import os
from apscheduler.schedulers.background import BackgroundScheduler
import gradio as gr
import pandas as pd

from src.control_panel import create_control_panel, create_control_callback
from src.leaderboard import (
    create_leaderboard_table,
     COLS,
    TYPES,
)

from src.llm_perf import get_llm_perf_df

from src.content import (
    LOGO,
    TITLE,
    ABOUT,
    INTRODUCTION,
    EXAMPLE_CONFIG,
    CITATION_BUTTON,
    CITATION_BUTTON_LABEL,
)


MACHINE_TO_HARDWARE = {"hf-dgx-01": "A100-80GB-275W 🖥️", "audace": "RTX4090-24GB-450W 💻"}
HF_TOKEN = os.environ.get("HF_TOKEN", None)

def restart_space():
    API.restart_space(repo_id=REPO_ID, token=HF_TOKEN)

import unicodedata

def is_valid_unicode(char):
    try:
        unicodedata.name(char)
        return True  # Valid Unicode character
    except ValueError:
        return False  # Invalid Unicode character

def remove_invalid_unicode(input_string):
    if isinstance(input_string, str):
        valid_chars = [char for char in input_string if is_valid_unicode(char)]
        return ''.join(valid_chars)
    else:
        return input_string  # Return non-string values as is

def preprocess_dataframe(df):
    # Apply the `remove_invalid_unicode` function to all string columns
    for column in df.columns:
        if df[column].dtype == 'object':  # Checking for string columns
            df[column] = df[column].apply(remove_invalid_unicode)
    return df

# Fetch and preprocess the leaderboard DataFrame
llm_perf_df = get_llm_perf_df()
llm_perf_df = preprocess_dataframe(llm_perf_df)

# Create the leaderboard table
leaderboard_table = create_leaderboard_table(llm_perf_df)

hidden_leaderboard_table_for_search = gr.components.Dataframe(
    leaderboard_table,
    headers=COLS,
    datatype=TYPES,
    visible=False,
    line_breaks=False,
    interactive=False
)

def display(x, y):
    # Assuming df is your DataFrame
    for column in leaderboard_table.columns:
        if leaderboard_table[column].dtype == 'object':
            leaderboard_table[column] = leaderboard_table[column].apply(remove_invalid_unicode)

    subset_df = leaderboard_table[COLS]
    return subset_df

dummy1 = gr.Textbox(visible=False)

INTRODUCTION_TEXT = """
This is a copied space from LLM Trustworthy Leaderboard. Instead of displaying
the results as table this space was modified to simply provides a gradio API interface. 
Using the following python script below, users can access the full leaderboard data easily.
Python on how to access the data:
```python
# Import dependencies
from gradio_client import Client
# Initialize the Gradio client with the API URL
client = Client("https://rodrigomasini-data-only-llm-perf-leaderboard.hf.space/")
try:
    # Perform the API call
    response = client.predict("","", api_name='/predict')
    # Check if response it's directly accessible
    if len(response) > 0:
        print("Response received!")
        headers = response.get('headers', [])
        data = response.get('data', [])
        print(headers)
        # Remove commenst if you want to download the dataset and save in csv format
        # Specify the path to your CSV file
        #csv_file_path = 'llm-perf-benchmark.csv'
        # Open the CSV file for writing
        #with open(csv_file_path, mode='w', newline='', encoding='utf-8') as file:
        #    writer = csv.writer(file)
            # Write the headers
        #    writer.writerow(headers)
            # Write the data
        #    for row in data:
        #        writer.writerow(row)
        #print(f"Results saved to {csv_file_path}")
    # If the above line prints a string that looks like JSON, you can parse it with json.loads(response)
    # Otherwise, you might need to adjust based on the actual structure of `response`
except Exception as e:
    print(f"An error occurred: {e}")
```
"""

interface = gr.Interface(
    fn=display,
    inputs=[gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text"), dummy1],
    outputs=[hidden_leaderboard_table_for_search]
)

scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()

interface.launch()