import os
import pandas as pd
import jsgf
from unidecode import unidecode
import string


def decode_prompt(prompt):
    prompt_decoded = unidecode(prompt)
    translator = str.maketrans('', '', string.punctuation)
    prompt_decoded = prompt_decoded.translate(translator)
    return prompt_decoded

grammar = jsgf.parse_grammar_file('book.jsgf')
data_files = []

for filename in os.listdir("data"):
    f = os.path.join("data", filename)
    if os.path.isfile(f):
        data_files.append(pd.read_csv(f, sep='\t', header=None))

recognized = 0
unrecognized = 0

for df in data_files:
    if len(df.columns) == 3:
        df.columns = ["agent", "message", "act"]
    elif len(df.columns) == 2:
        df.columns = ["agent", "message"]
    else:
        continue
    user_speech_rows = df[df['agent'] == "user"]
    user_speeches = user_speech_rows["message"]
    entries_count = len(user_speeches)


    parsed = user_speeches.apply(
        lambda x: bool(grammar.find_matching_rules(decode_prompt(x))))
    true_count = parsed.sum()
    false_count = len(parsed) - true_count
    recognized += true_count
    unrecognized += false_count

print(f"Recognized user utterances: {recognized}")
print(f"Unrecognized user utterances: {unrecognized}")
print(f"Accuracy: {recognized/(recognized+unrecognized)}")