model.py

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.distributions import Normal
import numpy as np


# GILD: Generalized Imitation Lerning from Demonstrations
class GILD_Network(nn.Module):
    def __init__(self, hidden_dim):
        super(GILD_Network, self).__init__()
        self.fc1 = nn.Linear(hidden_dim,256)
        self.fc2 = nn.Linear(256,256)
        self.fc3 = nn.Linear(256,1)
    def forward(self, x):
        x = torch.relu(self.fc1(x))
        x = torch.relu(self.fc2(x))
        x = F.softplus(self.fc3(x))
        return x

# Actor for DDPG and TD3
class Actor(nn.Module):
    def __init__(self, state_dim, action_dim, max_action):
        super(Actor, self).__init__()
        self.l1 = nn.Linear(state_dim, 256)
        self.l2 = nn.Linear(256, 256)
        self.l3 = nn.Linear(256, action_dim)
        self.max_action = max_action

    def forward(self, x):
        x = F.relu(self.l1(x))
        x1 = F.relu(self.l2(x))
        x2 = self.max_action * torch.tanh(self.l3(x1))
        return x2

# Critic for DDPG and TD3
class Critic(nn.Module):
    def __init__(self, state_dim, action_dim):
        super(Critic, self).__init__()

        self.l1 = nn.Linear(state_dim + action_dim, 256)
        self.l2 = nn.Linear(256, 256)
        self.l3 = nn.Linear(256, 1)

    def forward(self, x, u):
        x = F.relu(self.l1(torch.cat([x, u], 1)))
        x = F.relu(self.l2(x))
        x = self.l3(x)
        return x



# Initialize Policy weights
def weights_init_(m):
    if isinstance(m, nn.Linear):
        torch.nn.init.xavier_uniform_(m.weight, gain=1)
        torch.nn.init.constant_(m.bias, 0)

# For Critic for TD3 and SCA
class Critic_DoubleQ(nn.Module):
    def __init__(self, num_inputs, num_actions, hidden_dim):
        super(Critic_DoubleQ, self).__init__()

        # Q1 architecture
        self.linear1 = nn.Linear(num_inputs + num_actions, hidden_dim)
        self.linear2 = nn.Linear(hidden_dim, hidden_dim)
        self.linear3 = nn.Linear(hidden_dim, 1)

        # Q2 architecture
        self.linear4 = nn.Linear(num_inputs + num_actions, hidden_dim)
        self.linear5 = nn.Linear(hidden_dim, hidden_dim)
        self.linear6 = nn.Linear(hidden_dim, 1)

        self.apply(weights_init_)

    def forward(self, state, action):
        xu = torch.cat([state, action], 1)
        
        x1 = F.relu(self.linear1(xu))
        x1 = F.relu(self.linear2(x1))
        x1 = self.linear3(x1)

        x2 = F.relu(self.linear4(xu))
        x2 = F.relu(self.linear5(x2))
        x2 = self.linear6(x2)

        return x1, x2
    
    def Q1(self, state, action):
        sa = torch.cat([state, action], 1)

        q1 = F.relu(self.linear1(sa))
        q1 = F.relu(self.linear2(q1))
        q1 = self.linear3(q1)
        return q1
    
LOG_SIG_MAX = 2
LOG_SIG_MIN = -20
epsilon = 1e-6

class GaussianPolicy(nn.Module):
    def __init__(self, num_inputs, num_actions, hidden_dim, action_space=None):
        super(GaussianPolicy, self).__init__()
        
        self.linear1 = nn.Linear(num_inputs, hidden_dim)
        self.linear2 = nn.Linear(hidden_dim, hidden_dim)

        self.mean_linear = nn.Linear(hidden_dim, num_actions)
        self.log_std_linear = nn.Linear(hidden_dim, num_actions)

        self.apply(weights_init_)

        # action rescaling
        if action_space is None:
            self.action_scale = torch.tensor(1.)
            self.action_bias = torch.tensor(0.)
        else:
            self.action_scale = torch.FloatTensor(
                (action_space.high - action_space.low) / 2.)
            self.action_bias = torch.FloatTensor(
                (action_space.high + action_space.low) / 2.)

    def forward(self, state):
        x = F.relu(self.linear1(state))
        x = F.relu(self.linear2(x))
        mean = self.mean_linear(x)
        log_std = self.log_std_linear(x)
        log_std = torch.clamp(log_std, min=LOG_SIG_MIN, max=LOG_SIG_MAX)
        return mean, log_std

    def sample(self, state):
        mean, log_std = self.forward(state)
        std = log_std.exp()
        normal = Normal(mean, std)
        pi_action = normal.rsample()

        log_prob = normal.log_prob(pi_action).sum(axis=-1, keepdim=True)
        log_prob -= (2*(np.log(2) - pi_action - F.softplus(-2*pi_action))).sum(axis=1, keepdim=True)

        # Enforcing Action Bound
        action = torch.tanh(pi_action) * self.action_scale + self.action_bias
        mean = torch.tanh(mean) * self.action_scale + self.action_bias

        return action, log_prob, mean

    def log_prob(self, state, action):
        mean, log_std = self.forward(state)
        std = log_std.exp()
        normal = Normal(mean, std)
        return normal.log_prob(action).mean(dim=1, keepdim=True)

    def to(self, device):
        self.action_scale = self.action_scale.to(device)
        self.action_bias = self.action_bias.to(device)
        return super(GaussianPolicy, self).to(device)


class DeterministicPolicy(nn.Module):
    def __init__(self, num_inputs, num_actions, hidden_dim, action_space=None):
        super(DeterministicPolicy, self).__init__()
        self.linear1 = nn.Linear(num_inputs, hidden_dim)
        self.linear2 = nn.Linear(hidden_dim, hidden_dim)

        self.mean = nn.Linear(hidden_dim, num_actions)
        self.noise = torch.Tensor(num_actions)

        self.apply(weights_init_)

        # action rescaling
        if action_space is None:
            self.action_scale = 1.
            self.action_bias = 0.
        else:
            self.action_scale = torch.FloatTensor(
                (action_space.high - action_space.low) / 2.)
            self.action_bias = torch.FloatTensor(
                (action_space.high + action_space.low) / 2.)

    def forward(self, state):
        x = F.relu(self.linear1(state))
        x = F.relu(self.linear2(x))
        mean = torch.tanh(self.mean(x)) * self.action_scale + self.action_bias
        return mean

    def sample(self, state):
        mean = self.forward(state)
        noise = self.noise.normal_(0., std=0.1)
        noise = noise.clamp(-0.25, 0.25)
        action = mean + noise
        return action, torch.tensor(0.), mean

    def to(self, device):
        self.action_scale = self.action_scale.to(device)
        self.action_bias = self.action_bias.to(device)
        self.noise = self.noise.to(device)
        return super(DeterministicPolicy, self).to(device)