vihds/encoders.py [147:169]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        if self.condition_treatments:
            x = torch.cat((x, conds), 1)
        if self.condition_devices:
            x = torch.cat((x, devs), 1)
        params = OrderedDict()
        for free_name, constrained_name, free_to_constrained in zip(
            self.description.free_params, self.description.params, self.description.free_to_constrained,
        ):

            free_param = self.layers[free_name](x)
            # TODO: Torch equivalent of tf.stop_gradient
            # if stop_grad:
            #    free_param = tf.stop_gradient(free_param)  # eliminate score function term from autodiff
            constrained = constrain_parameter(free_param, free_to_constrained)
            params[free_name] = free_param
            params[constrained_name] = constrained

        for (other_param_name, other_param_value,) in self.description.other_params.items():
            params[other_param_name] = other_param_value

        new_distribution = self.description.class_type(wait_for_assigned=True, variable=True)
        new_distribution.assign_free_and_constrained(**params)
        return new_distribution
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



vihds/encoders.py [189:213]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        if self.condition_treatments:
            x = torch.cat((x, conds), 1)
        if self.condition_devices:
            x = torch.cat((x, devs), 1)
        params = OrderedDict()
        for free_name, constrained_name, free_to_constrained in zip(
            self.description.free_params, self.description.params, self.description.free_to_constrained,
        ):

            free_param = self.layers[free_name](x)
            # TODO: Torch equivalent of tf.stop_gradient
            # if stop_grad:
            #    free_param = tf.stop_gradient(free_param)  # eliminate score function term from autodiff
            # TODO: Implement Tensorboard summaries
            # variable_summaries(free_param, 'nn_%s'%name, plot_histograms)
            constrained = constrain_parameter(free_param, free_to_constrained)
            params[free_name] = free_param
            params[constrained_name] = constrained

        for (other_param_name, other_param_value,) in self.description.other_params.items():
            params[other_param_name] = other_param_value

        new_distribution = self.description.class_type(wait_for_assigned=True, variable=True)
        new_distribution.assign_free_and_constrained(**params)
        return new_distribution
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -