Created
October 12, 2016 01:06
-
-
Save tsuchm/82c598750fa687c0d5c21c28fa9dc00b to your computer and use it in GitHub Desktop.
Example of LSTM encoder which ignores labels for padding
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import chainer | |
from chainer import Link, Chain, Function, Variable, cuda | |
import chainer.functions as F | |
import chainer.links as L | |
import numpy as np | |
class LSTMEncoder(Chain): | |
def __init__(self, vocab_size, embed_size, hidden_size, ignore_label=-1): | |
super(LSTMEncoder, self).__init__( | |
xe = L.EmbedID(vocab_size, embed_size, ignore_label=-1), | |
eh = L.Linear(embed_size, 4 * hidden_size), | |
hh = L.Linear(hidden_size, 4 * hidden_size), | |
) | |
self.ignore_label = ignore_label | |
def __call__(self, x, c_prev, h_prev): | |
xp = cuda.get_array_module(x.data) | |
e = F.tanh(self.xe(x)) | |
c, h = F.lstm(c_prev, self.eh(e) + self.hh(h_prev)) | |
ignore = xp.broadcast_to(xp.reshape((x.data == self.ignore_label), (x.shape[0], 1)), c.shape) | |
c_next = F.where(ignore, c_prev, c) | |
h_next = F.where(ignore, h_prev, h) | |
return c_next, h_next |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment