import argparse
import gzip
import json

from sklearn.model_selection import GroupShuffleSplit


def read_groups():
    groups = [set()]
    for line in open("schema-groups.txt"):
        if line.strip() != "":
            groups[-1].add(line.strip())
        else:
            groups.append(set())

    return groups

def sample(random_state, train_pct):
    groups = read_groups()
    next_id = len(groups)
    names = []
    name_ids = []
    for line in gzip.open("all.jsonl.gz", "rt"):
        obj = json.loads(line)
        names.append(obj["name"])
        found = False
        for (i, group) in enumerate(groups):
            if obj["name"] in group:
                assert(not found)
                found = True
                name_ids.append(i)
        if not found:
            name_ids.append(next_id)
            next_id += 1

    gss = GroupShuffleSplit(n_splits=10, train_size=train_pct, random_state=random_state)
    train_idx, test_idx = next(gss.split(names, groups=name_ids))

    train_file = gzip.open("train.jsonl.gz", "wt")
    val_file = gzip.open("validation.jsonl.gz", "wt")
    for (idx, line) in enumerate(gzip.open("all.jsonl.gz", "rt")):
        if idx in train_idx:
            train_file.write(line)
        elif idx in test_idx:
            val_file.write(line)

    train_file.close()
    val_file.close()

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--train_pct", type=float, default=0.8)
    parser.add_argument("--random_state", type=int, default=16)
    args = parser.parse_args()
    sample(args.random_state, args.train_pct)