Pick Layer and LSTM peephole linear activation

Merge pull request #40 from Yoctol/LSTMPeephole_linear_activation

use linear activation in lstm peephole

Fix RNNCell constraint bug and mask_to_seq bug

Merge pull request #38 from Yoctol/mask_to_seq_bug

fix mask_to_seq bug

Fix RNNCell constraint bug

Merge pull request #37 from Yoctol/rnn_cell_constraint_fix

fix rnn_cell constraint bug

RNNCell and Seq2Seq RNN decoder

We provide a RNNCell layer to add a Dense layer behind any Keras recurrent layer. The cell behaves just like a Keras recurrent layer. We also extend the function of our RNNDecoder to decode sequences whose length are different with the input sequences.

Masked ConvNet wrapper and pooling wrapper

We refactor our convolution2D layer into a general MaskConvNet wrapper which supports masking function for any Conv layers in Keras. Also, we provide a similar wrapper for Keras pooling layers and a MaskConv layer to mask any high dimensional data for convolution network.

RNN encoder and decoder and 2D masked ConvNet

In this release, we extend LSTM encoder and decoder to RNN encoder and decoder wrapper.
Also, we provide masked 2D ConvNet to deal with unfixed length of sentence.

0.3.1: Merge pull request #7 from Yoctol/add_bidirectional_wrapper

Add bidirectional wrapper

0.3.0: Merge pull request #6 from Yoctol/refactor_lstm_decoder

refactor LSTM decoder