#!/usr/local/bin/python

from pathlib import Path
import polars as pl
from sklearn.model_selection import train_test_split, GroupShuffleSplit
import argparse

parser = argparse.ArgumentParser(description="Create dataset")
parser.add_argument("dataset", type=Path, help="dataset file name")
args = parser.parse_args()
filepath = args.dataset

data = pl.read_csv(
    filepath,
    use_pyarrow=True,
    columns=["INGRESO", "FECHANACIMIENTO", "SOSPECHA_DIAGNOSTICA", "GES"],
)

data = data.with_columns(
    age=((data["INGRESO"] - data["FECHANACIMIENTO"]).dt.days() / 365.25).cast(pl.Int16)
)
data = data.filter(pl.col("age") > 0)
data = data.with_columns(
    GES=data["GES"].cast(pl.Boolean).cast(str),
)
data = data.drop(["INGRESO", "FECHANACIMIENTO"])
data = data.with_columns(
    pl.struct(["age", "SOSPECHA_DIAGNOSTICA"])
    .apply(lambda x: f"Paciente de {x['age']} años. {x['SOSPECHA_DIAGNOSTICA']}")
    .alias("text")
).drop(["age", "SOSPECHA_DIAGNOSTICA"])
data = data.rename({"GES": "label"})

train_interim_splitter = GroupShuffleSplit(n_splits=1, train_size=0.7, random_state=11)
train_indices, interim_indices = next(
    train_interim_splitter.split(data["text"], data["label"], data["text"])
)
train = data[train_indices]
interim = data[interim_indices]

test_validation_splitter = GroupShuffleSplit(
    n_splits=1, train_size=2 / 3, random_state=11
)
test_indices, validation_indices = next(
    test_validation_splitter.split(interim["text"], interim["label"], interim["text"])
)
test = interim[test_indices]
validation = interim[validation_indices]

train.write_parquet("train.parquet")
test.write_parquet("test.parquet")
validation.write_parquet("validation.parquet")