log(prob_)
        total_words += sent_len
    return np.exp(ppl)/(float(total_words)) 

Please make sure that the boxes below are checked before you submit your issue. Thank you!