11685f68d52266188abc117687493c80a5580718,keras/layers/recurrent.py,MutatedRNN_1,__init__,#MutatedRNN_1#Any#Any#Any#Any#Any#Any#Any#Any#Any#,392

Before Change

        self.truncate_gradient = truncate_gradient
        self.return_sequences = return_sequences
        
        if self.input_dim != self.output_dim:
            raise Exception("input_dim must equal output_dim for this architecture")

        self.init = initializations.get(init)
        self.inner_init = initializations.get(inner_init)
        self.activation = activations.get(activation)
        self.inner_activation = activations.get(inner_activation)

After Change

        self.b_h = shared_zeros((self.output_dim))

        // P_h used to project X onto different dimension
        P = np.ones((self.input_dim, self.output_dim), dtype=theano.config.floatX)
        P = P / np.linalg.norm(P, axis=0)
        self.Pmat = theano.shared(P, name=None)

        self.params = [
            self.W_z, self.b_z,
            self.W_r, self.U_r, self.b_r,

Instances