Synopsis
Python module (using Rust) to transform a binary observation-feature1 matrix into a binary observation-feature2 matrix.
This is done by greedily applying a binary feature2-feature1 matrix from top to bottom.
Algorithm
Inputs:
-
an OxM binary input feature matrix. row = obseration, column = input feature.
-
an NxM binary "translation" or "mapping" matrix; row = output feature. column = input feature.
Output:
- an OxN binary output feature matrix; row = output feature. column = output feature.
Two procedures and a helper class are made available:
from feature_mapper import map_features, map_features_smin
from feature_mapper import FeatureMapper
The translation/mapping is performed, as follows (both variants). Here the activity diagrams:
The variant map_features_smin ensures, that all output features have
at least smin observations.
FeatureMapper helps by providing a wrapper that can match column
names of the input observation matrix and the mapping matrix. It also
computes column names the mapped output features by combinding
corresponding input feature names with underscore. Usage example:
from feature_mapper import FeatureMapper
fm = FeatureMapper().fit(mapping_matrix, mapping_matrix_column_names)
obs_out_features = fm.transform(obs_in_features, in_feature_names)
out_feature_names = fm.output_names
All matrices are internally processed as scipy.sparse.csr_matrix.
Install
First, you need to install Rust, e.g. using wget:
wget -O - https://sh.rustup.rs | sh -s
or see https://rustup.rs/
via pip
pip install feature-mapper
via poetry
poetry add feature-mapper
via repo-clone (using pyenv)
git clone https://github.com/complexity-science-hub/feature-mapper.git
cd feature-mapper
A fresh installation of Python can (and probably should) be obtained via
- installing
pyenv: https://github.com/pyenv/pyenv
Then (using fish):
pyenv install (cat .python-version)
pip install -U pip
pip install maturin
pip install tox
Build and run tests (takes a few minutes):
tox
If all is well, install:
pip install . -v
