Application of KANs to quartet inference tasks

Prerequisites

For all scripts Python 3.11.9 was used. The python packages used can be installed via

conda env create -f required_packages.yml

if python3 and pip are already installed.

Network for distinguishing Farris and Felsenstein trees

The training and test data for a network distinguishing alignments simulated under Farris and Felsenstein trees is saved in the folder data/processed/zone.

If it is not available the training data can be generated via

./1_preprocess_zone_train_data.sh

and the test data via

./1_preprocess_zone_test_data.sh

in the folder data/preprocessing.

A train and test scripts for the network can be found within the scripts folder.

Running

python3 train.py <config>

a network is trained using the hyperparameters defined in the config-file (see e.g. config/config_KAN.yaml). The trained models are saved within the models folder.

An already trained network can be tested by executing:

python3 test.py -m <model>

The results will be saved in the results folder.