from typing import List as PI_LIST, Tuple as PI_TUPLE

import polars as PI_POLARS
import numpy as PI_NUMPY

from sklearn.feature_extraction.text import TfidfVectorizer as PI_TFIDFVECTORIZER
from sklearn.preprocessing import normalize as PI_NORMALIZE


def FC_BUILD_TFIDF_AND_COSINE_BEST_MATCH(
    PI_LI_CATEGORIES: PI_LIST[str],
    PI_LI_TEXTS: PI_LIST[str],
    PI_ST_NGRAM: str = '1,2',
    PI_I_MIN_DF: int = 1
) -> PI_TUPLE[PI_NUMPY.ndarray, PI_NUMPY.ndarray]:

    if len(PI_LI_CATEGORIES) == 0:
        raise ValueError('PI_LI_CATEGORIES is empty')

    if len(PI_LI_TEXTS) == 0:
        return PI_NUMPY.array([]), PI_NUMPY.array([])

    ZV_LI_NGRAM = PI_ST_NGRAM.split(',')
    ZV_I_NGRAM_MIN = int(ZV_LI_NGRAM[0])
    ZV_I_NGRAM_MAX = int(ZV_LI_NGRAM[1])

    ZV_LI_CORPUS = []
    ZV_LI_CORPUS.extend(PI_LI_CATEGORIES)
    ZV_LI_CORPUS.extend(PI_LI_TEXTS)

    ZV_OB_VECT = PI_TFIDFVECTORIZER(
        ngram_range=(
            ZV_I_NGRAM_MIN,
            ZV_I_NGRAM_MAX
        ),
        min_df=PI_I_MIN_DF
    )

    ZV_MX_TFIDF_ALL = ZV_OB_VECT.fit_transform(ZV_LI_CORPUS)

    ZV_I_CATEGORIES_N = len(PI_LI_CATEGORIES)

    ZV_MX_CATEGORIES = ZV_MX_TFIDF_ALL[
        0:ZV_I_CATEGORIES_N,
        :
    ]

    ZV_MX_TEXTS = ZV_MX_TFIDF_ALL[
        ZV_I_CATEGORIES_N:,
        :
    ]

    ZV_MX_CATEGORIES_N = PI_NORMALIZE(
        ZV_MX_CATEGORIES,
        norm='l2',
        axis=1
    )

    ZV_MX_TEXTS_N = PI_NORMALIZE(
        ZV_MX_TEXTS,
        norm='l2',
        axis=1
    )

    ZV_MX_COS = (
        ZV_MX_TEXTS_N
        @
        ZV_MX_CATEGORIES_N.T
    ).toarray()

    ZV_AR_BEST_IDX = ZV_MX_COS.argmax(axis=1)
    ZV_AR_BEST_COS = ZV_MX_COS.max(axis=1)

    return ZV_AR_BEST_COS, ZV_AR_BEST_IDX