b954a0c942f08b69017eeaee77d2c856bcdd6ef7,ml/rl/training/ddpg_trainer.py,CriticNet,forward_split,#CriticNet#Any#Any#,361

Before Change

                x = torch.cat((x, action), dim=1)
            activation_func = getattr(F, activation)
            fc_func = self.layers[i]
            x = fc_func(x) if activation == "linear" else activation_func(fc_func(x))
        return x


def fan_in_init(weight_tensor, bias_tensor) -> None:

After Change

            // Actions skip input layer
            elif i == 1:
                x = self.batch_norm_ops[i](x)
                x = torch.cat((x, action), dim=1)

            x = self.layers[i](x)
            if activation == "linear":
                continue
            elif activation == "tanh":
                activation_func = torch.tanh
            else:
                activation_func = getattr(F, activation)
            x = activation_func(x)
        return x


def fan_in_init(weight_tensor, bias_tensor) -> None:

Instances