mlops-prototyping

MLOps Prototyping

Goal

To create standardized, reproducible, and production-ready prototypes in Jupyter notebooks. This skill enforces a structured layout (Imports -> Configs -> Load -> EDA -> Modeling -> Eval) and robust engineering practices (Pipelines, Split-Verification) to prevent technical debt and data leakage.

Prerequisites

• Language: Python 3.14
• Environment: uv managed project (.venv), with ipykernel in a notebook dependency group
• Context: Executed within a .ipynb file or converting to one.

Instructions

1. Notebook Structure

Enforce the following linear sections in every notebook to ensure readability and maintainability.

1. Title & Purpose: H1 Title and a brief description of the experiment goals.
2. Imports: Group standard libraries, third-party, and usage-specific imports.
3. Configs: Define Global Constants (paths, random seeds, hyperparameters) here. No magic numbers deeper in the code.
4. Datasets: Load, validate, and split data.
5. Analysis (EDA): Inspect target distributions and correlations.
6. Modeling: Define and train sklearn.pipeline.Pipeline objects.
7. Evaluations: Compute metrics and visualize performance on held-out data.

2. Configuration Standards

Expose all "knobs" at the top of the notebook for easy experimentation.

• Randomness: Define RANDOM_STATE = 42 and use it in splits and model initialization.

• Paths: Use pathlib for robust path handling.

  from pathlib import Path
  
  ROOT = Path("..")
  DATA_PATH = ROOT / "data" / "input.parquet"
  

• Hyperparameters: Group model params (e.g., N_ESTIMATORS, MAX_DEPTH).

• Toggles: Use booleans for expensive operations (e.g., USE_GPU = True, RUN_GRID_SEARCH = False).

3. Data Management

Ensure data integrity and prevent leakage.

• Loading: Prefer pd.read_parquet for speed/types, or pd.read_csv.
• Splitting:
  • Always split into X_train, X_test, y_train, y_test before any data-dependent transformations (imputation, scaling).
  • Random Split: Use sklearn.model_selection.train_test_split with stratify for balanced classification.
  • Time Series: Use sklearn.model_selection.TimeSeriesSplit if data has a temporal dimension (do NOT shuffle).
  • Use random_state=RANDOM_STATE.