CODEWITHSUNDEEP
pythontensorflowkeraslstm
user8038245
user8038245

Reputation:

Define custom LSTM Cell in Keras?

I use Keras with TensorFlow as back-end. If I want to make a modification to an LSTM cell, such as "removing" the output gate, how can I do it? It is a multiplicative gate, so somehow I will have to set it to fixed values so that whatever multiplies it, has no effect.

Upvotes: 7

Views: 10078

Answers (1)

Amir
Amir

Reputation: 16587

First of all, you should define your own custom layer. If you need some intuition how to implement your own cell see LSTMCell in Keras repository. E.g. your custom cell will be:

class MinimalRNNCell(keras.layers.Layer):

    def __init__(self, units, **kwargs):
        self.units = units
        self.state_size = units
        super(MinimalRNNCell, self).__init__(**kwargs)

    def build(self, input_shape):
        self.kernel = self.add_weight(shape=(input_shape[-1], self.units),
                                      initializer='uniform',
                                      name='kernel')
        self.recurrent_kernel = self.add_weight(
            shape=(self.units, self.units),
            initializer='uniform',
            name='recurrent_kernel')
        self.built = True

    def call(self, inputs, states):
        prev_output = states[0]
        h = K.dot(inputs, self.kernel)
        output = h + K.dot(prev_output, self.recurrent_kernel)
        return output, [output]

Then, use tf.keras.layers.RNN to use your cell:

cell = MinimalRNNCell(32)
x = keras.Input((None, 5))
layer = RNN(cell)
y = layer(x)

# Here's how to use the cell to build a stacked RNN:

cells = [MinimalRNNCell(32), MinimalRNNCell(64)]
x = keras.Input((None, 5))
layer = RNN(cells)
y = layer(x)

Upvotes: 11

Related Questions