ASF is a powerful library for algorithm selection and performance prediction. It allows users to easily create and use algorithm selectors with minimal code.
NOTE: ASF is still under construction (early alpha). Therefore, not only the API can change, but there might be some bugs in the implementations of the selectors. For the common methods (multi class classification, pairwise regression / classification as well as simple ranking) we checked the performance and the implementation and they can be safely used. We will release in the near future a benchmark of all methods on ASlib scenarios which will validate the performance.
- Easy-to-use API for creating algorithm selectors
- Supports various selection models including pairwise classifiers, multi-class classifiers, and performance models
- Integration with popular machine learning libraries like scikit-learn
You can create an algorithm selector with just 2 lines of code. Here is an example using the PairwiseClassifier:
from asf.selectors import PairwiseClassifier
from sklearn.ensemble import RandomForestClassifier
# Create a PairwiseClassifier
selector = PairwiseClassifier(model_class=RandomForestClassifier, metadata=your_metadata)
# Fit the selector with feature and performance data
selector.fit(dummy_features, dummy_performance)
# Predict the best algorithm for new instances
predictions = selector.predict(new_features)In the future, ASF will include more features such as:
- Empirical performance prediction
- Feature selection
- Support for ASlib scenarios
- And more!
To install ASF, use pip:
pip install asf-libFor detailed documentation and examples, please refer to the official documentation.
We welcome contributions! Please see our contributing guidelines for more details.
ASF is licensed under the MIT License. See the LICENSE file for more details.