v6 - Save Pipeline¶

Get the Video Course

Progress¶

  ✅ STEP 1: Load Data
  ✅ STEP 2: Split Train/Test
  ✅ STEP 3: Create Pipeline
  ✅ STEP 4: Train Pipeline
  ✅ STEP 5: Evaluate & Metrics
  🟡 STEP 6: Save Pipeline           ← NEW IN THIS VERSION
  ⬜ STEP 7: Save Metrics JSON

What's New¶

New Function¶

# utils/model_utils.py
def save_pipeline(pipeline, model_name, metrics=None, metadata=None):
    """Save Pipeline pickle + metadata JSON to models/ directory."""

What Gets Saved¶

models/p3_01_baseline/
├── model_latest.pkl              # Trained Pipeline (pickle)
├── model_20260323_143052.pkl     # Timestamped backup
├── metadata_latest.json          # Metrics + model info
└── metadata_20260323_143052.json # Timestamped backup

The latest files are always the most recent - used by the inference API later. Timestamped files provide version history.

Key Concept: Manual Model Saving¶

This is the manual way to save models. Notice what we have to manage ourselves: - File paths and directories - Pickle serialization - Metadata JSON with metrics - Timestamped backups

Coming later: MLflow replaces all of this with mlflow.sklearn.log_model(pipeline) - one line instead of 30+.

How to Run¶

cd v6_save_pipeline_ccfd-project/

# PRE-REQUISITE: Generate data first
python p1_01_generate_initial_dataset.py

# Run training script
python p3_01_train_model_baseline.py

Expected Output (STEP 6):

STEP 6: Saving Pipeline

Model Saved: p3_01_baseline
   Location: models/p3_01_baseline/
   Model: LogisticRegression
   Timestamp: 20260323_143052

Next Version¶

v7 → Implement save_metrics_json() - save detailed metrics to a JSON file for tracking.

Back to 04 A - Train Baseline Model