nikolay-lysenko/dsawl

A set of tools for machine learning (for the current day, there are active learning utilities and implementations of some stacking-based techniques).

dsawl

What is it?

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.

How to install the package?

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.