Data, code and models for: Simple linear classifiers outperform deep models for SARS-CoV-2 lineage assignment from k-mer profiles

Companion deposit for the article "Simple linear classifiers outperform deep models for SARS-CoV-2 lineage assignment from k-mer profiles" (Jamhuri, Irawan, Alvi, Safruddin, Rozi; submitted to Computational Biology and Chemistry). The archive contains 3-, 4- and 5-mer count matrices for 256,225 NCBI Virus genomes, accessions and lineage labels, the group-aware partition, a 36-line NumPy implementation of the single-layer softmax classifier with DGN-EF, experiment and figure scripts, trained checkpoints of the SLP and MLP (ten seeds each), and result files. Training code for the fcgr4-CNN and sae4-MLP baselines is not included; their recorded scores are in results/hasil_gabungan.json. See README.md for the layout and commands. Code: MIT License. Data derived from NCBI Virus.