ICTV Challenge
This project is submitted as part of the 2024 ICTV Computational Virus Taxonomy Challenge.
The results are in ./results
:
results/vanjari-0.1.csv: The results for the main Vanjari model.
results/vanjari-fast-0.1.csv: The results for the fast Vanjari model.
results/vanjari-ensemble-0.1.csv: The results for the fast Vanjari model.
There are also versions of the results with a threshold of 0.5:
To reproduce the results, use the following command to download the dataset:
wget "https://github.com/ICTV-VBEG/ICTV-TaxonomyChallenge/raw/refs/heads/main/dataset/dataset_challenge.tar.gz?download=" -O ictv-challenge.tar.gz
tar zxvf ictv-challenge.tar.gz
This will create a directory called dataset_challenge
with the sequences. Now run the following commands to classify the sequences using the two Vanjari models:
# Generage results for single models
vanjari --input dataset_challenge/ --output-csv ictv-challenge/vanjari-0.1.csv --memmap-array-path ictv-challenge/embeddings.npy --memmap-index ictv-challenge/embeddings.txt
vanjari-fast --input dataset_challenge/ --output-csv ictv-challenge/vanjari-fast-0.1.csv
# Generate results for ensemble
vanjari-tools ensemble-csvs --input ictv-challenge/vanjari-0.1.csv --input ictv-challenge/vanjari-fast-0.1.csv --output ictv-challenge/vanjari-ensemble-0.1.csv
# Set the threshold for the all results to 0.5
vanjari-tools increase-threshold --input ictv-challenge/vanjari-0.1.csv --output ictv-challenge/vanjari-0.1-threshold0.5.csv --threshold 0.5
vanjari-tools increase-threshold --input ictv-challenge/vanjari-fast-0.1.csv --output ictv-challenge/vanjari-fast-0.1-threshold0.5.csv --threshold 0.5
vanjari-tools increase-threshold --input ictv-challenge/vanjari-ensemble-0.1.csv --output ictv-challenge/vanjari-ensemble-0.1-threshold0.5.csv --threshold 0.5