Experiment Tracking

import wandb

wandb.init(project="tracking-demo")

for step in range(1000):
    # Log scalar metrics
    wandb.log({
        "train/loss": compute_loss(train_data),
        "train/accuracy": compute_accuracy(train_data),
        "val/loss": compute_loss(val_data),
        "val/accuracy": compute_accuracy(val_data),
        "learning_rate": scheduler.get_last_lr()[0],
    })

# Log summary metrics (final values)
wandb.summary["best_accuracy"] = best_accuracy
wandb.summary["total_params"] = count_params(model)

import wandb
import matplotlib.pyplot as plt
import numpy as np

# Log images
images = [wandb.Image(img, caption=f"Sample {i}")
          for i, img in enumerate(sample_images[:8])]
wandb.log({"predictions": images})

# Log matplotlib plots
fig, ax = plt.subplots()
ax.plot(history['loss'])
ax.set_title("Training Loss")
wandb.log({"loss_plot": wandb.Image(fig)})
plt.close()

# Log confusion matrix
wandb.log({"confusion_matrix": wandb.plot.confusion_matrix(
    y_true=y_test, preds=y_pred,
    class_names=class_names
)})

# Log audio
wandb.log({"audio_sample": wandb.Audio(
    audio_array, sample_rate=16000, caption="Prediction"
)})

# Log 3D point clouds
wandb.log({"point_cloud": wandb.Object3D(points)})

# Create a table for detailed analysis
table = wandb.Table(columns=["image", "prediction", "actual", "confidence"])

for img, pred, actual, conf in zip(images, preds, actuals, confidences):
    table.add_data(
        wandb.Image(img),
        class_names[pred],
        class_names[actual],
        conf
    )

wandb.log({"predictions_table": table})

# Log a pandas DataFrame as a table
import pandas as pd
df = pd.DataFrame(results)
wandb.log({"results": wandb.Table(dataframe=df)})

import wandb

wandb.init(project="pytorch-demo")

# Watch model gradients and parameters
wandb.watch(model, log="all", log_freq=100)

# Training loop
for epoch in range(epochs):
    for batch in dataloader:
        loss = train_step(model, batch)
        wandb.log({"loss": loss})

# Log the final model
torch.save(model.state_dict(), "model.pt")
wandb.save("model.pt")

from sklearn.ensemble import RandomForestClassifier
import wandb

wandb.init(project="sklearn-demo")

# Log sklearn model params automatically
model = RandomForestClassifier(n_estimators=100, max_depth=10)
wandb.config.update(model.get_params())

model.fit(X_train, y_train)
accuracy = model.score(X_test, y_test)
wandb.log({"accuracy": accuracy})

# Log ROC curve
wandb.log({"roc": wandb.plot.roc_curve(y_test, model.predict_proba(X_test))})