Epic sklearn — An expansion pack for scikit-learn

What is it?

The epic-sklearn Python library is a companion to the scikit-learn library for machine learning. It provides additional components and utilities, which can make working within the scikit-learn framework even more convenient and productive.

The main difference in base-assumptions between scikit-learn and epic-sklearn is that epic-sklearn has pandas as a dependency. Moreover, most epic-sklearn components support pandas objects (DataFrame and Series) and "pass along" as much information as possible. For example, in most transformers, if the features matrix is provided as a DataFrame, the transformed matrix will also be a DataFrame, and the index (and columns, if applicable) will be preserved. There are also a few components specifically designed for working only with pandas objects.

Content Highlights

• composite: Classifiers acting on other classifiers.
• feature_selection:
  • mutual_info: Calculation of conditional mutual information between a feature and the target given another feature, and feature selection algorithms based on conditional mutual information.
• metrics:
  • Metrics and scores for evaluating classification results and other data sets.
  • Also includes the leven module, allowing parallel computation of pairwise Levenshtein distances between python strings.
• neighbors: Utilities relevant for nearest neighbors algorithms.
• pipeline: Transformers for constructing transformation pipelines.
  • Contains a transformer that splits the samples based on a criterion, and applies different transformations on each sample group.
• plot: Plotting utilities.
• preprocessing:
  • categorical: Transformers for encoding and processing categorical data.
  • data: Transformers for binning and manipulating data distribution. Includes the Yeo–Johnson transformation.
  • general: General-purpose transformers (e.g. select DataFrame columns, apply a function in parallel, generate features from an iterator).
  • label: Utilities for encoding labels.
• utils:
  • data: Generate random batches from data.
  • validation: Functions for input validation and normalization.
  • kneedle: Implementation of the "Kneedle in a Haystack" algorithm.
  • thresholding: A helper for setting and applying a threshold based on classification metrics.

Contributors

Thanks to Yaron Cohen for his contribution to this project.