import pandas as pd
import re



def load_pdf_data(file_path):
    data = pd.read_csv(file_path, encoding="ISO-8859-1")
    return data

def clean_text(text):
    # Function to clean text
    text = text.lower()
    text = re.sub(r'\W+', ' ', text)  # replace all non-alphanumeric characters with a space
    text = re.sub(r'\d+', '', text)  # remove all digits
    text = text.strip()
    return text

def preprocess_data(data):
    data['utterance'] = data['utterance'].apply(clean_text)
    return data

def save_filter_intents(data, file_path):
    # Get the 50 most balanced intents
    filtered_df = data.groupby('intent', group_keys=False).apply(lambda x: x.sample(min(len(x), 15)))
    # Save the filtered DataFrame as a CSV file
    filtered_df.to_csv(file_path+'\Pager_filtered_Intents.csv', index=False)

def save_cleaned_data(data, file_path):
    # Save the cleaned DataFrame as a CSV file
    data.to_csv(file_path+'\Pager_Intents_cleaned.csv', index=False)

# data = load_data(r'C:\Users\serban.tica\Documents\tobi_llm_intent_recognition\data\Pager_Intents.csv')
# cleaned_data = preprocess_data(data)
# save_cleaned_data(data, r'C:\Users\serban.tica\Documents\tobi_llm_intent_recognition\data')
# save_filter_intents(cleaned_data, r'C:\Users\serban.tica\Documents\tobi_llm_intent_recognition\data')

# split train test
def split_train_test(data, test_size=0.3):
    # Split the data into training and testing sets
    train_data = data.sample(frac=1-test_size, random_state=42)
    test_data = data.drop(train_data.index)
    print(f"Train data shape: {train_data.shape}")

    return train_data, test_data

def save_train_test_data(train_data, test_data, file_path):
    # Save the training and testing data to CSV files
    train_data.to_csv(file_path+'\\train_data.csv', index=False)
    test_data.to_csv(file_path+'\\test_data.csv', index=False)


# train_data, test_data = split_train_test(cleaned_data)
# save_train_test_data(train_data, test_data, 'C:\\Users\\serban.tica\\Documents\\tobi_llm_intent_recognition\\data')