Weights Biases Low Code No Code

Weights & Biases Low-Code MLOps

Weights Biases W&B wandb MLOps Experiment Tracking Sweeps Hyperparameter Reports Artifacts Model Registry Low Code No Code

Experiment Tracking

# === W&B Experiment Tracking ===

# pip install wandb
# wandb login  # paste API key from wandb.ai/authorize

# Basic Usage (3 lines added to any training script)
# import wandb
#
# # Initialize run
# wandb.init(
#     project="image-classification",
#     config={
#         "learning_rate": 0.001,
#         "batch_size": 32,
#         "epochs": 100,
#         "model": "resnet50",
#         "optimizer": "adam",
#         "dropout": 0.3,
#     }
# )
#
# # Training loop
# for epoch in range(config.epochs):
#     train_loss, train_acc = train_one_epoch(model, train_loader)
#     val_loss, val_acc = evaluate(model, val_loader)
#
#     # Log metrics (1 line)
#     wandb.log({
#         "train/loss": train_loss,
#         "train/accuracy": train_acc,
#         "val/loss": val_loss,
#         "val/accuracy": val_acc,
#         "epoch": epoch,
#     })
#
# # Log final model
# wandb.save("model_best.pth")
# wandb.finish()

# PyTorch Lightning (1 line integration)
# from pytorch_lightning.loggers import WandbLogger
# wandb_logger = WandbLogger(project="my-project")
# trainer = Trainer(logger=wandb_logger)

# Hugging Face (1 config change)
# TrainingArguments(report_to="wandb", ...)

from dataclasses import dataclass

@dataclass
class Integration:
    framework: str
    code_lines: int
    method: str
    auto_tracked: str

integrations = [
    Integration("PyTorch (Manual)",
        3,
        "wandb.init() + wandb.log() + wandb.finish()",
        "Metrics Config System Info GPU"),
    Integration("PyTorch Lightning",
        1,
        "WandbLogger(project='name')",
        "Loss Metrics LR Schedule Gradients Model Graph"),
    Integration("Hugging Face",
        1,
        "report_to='wandb' ใน TrainingArguments",
        "Loss Eval Metrics Config Model Card"),
    Integration("Keras/TensorFlow",
        1,
        "WandbCallback() ใน model.fit(callbacks=[...])",
        "Loss Metrics Layer Weights Gradients"),
    Integration("Scikit-learn",
        3,
        "wandb.init() + wandb.sklearn.plot_*() + wandb.finish()",
        "Confusion Matrix ROC Feature Importance"),
    Integration("XGBoost/LightGBM",
        2,
        "wandb.xgboost.wandb_callback() หรือ lightgbm",
        "Training Metrics Feature Importance Tree"),
]

print("=== Framework Integrations ===")
for i in integrations:
    print(f"  [{i.framework}] Code Lines: {i.code_lines}")
    print(f"    Method: {i.method}")
    print(f"    Auto-tracked: {i.auto_tracked}")

Sweeps (Hyperparameter Tuning)

# === W&B Sweeps Configuration ===

# sweep_config.yaml
# method: bayes
# metric:
#   name: val/loss
#   goal: minimize
# parameters:
#   learning_rate:
#     min: 0.0001
#     max: 0.01
#   batch_size:
#     values: [16, 32, 64, 128]
#   optimizer:
#     values: ["adam", "sgd", "adamw"]
#   dropout:
#     min: 0.1
#     max: 0.5
#   weight_decay:
#     min: 0.0001
#     max: 0.01
# early_terminate:
#   type: hyperband
#   min_iter: 10

# Python:
# import wandb
#
# sweep_config = {
#     "method": "bayes",
#     "metric": {"name": "val/loss", "goal": "minimize"},
#     "parameters": {
#         "learning_rate": {"min": 0.0001, "max": 0.01},
#         "batch_size": {"values": [16, 32, 64, 128]},
#         "optimizer": {"values": ["adam", "sgd", "adamw"]},
#     }
# }
#
# sweep_id = wandb.sweep(sweep_config, project="sweep-demo")
# wandb.agent(sweep_id, function=train, count=50)

