af461df627ef660d1a71b6981dedb4e4b504ba9a,txtgen/modules/connectors/connectors.py,StochasticConnector,_build,#StochasticConnector#Any#,256

Before Change

        modules = ["txtgen.custom", "txtgen.core.distributions"]
        sampler = get_function(self.hparams.distribution, modules)

        if sampler is distributions.sample_gaussian:
            if not isinstance(inputs, tuple):
                raise ValueError(
                    "Gaussian connector requires tuple input tensors.")

            if len(inputs) == 2:
                input_mu, input_log_var = inputs
                output = sampler(input_mu, input_log_var)

            elif len(inputs) == 3:
                input_mu, input_log_var, context = inputs
                sample = sampler(input_mu, input_log_var)
                output = tf.concat([sample, context], axis=1)
            else:
                raise ValueError("Gaussian connector supports either "
                                 "(mu, logvar) or (mu, logvar, context)")

        else:
            raise ValueError("Unsupported distribution")

        self._add_internal_trainable_variables()
        self._built = True

        return output

After Change

        try:
            nest.assert_same_structure(inputs, self._decoder_state_size)
        except (ValueError, TypeError):
            flat_input = nest.flatten(inputs)
            output = nest.pack_sequence_as(
                self._decoder_state_size, flat_input)

        self._add_internal_trainable_variables()
        self._built = True

Instances