TensorBoard & W&B — Experiment Tracking

# pip install wandb tensorboard

import torch
import wandb

# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
# WEIGHTS & BIASES (W&B) — industry standard
# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

# Initialize run (creates a new experiment in W&B dashboard)
wandb.init(
    project="ai-course-experiments",
    name="resnet50-flowers-baseline",
    config={
        "architecture": "resnet50",
        "learning_rate": 1e-3,
        "batch_size": 64,
        "n_epochs": 30,
        "n_classes": 5,
        "augmentation": "random_flip_crop",
    }
)

# Log metrics every step
def training_loop_with_wandb(model, train_loader, val_loader, n_epochs=30):
    optimizer = torch.optim.AdamW(model.parameters(), lr=wandb.config.learning_rate)
    loss_fn = torch.nn.CrossEntropyLoss()

    for epoch in range(n_epochs):
        model.train()
        train_loss = 0
        for X, y in train_loader:
            optimizer.zero_grad()
            loss = loss_fn(model(X), y)
            loss.backward()
            optimizer.step()
            train_loss += loss.item()

            # Log every batch
            wandb.log({"train/batch_loss": loss.item()})

        # Log per-epoch metrics
        wandb.log({
            "train/epoch_loss": train_loss / len(train_loader),
            "epoch": epoch,
            "lr": optimizer.param_groups[0]["lr"],
        })

    # Save model artifact to W&B
    wandb.save("best_model.pt")

# Watch gradients (log gradient histograms automatically)
# wandb.watch(model, log="all", log_freq=100)

wandb.finish()

# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
# TENSORBOARD — built into PyTorch
# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
from torch.utils.tensorboard import SummaryWriter

writer = SummaryWriter("runs/experiment_1")

for epoch in range(10):
    train_loss = 0.5 - epoch * 0.04  # simulated
    val_loss   = 0.6 - epoch * 0.035

    writer.add_scalar("Loss/train", train_loss, epoch)
    writer.add_scalar("Loss/val",   val_loss,   epoch)
    writer.add_scalar("LR", 1e-3 * (0.9 ** epoch), epoch)

# Add model graph to TensorBoard
import torch.nn as nn
model = nn.Linear(10, 2)
dummy = torch.randn(1, 10)
writer.add_graph(model, dummy)

writer.close()

# Launch TensorBoard:
# tensorboard --logdir runs/
# Then open: http://localhost:6006

print("Experiment tracking setup complete!")
print("W&B dashboard: https://wandb.ai/your-username/ai-course-experiments")