gogo2/NN/models/simple_cnn.py
2025-03-31 11:12:45 +03:00

120 lines
4.0 KiB
Python

import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import os
import logging
import torch.nn.functional as F
from typing import List, Tuple
# Configure logger
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class CNNModelPyTorch(nn.Module):
"""
CNN model for trading with multiple timeframes
"""
def __init__(self, window_size, num_features, output_size, timeframes):
super(CNNModelPyTorch, self).__init__()
self.window_size = window_size
self.num_features = num_features
self.output_size = output_size
self.timeframes = timeframes
# Calculate total input features across all timeframes
self.total_features = num_features * len(timeframes)
# Device configuration
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
logger.info(f"Using device: {self.device}")
# Convolutional layers
self.conv1 = nn.Conv1d(self.total_features, 64, kernel_size=3, padding=1)
self.bn1 = nn.BatchNorm1d(64)
self.conv2 = nn.Conv1d(64, 128, kernel_size=3, padding=1)
self.bn2 = nn.BatchNorm1d(128)
self.conv3 = nn.Conv1d(128, 256, kernel_size=3, padding=1)
self.bn3 = nn.BatchNorm1d(256)
# Calculate size after convolutions
conv_output_size = window_size * 256
# Fully connected layers
self.fc1 = nn.Linear(conv_output_size, 512)
self.fc2 = nn.Linear(512, 256)
# Advantage and Value streams (Dueling DQN architecture)
self.fc3 = nn.Linear(256, output_size) # Advantage stream
self.value_fc = nn.Linear(256, 1) # Value stream
# Initialize optimizer and scheduler
self.optimizer = optim.Adam(self.parameters(), lr=0.001)
self.scheduler = optim.lr_scheduler.ReduceLROnPlateau(
self.optimizer, mode='max', factor=0.5, patience=5, verbose=True
)
# Move model to device
self.to(self.device)
def forward(self, x: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
"""Forward pass through the network"""
# Ensure input is on the correct device
x = x.to(self.device)
# Reshape input: [batch, window_size, features] -> [batch, channels, window_size]
batch_size = x.size(0)
x = x.permute(0, 2, 1)
# Convolutional layers
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = F.relu(self.bn3(self.conv3(x)))
# Flatten
x = x.view(batch_size, -1)
# Fully connected layers
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
# Split into advantage and value streams
advantage = self.fc3(x)
value = self.value_fc(x)
# Combine value and advantage
q_values = value + (advantage - advantage.mean(dim=1, keepdim=True))
return q_values, value
def predict(self, X):
"""Make predictions"""
self.eval()
# Convert to tensor if not already
if not isinstance(X, torch.Tensor):
X_tensor = torch.tensor(X, dtype=torch.float32).to(self.device)
else:
X_tensor = X.to(self.device)
with torch.no_grad():
q_values, value = self(X_tensor)
q_values_np = q_values.cpu().numpy()
actions = np.argmax(q_values_np, axis=1)
return actions, q_values_np
def save(self, path: str):
"""Save model weights"""
os.makedirs(os.path.dirname(path), exist_ok=True)
torch.save(self.state_dict(), f"{path}.pt")
logger.info(f"Model saved to {path}.pt")
def load(self, path: str):
"""Load model weights"""
self.load_state_dict(torch.load(f"{path}.pt", map_location=self.device))
self.eval()
logger.info(f"Model loaded from {path}.pt")