FImdlp

Discretization algorithm based on the paper by Usama M. Fayyad and Keki B. Irani

Multi-Interval Discretization of Continuous-Valued Attributes for Classification Learning. In Proceedings of the 13th International Joint Conference on Artificial Intelligence (IJCAI-95), pages 1022-1027, Montreal, Canada, August 1995.

Installation

From PyPI

pip install FImdlp

From source (development)

The project no longer relies on a git submodule — the C++ sources live under src/cpp/. A regular clone is enough:

git clone https://github.com/doctorado-ml/FImdlp.git
cd FImdlp
make deps      # install the [dev] extras (build, twine, pip-audit, black, flake8, coverage)
make install   # editable install; compiles the Cython/C++ extension in place

Run make help to list every available target.

Quick start

from sklearn.datasets import load_iris
from fimdlp.mdlp import FImdlp

X, y = load_iris(return_X_y=True)
clf = FImdlp().fit(X, y)

# Discretize
X_disc = clf.transform(X)

# Inspect cut points: [vmin, c1, ..., cn, vmax] per feature
for f, cuts in enumerate(clf.get_cut_points()):
    print(f"feature {f}: {cuts}")

Constructor parameters:

Parameter	Default	Description
n_jobs	-1	Threads for per-feature fit/transform. -1 uses all cores.
min_length	3	Minimum samples in an interval to consider further splits.
max_depth	1e6	Maximum recursion depth of the splitting procedure.
max_cuts	0	Cap on intermediate cut points per feature (0 = unlimited; <1 is interpreted as a fraction of samples).

Make targets

Target	What it does
make help	List every target.
make deps	Install the [dev] extras (build, twine, pip-audit, black, flake8, coverage).
make install	Editable install (pip install -e .); rebuilds the C++/Cython extension.
make test	Run unit tests with coverage. Rebuilds the extension if the .so is missing.
make coverage	Run tests then print the coverage report.
make lint	Format with black and lint with flake8.
make build	Produce wheel + sdist in dist/.
make publish	make build + twine check + twine upload.
make audit	Run pip-audit on the installed packages.
make sample_py	Run the Python sample on the iris dataset.
make sample_cpp	Build and run the C++ sample on the iris dataset.
make version	Show Python, FImdlp and bundled mdlp versions.
make clean	Remove build artifacts, caches and the compiled extension.

Running the samples

Python sample

make sample_py
# equivalent to:
#   cd samples && python sample.py iris

Other options:

python samples/sample.py iris            # default settings
python samples/sample.py iris -c 2       # cap intermediate cut points to 2
python samples/sample.py iris -m 3       # cap recursion depth to 3
python samples/sample.py iris -n 25      # set min_length to 25
python samples/sample.py -h              # full option list

C++ sample

make sample_cpp
# equivalent to:
#   cd samples && cmake -B build -S . && cmake --build build && cd build && ./sample -f iris

Other options:

cd samples/build
./sample -f iris -c 2     # cap intermediate cut points to 2
./sample -f glass -m 3    # change dataset and depth
./sample -h               # full option list

Based on

https://github.com/rmontanana/mdlp