ZeroTune provides instant zero-shot hyperparameter optimization using advanced pre-trained models. Get competitive hyperparameters for your machine learning models in sub-millisecond time with robust performance across diverse datasets!
🏆 Decision Tree: 100% win rate • 🌲 Random Forest: 100% win rate • 🔧 XGBoost: 100% win rate • 🚀 +7.08%, +1.47% & +0.80% improvements • ⚡ <1ms prediction • 📊 50-seed validated
from zerotune import ZeroTunePredictor
from sklearn.tree import DecisionTreeClassifier
import pandas as pd
# Load your dataset
df = pd.read_csv('your_dataset.csv')
X = df.drop('target', axis=1)
y = df['target']
# Get optimal hyperparameters instantly
predictor = ZeroTunePredictor(model_name='decision_tree', task_type='binary')
best_params = predictor.predict(X, y)
# Train model with predicted hyperparameters
model = DecisionTreeClassifier(**best_params)
model.fit(X, y)
print(f"Optimal hyperparameters: {best_params}")
# Expected: +7.08% improvement over random hyperparameters- 🏆 100% Win Rate: All three models (Decision Tree, Random Forest, XGBoost) beat random hyperparameters on every test dataset
- ⚡ Instant Predictions: Sub-millisecond hyperparameter optimization (vs hours of traditional HPO)
- 🎯 Significant Improvements: +7.08%, +1.47%, +0.80% average performance gains respectively
- 🔬 Scientifically Validated: 50-seed evaluation across diverse datasets with statistical rigor
- 🚀 Production Ready: Pre-trained models included - no training required
- 🔧 Optuna Integration: Warm-start TPE optimization with perfect baseline consistency
| Model | Binary Classification | Performance |
|---|---|---|
| 🏆 Decision Tree | ✅ | 100% win rate, +7.08% |
| 🌲 Random Forest | ✅ | 100% win rate, +1.47% |
| 🔧 XGBoost | ✅ | 100% win rate, +0.80% |
All models achieve 100% win rates - every single prediction outperforms random hyperparameter selection.
# Install Poetry (if not already installed)
curl -sSL https://install.python-poetry.org | python3 -
# Install ZeroTune
git clone https://github.com/your-repo/zerotune.git
cd zerotune
poetry install🚀 Ready-to-Use: All trained models are included - start predicting immediately!
from zerotune import ZeroTunePredictor
# For different models
predictor_dt = ZeroTunePredictor(model_name='decision_tree', task_type='binary')
predictor_rf = ZeroTunePredictor(model_name='random_forest', task_type='binary')
predictor_xgb = ZeroTunePredictor(model_name='xgboost', task_type='binary')
# Get instant predictions
hyperparams = predictor_dt.predict(X, y)from zerotune.core.optimization import optimize_hyperparameters
# Use zero-shot predictions to warm-start Optuna TPE
best_params, study = optimize_hyperparameters(
X=X, y=y,
model_type='decision_tree',
param_grid=param_grid,
n_trials=20,
warm_start=True, # Uses ZeroTune predictions
n_jobs=1
)# Quick evaluation on test datasets
poetry run python decision_tree_experiment.py eval-test
poetry run python random_forest_experiment.py eval-test
poetry run python xgb_experiment.py eval-test
# Full evaluation with Optuna benchmarking
poetry run python decision_tree_experiment.py eval-full --optuna_trials 25 --seeds 50┌─────────────────────┐ ┌──────────────────────┐ ┌─────────────────────┐
│ Knowledge Base │───▶│ Pre-trained Model │───▶│ Zero-Shot Predict │
│ Building │ │ Training │ │ (ZeroTunePredictor)│
│ (ZeroTune) │ │ │ │ │
│ │ │ │ │ │
│ • Multi-seed HPO on │ │ • RFECV feature │ │ • Sub-ms prediction │
│ many datasets │ │ selection (15/22) │ │ • 100% win rate │
│ • Extract 22+ meta- │ │ • Top-K filtering │ │ • Feature selection │
│ features │ │ • RandomForest +HPO │ │ • High performance │
│ • Store full trials │ │ • Meta-features → │ │ │
│ dataframes │ │ Hyperparameters │ │ │
└─────────────────────┘ └──────────────────────┘ └─────────────────────┘
│
▼
┌──────────────────────┐ ┌─────────────────────┐
│ Optuna TPE │◀───│ Your ML Pipeline │
│ Warm-Start │ │ │
│ │ │ │
│ • Zero-shot init │ │ • Train your model │
│ • Faster convergence │ │ • Better performance│
│ • study.enqueue() │ │ • Production deploy │
│ • Perfect baseline │ │ • Instant results │
└──────────────────────┘ └─────────────────────┘
- Knowledge Base: Multi-dataset HPO experiments with 22+ meta-features extracted
- Model Training: RFECV feature selection + RandomForest predictor with hyperparameter optimization
- Zero-Shot Prediction: Instant hyperparameter prediction based on dataset characteristics
- Optional Warm-Start: Use predictions to initialize Optuna TPE for further optimization
| Model | Win Rate | Avg Improvement | Best Single Win | Statistical Significance |
|---|---|---|---|---|
| Decision Tree | 100% | +7.08% | +17.4% | 90% of datasets |
| Random Forest | 100% | +1.47% | +4.4% | 50% of datasets |
| XGBoost | 100% | +0.80% | +2.6% | 90% of datasets |
Key Benefits:
- ✅ Perfect Reliability: 100% win rate across all models and test datasets
- ✅ Instant Results: Sub-millisecond prediction vs hours of traditional HPO
- ✅ Statistical Rigor: 50 random seeds × 10 datasets = 500 total experiments
- ✅ Production Ready: No training required, robust error handling
For detailed performance analysis, see PERFORMANCE_ANALYSIS.md
For researchers and advanced users:
# Generate publication-ready analysis and charts
poetry run python publication_analysis.py DecisionTree --auto-detect
poetry run python publication_analysis.py RandomForest --auto-detect
poetry run python publication_analysis.py XGBoost --auto-detectSee PUBLICATION_CHARTS_GUIDE.md for detailed documentation.
from zerotune import ZeroTune
# Build knowledge base from your datasets
zt = ZeroTune(model_type='xgboost', kb_path='my_knowledge_base.json')
dataset_ids = [31, 38, 44, 52, 151] # OpenML dataset IDs
kb = zt.build_knowledge_base(dataset_ids=dataset_ids, n_iter=20)from zerotune.core.predictor_training import train_predictor_from_knowledge_base
# Train predictor from knowledge base
model_path = train_predictor_from_knowledge_base(
kb_path='my_knowledge_base.json',
model_name='xgboost',
task_type='binary',
top_k_per_seed=3
)See CONTRIBUTING.md for development setup and contribution guidelines.
This project is licensed under the MIT License - see the LICENSE file for details.