model.py

# This is inspired from:
# https://github.com/udacity/deep-reinforcement-learning/blob/55474449a112fa72323f484c4b7a498c8dc84be1/ddpg-bipedal/model.py

import numpy as np

import torch
import torch.nn as nn
import torch.nn.functional as F

# use for NN weight initialization
def hidden_init(layer):
    fan_in = layer.weight.data.size()[0]
    lim = 1. / np.sqrt(fan_in)
    return (-lim, lim)

class Actor(nn.Module):
    """Actor (Policy) Model."""

    def __init__(self, state_size, action_size, seed, fc1_units=128, fc2_units=256):
        super(Actor, self).__init__()
        self.seed = torch.manual_seed(seed)
        self.fc1 = nn.Linear(state_size, fc1_units)
        # After reading the slack channel, one suggest improvement using a batch normalisation layer
        # I do not notice better convergence with this additional layer, so I remove it before the submission
        #self.bn1 = nn.BatchNorm1d(fc1_units)
        self.fc2 = nn.Linear(fc1_units, fc2_units)
        self.fc3 = nn.Linear(fc2_units, action_size)
        self.reset_parameters()

    def reset_parameters(self):
        self.fc1.weight.data.uniform_(*hidden_init(self.fc1))
        self.fc2.weight.data.uniform_(*hidden_init(self.fc2))
        self.fc3.weight.data.uniform_(-3e-3, 3e-3)

    def forward(self, state):
        """Build an actor (policy) network that maps states -> actions."""
        # I do not notice better convergence with this additional batch normalization layer, so I remove it before the submission
        #x = F.relu(self.bn1(self.fc1(state)))
        x = F.relu(self.fc1(state))
        x = F.relu(self.fc2(x))
        return F.tanh(self.fc3(x))


class Critic(nn.Module):
    """Critic (Value) Model."""

    def __init__(self, state_size, action_size, seed, fc1_units=128, fc2_units=256):
        super(Critic, self).__init__()
        self.seed = torch.manual_seed(seed)
        self.fc1 = nn.Linear(state_size, fc1_units)
        # After reading the slack channel, one suggest improvement using a batch normalisation layer
        # I do not notice better convergence with this additional layer, so I remove it before the submission
        #self.bn1 = nn.BatchNorm1d(fc1_units)
        self.fc2 = nn.Linear(fc1_units+action_size, fc2_units)
        self.fc3 = nn.Linear(fc2_units, 1)
        self.reset_parameters()

    def reset_parameters(self):
        self.fc1.weight.data.uniform_(*hidden_init(self.fc1))
        self.fc2.weight.data.uniform_(*hidden_init(self.fc2))
        self.fc3.weight.data.uniform_(-3e-3, 3e-3)

    def forward(self, state, action):
        """Build a critic (value) network that maps (state, action) pairs -> Q-values."""
        # I do not notice better convergence with this additional batch normalization layer, so I remove it before the submission
        # x = F.relu(self.bn1(self.fc1(state)))
        x = F.relu(self.fc1(state))
        # Let's try leaky relu as suggested in the slack channel (it seems to be much slower)
        #x = F.leaky_relu(self.fc1(state))
        x = torch.cat((x, action), dim=1)
        x = F.relu(self.fc2(x))
        #x = F.leaky_relu(self.fc2(x))
        return self.fc3(x)