import polars as pl
import numpy as np
import re
from datasets import load_dataset, Dataset, concatenate_datasets

#region Preprocessing functions
def remove_newlines(text):
        return text.replace('\n', '')

def remove_urls(text):
    url_regex = r'http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+'
    return re.sub(url_regex, '', text)

def remove_html_tags(text):
    html_regex = r'<[^>]+>'
    return re.sub(html_regex, '', text)

def remove_special_characters(text):
    special_chars_regex = r'[^a-zA-Z0-9\s]'  
    return re.sub(special_chars_regex, '', text)

def remove_numbers(text):
    return re.sub(r'\d+', '', text)

def remove_extra_spaces(text):
    return re.sub(r'\s+', ' ', text)

def remove_twitter_mentions(text):
    return re.sub(r'@([A-Za-z0-9_]+)', '', text)

def remove_emoticons(text):
    emoticon_regex = r'[\U0001F600-\U0001F64F\U0001F300-\U0001F5FF\U0001F680-\U0001F6FF\U0001F700-\U0001F77F\U0001F780-\U0001F7FF\U0001F800-\U0001F8FF\U0001F900-\U0001F9FF\U0001FA00-\U0001FA6F\U0001FA70-\U0001FAFF\U00002702-\U000027B0\U000024C2-\U0001F251\U0001F004\U0001F0CF\U0001F170-\U0001F251\U0001F600-\U0001F64F\U00002702-\U000027B0\U000024C2-\U0001F251\U0001F300-\U0001F5FF\U0001F680-\U0001F6FF\U0001F700-\U0001F773\U0001F780-\U0001F7D8\U0001F7E0-\U0001F7EB\U0001F7F0-\U0001F7FF\U0001F800-\U0001F80B\U0001F90D-\U0001F9FF\U0001FA70-\U0001FA74\U0001F600-\U0001F64F\U0001F90D-\U0001F971\U0001F973-\U0001F978\U0001F97A-\U0001F9CB\U0001F9CD-\U0001F9FF]+'
    return re.sub(emoticon_regex, '', text)

def normalize_case(text):
    return text.lower()

def remove_unnecessary_spaces(text):
    text = text.strip()  
    text = re.sub(r'\s+', ' ', text)
    return text

def remove_punctuation_and_brackets(text):
    text = re.sub(r'[^\w\s\[\]]', '', text)
    return text

def remove_numbered_brackets(text):
    text = re.sub(r'\[\d+\]', '', text)
    return text

def remove_initial_article(text):
    words_to_remove = [
    'a', 'an', 'the', 'some', 'many', 'much', 'few', 'little', 
    'several', 'a few', 'a little', 'a lot of', 'lots of', 'plenty of', 
    'this', 'that', 'these', 'those', 'its' ]
    words = text.split()
    if words and words[0].lower() in words_to_remove:
        words.pop(0)
    return ' '.join(words)
    
def preprocess_text(text):
    text = remove_newlines(text)
    text = remove_punctuation_and_brackets(text)
    text = remove_special_characters(text)
    text = remove_urls(text)
    text = remove_html_tags(text)
    text = remove_numbers(text)
    text = remove_extra_spaces(text)
    text = remove_twitter_mentions(text)
    text = remove_emoticons(text)
    text = normalize_case(text)
    text = remove_unnecessary_spaces(text)
    text = remove_numbered_brackets(text)
    text=remove_initial_article(text)
    return text
#endregion

def softmax(x):
    r=np.exp(x - np.max(x))
    return r/r.sum(axis=0)

# Load the dataset
dataset1 = load_dataset("Fizzarolli/wattpad2", "default", split="train")

df2 = pl.read_parquet("wattpad_stories.parquet")
df2 = df2.to_pandas()
dataset2 = Dataset.from_pandas(df2)

dataset = concatenate_datasets([dataset1, dataset2]).shuffle(seed=42)

# Language detection
from lingua import LanguageDetectorBuilder
import unicodedata

detector = LanguageDetectorBuilder.from_all_languages().with_preloaded_language_models().build()

def add_language_column(example):
    res = detector.compute_language_confidence_values(unicodedata.normalize('NFKD', example["description"]))
    example["language"] = res[0].language.iso_code_639_3.name
    confidences = list(map(lambda x: x.value, res))
    example["language_confidence"] = np.max(confidences)
    return example

dataset = dataset.map(add_language_column)

print(dataset)
print(dataset[0])

# NSFW detection
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("eliasalbouzidi/distilbert-nsfw-text-classifier")
model = AutoModelForSequenceClassification.from_pretrained("eliasalbouzidi/distilbert-nsfw-text-classifier", device_map="cuda")

def add_nsfw_column(example):
    if "overall_nsfw_score" not in example:
        nsfw_scores = []
        for chapter_text in example["chapter_contents"]:
            preprocessed_text = preprocess_text(chapter_text)
            inputs = tokenizer(preprocessed_text, return_tensors="pt", padding=True, truncation=True, max_length=512).to("cuda")
            outputs = model(**inputs).logits
            probs = torch.softmax(outputs, dim=1).tolist()[0]
            nsfw_scores.append(probs[1])
        example["overall_nsfw_score"] = sum(nsfw_scores) / len(nsfw_scores)
        example["chapter_nsfw_scores"] = nsfw_scores
    return example

dataset = dataset.map(add_nsfw_column)

print(dataset)
print(dataset[0])

dataset.push_to_hub("Fizzarolli/wattpad2", "default", commit_message="Add more stories")