@dataclass
class SweepMethod:
    method: str
    algorithm: str
    best_for: str
    efficiency: str

methods = [
    SweepMethod("bayes",
        "Bayesian Optimization (Gaussian Process)",
        "ทั่วไป แนะนำ ใช้ Results เก่าเลือก Parameters ใหม่",
        "สูง (หา Optimum เร็ว 20-50 runs)"),
    SweepMethod("random",
        "Random Search สุ่มจาก Search Space",
        "Search Space กว้าง เริ่มต้น Explore",
        "ปานกลาง (ต้อง 50-100+ runs)"),
    SweepMethod("grid",
        "Grid Search ทดสอบทุก Combination",
        "Search Space เล็ก Parameters น้อย",
        "ต่ำ (Combination มาก ใช้เวลานาน)"),
]

print("=== Sweep Methods ===")
for m in methods:
    print(f"  [{m.method}] {m.algorithm}")
    print(f"    Best for: {m.best_for}")
    print(f"    Efficiency: {m.efficiency}")

Reports & Model Registry

# === W&B Reports & Model Registry ===

# Artifacts (Model Versioning)
# import wandb
#
# # Log Model as Artifact
# run = wandb.init(project="my-project")
# artifact = wandb.Artifact("my-model", type="model")
# artifact.add_file("model_best.pth")
# run.log_artifact(artifact)
#
# # Download Model Artifact
# run = wandb.init()
# artifact = run.use_artifact("my-model:latest")
# artifact_dir = artifact.download()

# Model Registry (No Code - UI)
# 1. Go to Model Registry tab
# 2. Link Artifact to Registry
# 3. Set Stage: Staging → Production
# 4. Add Description, Tags
# 5. View Lineage: Dataset → Run → Model

@dataclass
class WBFeature:
    feature: str
    code_needed: str
    ui_available: str
    use_case: str

features = [
    WBFeature("Experiment Dashboard",
        "No (Auto-generated)",
        "Full Dashboard Compare Filter Sort",
        "เปรียบเทียบ Run ดู Metrics Trend"),
    WBFeature("Reports",
        "No (Drag-and-drop UI)",
        "Interactive Markdown Charts Tables",
        "แชร์ Results กับทีม สรุป Experiment"),
    WBFeature("Artifacts",
        "2 บรรทัด (log_artifact)",
        "Browse Download Compare",
        "Version Dataset Model Files"),
    WBFeature("Model Registry",
        "No (UI + optional API)",
        "Stage Management Lineage Link",
        "Model Lifecycle Staging → Production"),
    WBFeature("Tables",
        "1 บรรทัด (wandb.Table)",
        "Interactive Filter Sort Visualize",
        "Data Analysis EDA Error Analysis"),
    WBFeature("Alerts",
        "No (UI Config)",
        "Slack Email Webhook",
        "แจ้งเตือนเมื่อ Metric ถึง Threshold"),
]

print("=== W&B Features ===")
for f in features:
    print(f"  [{f.feature}] Code: {f.code_needed}")
    print(f"    UI: {f.ui_available}")
    print(f"    Use: {f.use_case}")

เคล็ดลับ

• wandb.init: ใส่ config ทุกครั้ง เปรียบเทียบ Hyperparameters ง่าย
• Sweeps: ใช้ Bayes เริ่มต้น 20-50 runs หา Best Parameters
• Tags: ใช้ Tags แบ่งกลุ่ม Run เช่น "baseline" "v2" "experiment"
• Artifacts: Log Model ทุกครั้ง ใช้ Model Registry จัดการ Stage
• Reports: สร้าง Report ทุกสัปดาห์ แชร์ Progress กับทีม

Weights & Biases คืออะไร

MLOps Platform Experiment Tracking Sweeps Artifacts Reports Model Registry wandb PyTorch Hugging Face Keras Low-Code Free Team

สรุป

Weights Biases W&B wandb Low Code Experiment Tracking Sweeps Bayes Artifacts Model Registry Reports Dashboard MLOps Production