pneuma-pygame/agents/ppo/brain.py

import os
import numpy as np
import torch as T
import torch.nn as nn
import torch.optim as optim
from torch.distributions.categorical import Categorical


class PPOMemory:
    def __init__(self, batch_size):
        self.states = []
        self.probs = []
        self.vals = []
        self.actions = []
        self.rewards = []
        self.dones = []

        self.batch_size = batch_size

    def generate_batches(self):

        n_states = len(self.states)
        batch_start = np.arange(0, n_states, self.batch_size)
        indices = np.arange(n_states, dtype=np.int64)
        np.random.shuffle(indices)
        batches = [indices[i:i+self.batch_size] for i in batch_start]

        return np.array(self.states),\
            np.array(self.actions),\
            np.array(self.probs),\
            np.array(self.vals),\
            np.array(self.rewards),\
            np.array(self.dones),\
            batches

    def store_memory(self, state, action, probs, vals, reward, done):
        self.states.append(state)
        self.actions.append(action)
        self.probs.append(probs)
        self.vals.append(vals)
        self.rewards.append(reward)
        self.dones.append(done)

    def clear_memory(self):
        self.states = []
        self.probs = []
        self.vals = []
        self.actions = []
        self.rewards = []
        self.dones = []


class ActorNetwork(nn.Module):

    def __init__(self, input_dim, output_dim, alpha, fc1_dims=256, fc2_dims=256, chkpt_dir='tmp/ppo'):
        super(ActorNetwork, self).__init__()

        self.chkpt_dir = chkpt_dir

        self.actor = nn.Sequential(
            nn.Linear(input_dim, fc1_dims),
            nn.ReLU(),
            nn.Linear(fc1_dims, fc2_dims),
            nn.ReLU(),
            nn.Linear(fc2_dims, output_dim),
            nn.Softmax(dim=-1)
        )

        self.optimizer = optim.Adam(self.parameters(), lr=alpha, eps=1e-5)

        self.device = T.device('cuda:0' if T.cuda.is_available() else 'cpu')

        self.to(self.device)

    def forward(self, state):
        dist = self.actor(state)
        dist = Categorical(dist)

        return dist

    def save_checkpoint(self, filename = 'actor_torch_ppo'):
        T.save(self.state_dict(), os.path.join(self.chkpt_dir, filename))

    def load_checkpoint(self, filename = 'actor_torch_ppo'):
        self.load_state_dict(T.load(os.path.join(self.chkpt_dir, filename), map_location=self.device))


class CriticNetwork(nn.Module):

    def __init__(self, input_dims, alpha, fc1_dims=256, fc2_dims=256, chkpt_dir='tmp/ppo'):
        super(CriticNetwork, self).__init__()

        self.chkpt_dir = chkpt_dir

        self.critic = nn.Sequential(
            nn.Linear(input_dims, fc1_dims),
            nn.ReLU(),
            nn.Linear(fc1_dims, fc2_dims),
            nn.ReLU(),
            nn.Linear(fc2_dims, 1)
        )

        self.optimizer = optim.Adam(self.parameters(), lr=alpha, eps=1e-5)

        self.device = T.device('cuda:0' if T.cuda.is_available() else 'cpu')

        self.to(self.device)

    def forward(self, state):
        value = self.critic(state)
        return value

    def save_checkpoint(self, filename = 'critic_torch_ppo'):
        T.save(self.state_dict(), os.path.join(self.chkpt_dir, filename))

    def load_checkpoint(self, filename = 'critic_torch_ppo'):
        self.load_state_dict(T.load(os.path.join(self.chkpt_dir, filename), map_location=self.device))