make-data.sh

#!/usr/bin/env bash

set -xeuo pipefail

python -m venv .venv \
&& .venv/bin/pip install -r requirements.txt \
&& source .venv/bin/activate

FILENAME=csd_smiles.csv
COLUMN=huckel_smiles_CSD_xyz
outdir=$(readlink -f $(dirname $0))
full_input="$outdir/$FILENAME"
converted="$outdir/$(basename $FILENAME .csv)-split.csv"

curl https://raw.githubusercontent.com/jensengroup/xyz2mol_tm/refs/heads/main/SMILES_csvs/csd_smiles.csv > "$full_input"

schemist convert "$full_input" \
    --format csv \
    --column $COLUMN \
    --to smiles scaffold mwt clogp tpsa \
2> "$outdir/convert.log" \
| schemist split \
    --format csv \
    --column smiles \
    --type scaffold \
    --train 0.85 --test 0.15 \
    --seed 1 \
2> "$outdir/split.log" \
> "$converted"

for split_name in train test
do  
    split="$outdir/$(basename $converted .csv)_$split_name.csv"
    filter_field=$(head -n1 "$converted" | tr , $'\n' | grep -nwF "is_$split_name" | cut -d: -f1)
    
    head -n1 "$converted" \
    | cat - <(tail -n+2 "$converted" | awk -F, -v OFS=, -v col="$filter_field" '$col == "True"') \
    > "$split"
done

for f in *.csv
do 
    gzip --best "$f"
done