Feature extraction, processing and interpretation algorithms and functions for machine learning and data science.
Project description
feature_stuff: a python machine learning library for advanced feature extraction, processing and interpretation.
Latest Release | see on pypi.org |
Package Status | see on pypi.org |
License | see on github |
Build Status | see on travis |
What is it
feature_stuff is a Python package providing fast and flexible algorithms and functions for extracting, processing and interpreting features:
Numeric feature extraction
add_interactions | generic function for adding interaction features to a data frame either by passing them as a list or by passing a boosted trees model to extract the interactions from. |
target_encoding | target encoding of a feature column using exponential prior smoothing or mean prior smoothing |
cv_target_encoding | target encoding of a feature column taking cross-validation folds as input |
add_knn_values | creates a new feature with the K-nearest-neighbours of the values of a given feature |
add_group_values | generic and memory efficient enrichment of features dataframe with group values |
Model feature insights extraction
get_xgboost_interactions | takes a trained xgboost model and returns a list of interactions between features, to the order of maximum depth of all trees. |
Installation
Binary installers for the latest released version are available at the Python package index .
# or PyPI
pip install feature_stuff
The source code is currently hosted on GitHub at: https://github.com/hiflyin/Feature-Stuff
Installation from sources
In the Feature-Stuff
directory (same one where you found this file after
cloning the git repo), execute:
python setup.py install
or for installing in development mode:
python setup.py develop
Alternatively, you can use pip
if you want all the dependencies pulled
in automatically (the -e
option is for installing it in development
mode):
pip install -e .
How to use it
< see the attached API of each function/ algorithm >
Example on extracting interactions form tree based models and adding them as new features to your dataset.
import feature_stuff as fs
import pandas as pd
import xgboost as xgb
data = pd.DataFrame({"x0":[0,1,0,1], "x1":range(4), "x2":[1,0,1,0]})
print data
x0 x1 x2
0 0 0 1
1 1 1 0
2 0 2 1
3 1 3 0
target = data.x0 * data.x1 + data.x2*data.x1
print target.tolist()
[0, 1, 2, 3]
model = xgb.train({'max_depth': 4, "seed": 123}, xgb.DMatrix(data, label=target), num_boost_round=2)
fs.addInteractions(data, model)
# at least one of the interactions in target must have been discovered by xgboost
print data
x0 x1 x2 inter_0
0 0 0 1 0
1 1 1 0 1
2 0 2 1 0
3 1 3 0 3
# if we want to inspect the interactions extracted
from feature_stuff import model_features_insights_extractions as insights
print insights.get_xgboost_interactions(model)
[['x0', 'x1']]
Contributing to feature-stuff
All contributions, bug reports, bug fixes, documentation improvements, enhancements and ideas are welcome.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for feature_stuff-0.0.dev5-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | def5309ed0907cf8aff09a380362baedef313fafb9f77141c394daf16cf22c4f |
|
MD5 | 379474c0a948a0534d6bb055d41de22a |
|
BLAKE2b-256 | ef27f01d3c33e4046c2b513b38cf38746a78f7a5391df90df99182c0a7161053 |