This is a set of tools for machine learning. Provided by the package utilities are described in the below table:
Subject | Description | Docs |
---|---|---|
Active Learning | Highly-modular system that recommends which previously unlabelled examples should be labelled in order to increase model quality quickly and significantly. Special features: various options for both exploitation and exploration. | Read more |
Stacking | A method that applies machine learning algorithm to out-of-fold predictions or transformations made by other machine learning models. Special features: support of any sklearn -compatible estimators (in particular, pipelines). |
Read more |
Target Encoding | An alternative to one-hot encoding and hashing trick that attempts to have both memory efficiency and incorporation of all useful information from initial features. Special features: sklearn -compatible wrapper that can transform data out-of-fold and apply an estimator to the result. |
Read more |
Repository name is a combination of three words: DS, saw, and awl. DS is as an abbreviation for Data Science and the latter two words represent useful tools.
The package is compatible with Python 3.5 or newer. A virtual environment where it is guaranteed that the package works can be created based on the file named requirements.txt
.
To install a stable release of the package, run this command:
pip install dsawl
To install the latest version from sources, execute this from your terminal:
cd path/to/your/destination
git clone https://github.com/Nikolay-Lysenko/dsawl
cd dsawl
pip install -e .
If you have any troubles with installation, your questions are welcome.