Training Loop

Code

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader

def train(model, loader, epochs=5, lr=1e-3, device="cpu"):
    model.to(device).train()
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=lr)

    for epoch in range(epochs):
        total = 0.0
        for xb, yb in loader:
            xb, yb = xb.to(device), yb.to(device)
            optimizer.zero_grad()
            logits = model(xb)
            loss = criterion(logits, yb)
            loss.backward()
            optimizer.step()
            total += loss.item() * xb.size(0)
        print(f"epoch {epoch} loss={total / len(loader.dataset):.4f}")

@torch.no_grad()
def evaluate(model, loader, device="cpu"):
    model.to(device).eval()
    correct = 0
    for xb, yb in loader:
        xb, yb = xb.to(device), yb.to(device)
        preds = model(xb).argmax(1)
        correct += (preds == yb).sum().item()
    return correct / len(loader.dataset)

Explanation

Each step zeroes gradients, runs forward to compute a loss, calls backward to populate gradients, and steps the optimizer. train() and eval() toggle dropout and batch-norm behavior, and torch.no_grad() disables gradient tracking during evaluation. Averaging loss by sample count gives a per-epoch metric.

More PyTorch Snippets