"""
Utilities for processing lists of parquet files
- Get number of rows
- Get number of unique values
"""

from tqdm import tqdm
from pyarrow.parquet import ParquetDataset
import pyarrow.dataset as ds
import pyarrow as pa
import os
import pyarrow.acero as ac
import pandas as pd

def get_num_rows(dataset):
    return sum(p.count_rows() for p in tqdm(dataset.fragments))

def get_num_uniques(dataset: ds.Dataset, columns: list) -> int:
    table = ac.Declaration.from_sequence([
        ac.Declaration('scan', ac.ScanNodeOptions(dataset)),
        ac.Declaration('aggregate', ac.AggregateNodeOptions([], columns)),
    ]).to_table()
    # return len(set(table.column(columns[0])))
    return pa.compute.count(table.column(columns[0]).combine_chunks().unique())

def listdir_fullpath(d):
    return [os.path.join(d, f) for f in os.listdir(d)]

def get_num_uniques_slow(path, column):
    column_set = set()
    for shard_path in tqdm(listdir_fullpath(path)):
        df = pd.read_parquet(listdir_fullpath(shard_path)[0], columns=[column])
        column_set = column_set.union(set(df[column]))
    return len(column_set)

def get_schema() -> pa.Schema:
    # this is the schema as in min-hash parquet files
    # (just the desired similarity hash list selected from it)
    return pa.schema([
        ("id", pa.string()), # this is (col name, col type)
        ("id_int", pa.uint64()),
        ("cluster_id", pa.uint64()),
        ("shard_id", pa.string())
    ])

if __name__ == "__main__":
    path = "/home1/BharatGPT_Data/RedPajamaV2/data/minhash_filtered/2023-14"

    # # specific files if needed
    # shards = listdir_fullpath(path)
    # file_paths = []
    # for shard in shards:
    #     for file_name in os.listdir(shard):
    #         if "head" in file_name and "4513" not in shard:
    #             file_paths.append(os.path.join(shard, file_name))
    # dataset = ParquetDataset(file_paths, use_legacy_dataset=False)
    # dataset = ds.dataset(source=file_paths, format="parquet") # not ParquetDataset for acero

    # else: all files
    dataset = ParquetDataset(path, use_legacy_dataset=False)

    print(get_num_rows(dataset))
    # print(get_num_uniques(dataset, ["cluster_id"]))
    # print(get_num_uniques_slow(path, "cluster_id"))