Implementation of the Convolutional Motif Kernel Network (CMKN). This method allows to build neural network models that incorporate learning within the reproducing kernel Hilbert space (RKHS) of the position-aware motif kernel into simple end-to-end learning. This approach results in artificial neural networks that can robustly learn on small datasets and are inherently interpretable.
cmkn/: CMKN's source codedocs/: CMKN's documentationscripts/: Scripts to perform experiments and analysisdata/: Preprocessing scripts for the datasets used in the experiments shown in the corresponding paper
You can perform a user-specific installation by running
$ python -m pip install .
from the root of the project. We strongly advise an installation in a virtual environment. You can create and activate one by executing the following two commands from the root of the project
$ python -m venv venv
$ . venv/bin/activate
If you are using anaconda, you can create a separate environment with the following commands
$ conda create -n venv python=3.9
$ conda activate venv
and then performing the installation as usual by running
(venv) $ python -m pip install .
If you plan to extend the code, then you should perform an editable installation with
(venv) $ python -m pip install -e .
You can run the unit-tests by executing
$ python -m unittest
from the root of the project. The ground truth needed for the tests is stored in the folder data/gound_truth/.
The documentation is written with sphinx. You can build it by running
$ cd docs && make html
from the root of the project. The entry point for the documentation will be placed in doc/_build/html/index.html which you can open with a browser of your choice.