import pandas as pd
from sklearn.model_selection import train_test_split

train = pd.read_csv('https://services.healthtech.dtu.dk/services/DeepLoc-2.0/data/Swissprot_Train_Validation_dataset.csv')
train = train.loc[:,['Sequence','Cell membrane', 'Cytoplasm','Endoplasmic reticulum', 'Extracellular', 'Golgi apparatus', 'Lysosome/Vacuole', 'Mitochondrion', 'Nucleus', 'Peroxisome', 'Plastid']]
train = train.melt('Sequence', var_name='Location').query('value == 1.0').drop(labels='value', axis=1)

train, val = train_test_split(train)
val = val.reset_index(drop=True)

test = pd.read_csv('https://services.healthtech.dtu.dk/services/DeepLoc-2.0/data/hpa_testset.csv')
test = test.loc[:,['fasta','Cell membrane', 'Cytoplasm','Endoplasmic reticulum', 'Golgi apparatus', 'Lysosome/Vacuole', 'Mitochondrion', 'Nucleus', 'Peroxisome']].rename(columns={'fasta':'Sequence'})
test = test.melt('Sequence', var_name='Location').query('value == 1.0').drop(labels='value', axis=1).sample(frac=1).reset_index(drop=True)

train.to_parquet('data/deeploc/deeploc-train.parquet', index=False)
val.to_parquet('data/deeploc/deploc-val.parquet', index=False)
test.to_parquet('data/deeploc/deeploc-test.parquet', index=False)