model.py

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import os
import numpy as np

# source: Goswami, V. (n.d.). SnakeGameAI [Computer software]. GitHub. https://github.com/vedantgoswami/SnakeGameAI

class Linear_QNet(nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super().__init__()
        self.linear1 = nn.Linear(input_size, hidden_size)
        self.linear2 = nn.Linear(hidden_size, output_size)

    def forward(self, x):
        x = F.relu(self.linear1(x))
        x = self.linear2(x)
        return x

    def save(self, file_name='model.pth'):
        model_folder_path = os.path.join(os.path.dirname(__file__), 'model')
        os.makedirs(model_folder_path, exist_ok=True)
        file_path = os.path.join(model_folder_path, file_name)
        torch.save(self.state_dict(), file_path)

    def load_model(self, file_name='model.pth'):
        model_folder_path = os.path.join(os.path.dirname(__file__), 'model')
        file_path = os.path.join(model_folder_path, file_name)
        self.load_state_dict(torch.load(file_path))


class QTrainer:
    def __init__(self, model, lr, gamma):
        self.lr = lr
        self.gamma = gamma
        self.model = model
        self.optimer = optim.Adam(model.parameters(), lr=self.lr)
        self.criterion = nn.MSELoss()
        for i in self.model.parameters():
            print(i.is_cuda)

    def train_step(self, state, action, reward, next_state, done):
        state = torch.tensor(np.array(state), dtype=torch.float)
        next_state = torch.tensor(np.array(next_state), dtype=torch.float)
        action = torch.tensor(action, dtype=torch.long)
        reward = torch.tensor(reward, dtype=torch.float)

        if (len(state.shape) == 1):  # only one parameter to train , Hence convert to tuple of shape (1, x)
            # (1 , x)
            state = torch.unsqueeze(state, 0)
            next_state = torch.unsqueeze(next_state, 0)
            action = torch.unsqueeze(action, 0)
            reward = torch.unsqueeze(reward, 0)
            done = (done,)

        # 1. Predicted Q value with current state
        pred = self.model(state)
        target = pred.clone()
        for idx in range(len(done)):
            Q_new = reward[idx]
            if not done[idx]:
                Q_new = reward[idx] + self.gamma * torch.max(self.model(next_state[idx]))
            target[idx][torch.argmax(action).item()] = Q_new
            # 2. Q_new = reward + gamma * max(next_predicted Qvalue) -> only do this if not done
        # pred.clone()
        # preds[argmax(action)] = Q_new
        self.optimer.zero_grad()
        loss = self.criterion(target, pred)
        loss.backward()

        self.optimer.step()