🏠 Welcome to GridMaster¶

Welcome to GridMaster — an advanced Python toolkit I built to automate hyperparameter tuning and model selection across multiple classifiers.

With just a few lines, GridMaster helps you:

✅ Automatically optimize key classifiers over your dataset
✅ Narrow down from broad, industry-recommended parameter grids
✅ Fine-tune around best parameter ranges using smart linear or logarithmic scaling
✅ Run multiple models in parallel, automatically selecting the top performer — no manual, repetitive grid search loops needed
✅ Fully compatible with GridSearchCV workflows — migrate easily without added learning costs, including advanced settings like GPU acceleration
✅ Balance system load and performance by default, using half of available CPU cores for parallel search — adjustable for advanced users (see Advanced Setting – CPU Parallelism (n_jobs))

⚙️ New in v0.5.x

Smart mode selector: Switch between 'fast' (default) for lightweight search, or 'industrial' for expanded grids and deeper optimization.

Solver auto-selection for Logistic Regression: Automatically picks 'liblinear' or 'saga' depending on dataset size and mode, enabling elasticnet where applicable.

Expanded coarse grids under 'industrial' mode for XGBoost, LightGBM, CatBoost, Random Forest.

⚙️ New in v0.3.x

Smart, Expert, and Custom fine-tuning modes:

Smart: Automatically selects top 2 impactful hyperparameters based on variation analysis.

Expert: Focuses on commonly sensitive parameters like learning_rate and max_depth.

Custom: User-defined fine grids.

Parallel CPU support: Automatically detects available CPU cores and assigns half for faster parallel search (n_jobs), balancing speed and system performance. You can override this manually if needed.

Enable GPU Acceleration: Directly pass GPU-specific flags (like tree_method='gpu_hist') to XGBoost, LightGBM, and CatBoost via custom_estimator_params.

🚀 Supported Models¶

GridMaster currently supports classification models only:

✅ Logistic Regression
✅ Random Forest
✅ XGBoost
✅ LightGBM
✅ CatBoost

⚠️ Note: Decision Trees are currently excluded, as they are rarely used in industrial hyperparameter optimization workflows.

🏗️ GridMaster is built on top of scikit-learn and integrates seamlessly with popular libraries like XGBoost, LightGBM, and CatBoost, providing a familiar interface for model tuning and evaluation.

🔍 How It Works¶

Coarse Search
Starts with broad, commonly recommended parameter grids for each classifier (e.g., C, max_depth, learning_rate), providing a robust initial exploration of the search space.

Mode Selection:

GridMaster now supports two modes:

- 'fast': Focused, minimal grids for fast iteration.

- 'industrial': Wide grids with additional hyperparameters, better suited for production-scale optimization.
Fine Search
Automatically refines parameter ranges around the best coarse result:
For linear-scale parameters, narrows range by ±X% (default ±50%)
For log-scale parameters (like C, learning_rate), adjusts intelligently on the log scale ensuring meaningful search coverage without wasting runs.
Multi-Stage Search
Allows multiple fine-tuning rounds with custom precision, eliminating the need to manually loop grid searches for each model.

By default, both Fine Search and Multi-Stage Search focus on the top 2 most impactful parameters (based on performance variation observed during the coarse search stage) for refinement.
Multi-Model Comparison
Trains and tunes all supported models in parallel, automatically identifies the top performer, and outputs detailed metrics and plots for interpretation.

✨ Why I Built GridMaster¶

I designed GridMaster to free myself (and others) from the repetitive burden of per-model grid search, offering a clean, unified, and automated workflow for classification tasks.

It encapsulates the entire ML pipeline — from preprocessing and feature selection to model training and evaluation — ensuring reproducibility, modularity, and smooth end-to-end optimization.

“Laziness fuels productivity.”

🛠️ Get Started¶

Check the Quickstart Guide
Run your first multi-model search pipeline
Visualize and compare model performances
Dive into Essential Tools and Advanced Utilities

📇 About the Author¶

Hi, I’m Winston Wang — a data scientist passionate about making a meaningful contribution to the world, one data-driven solution at a time.

For feedback or suggestions, feel free to email me at:
📧 mail@winston-wang.com

For more about me, please visit my personal website.