app.py

from shiny import App, render, reactive, ui
import pandas as pd
import numpy as np
import random
import io
import csv
import urllib3
from io import StringIO
import os
# HTTP & API
import requests
import urllib3
from huggingface_hub import login, hf_hub_download

# Constants
CSV_FILE = "hf://datasets/TJStatsApps/scouting_reports/elo_rankings.csv"
CSV_FILE_NAME = "elo_rankings.csv"
HF_REPO_ID = "TJStatsApps/scouting_reports"
HF_FILENAME = "elo_rankings.csv"

PITCH_TYPES = ["Hit", "Decisions", "Power", "Speed", "Defense"]
EVAL_CATEGORIES = ["ETA", "FV"]

# Hugging Face login using access token from environment variable
ACCESS_TOKEN = os.environ.get("ACCESS_TOKEN")
if ACCESS_TOKEN:
    login(token=ACCESS_TOKEN)


ADMIN_PW = os.getenv("ADMIN_PW")
VALID_PASSWORDS = []
VALID_PASSWORDS.append(ADMIN_PW)

# Initial ELO rating for new players
INITIAL_ELO = 1200
K_FACTOR = 32  # ELO adjustment factor

# Pre-defined list of 20 players (fallback if CSV can't be loaded)
df_player_names = pd.read_csv('top_100_prospects_canonical_names.csv')

DEFAULT_PLAYERS = df_player_names['canonical_name'].to_list()

def load_rankings_from_hf():
    """Load ELO rankings from Hugging Face dataset"""
    try:
        http = urllib3.PoolManager()
        response = http.request(
            'GET',
            "https://huggingface.co/datasets/TJStatsApps/scouting_reports/resolve/main/elo_rankings.csv"
        )
        
        if response.status == 200:
            csv_data = StringIO(response.data.decode('utf-8'))
            df = pd.read_csv(
                csv_data,
                quoting=csv.QUOTE_ALL,
                escapechar='\\',
                on_bad_lines='warn'
            )
            
            # Convert back to internal format (Player -> name, etc.)
            df = df.rename(columns={
                "Player": "name",
                "ELO": "elo",
                "Matches": "matches",
                "Wins": "wins"
            })
            
            # Remove rank and win% columns if they exist
            df = df[["name", "elo", "matches", "wins"]]
            
            return df
        else:
            print(f"Failed to load rankings: HTTP {response.status}")
            return None
            
    except Exception as e:
        print(f"Error loading rankings from HF: {e}")
        return None


# Define the login UI
login_ui = ui.page_fluid(
    ui.card(
        ui.input_password("password", "Enter Password:", width="50%"),
        ui.tags.input(
            type="checkbox",
            id="authenticated",
            value=False,
            disabled=True
        ),
        ui.input_action_button("login", "Login", class_="btn-primary"),
        ui.output_text("login_message"),
    )
)

main_ui = ui.page_sidebar(
    ui.sidebar(
        ui.h3("Current Matchup"),
        ui.output_ui("current_matchup"),
        ui.br(),
        ui.input_action_button("new_matchup", "Generate New Matchup", class_="btn-primary"),
        ui.br(),
        ui.br(),
        ui.input_action_button("reset_all", "Reset All Data", class_="btn-danger"),
        width='600px'
    ),
    ui.div(
        ui.h2("ELO Rankings"),
        ui.div(
            ui.input_action_button("update_rankings", "Update Rankings", class_="btn-info", style="margin-right: 10px;"),
            ui.input_action_button("upload_rankings", "Upload to HF", class_="btn-warning"),
            style="float: right; margin-bottom: 10px;"
        ),
        style="overflow: hidden;"
    ),
    ui.card(
        ui.output_data_frame("rankings_table")
    ),
    ui.h2("Match History"),
    ui.card(
        ui.output_data_frame("match_history")
    )
)


# Combined UI with conditional panel
app_ui = ui.page_fluid(
    ui.tags.head(
        ui.tags.script(src="script.js")
    ),

    ui.panel_conditional(
        "!input.authenticated",
        login_ui
    ),
    ui.panel_conditional(
        "input.authenticated",
        main_ui
    )
)


def server(input, output, session):
    # Try to load rankings from Hugging Face, fallback to default players

    @reactive.Effect
    @reactive.event(input.login)
    def check_password():
        if input.password() in VALID_PASSWORDS:
            ui.update_checkbox("authenticated", value=True)
            ui.update_text("login_message", value="")
        else:
            ui.update_text("login_message", value="Invalid password!")
            ui.update_text("password", value="")

    @output
    @render.text
    def login_message():
        return ""    
    
    loaded_df = load_rankings_from_hf()
    if loaded_df is not None:
        initial_df = loaded_df
        ui.notification_show("Rankings loaded from Hugging Face!", type="message")
    else:
        initial_df = pd.DataFrame({
            "name": DEFAULT_PLAYERS,
            "elo": [INITIAL_ELO] * len(DEFAULT_PLAYERS),
            "matches": [0] * len(DEFAULT_PLAYERS),
            "wins": [0] * len(DEFAULT_PLAYERS)
        })
        ui.notification_show("Using default players (couldn't load from HF)", type="warning")
    
    # Reactive values to store data
    players_data = reactive.value(initial_df)
    matches_data = reactive.value(pd.DataFrame(columns=["player1", "player2", "winner", "player1_elo_before", "player2_elo_before", "player1_elo_after", "player2_elo_after"]))
    current_player1 = reactive.value("")
    current_player2 = reactive.value("")
    rankings_update_trigger = reactive.value(0)  # Trigger for manual rankings update
    
    def calculate_elo_change(rating_a, rating_b, score_a):
        """Calculate ELO rating change for player A"""
        expected_a = 1 / (1 + 10 ** ((rating_b - rating_a) / 400))
        return K_FACTOR * (score_a - expected_a)
    
    # def generate_random_matchup():
    #     """Generate a random matchup between two players, prioritizing those with fewer matchups"""
    #     df = players_data()
    #     if len(df) >= 2:
    #         # Calculate weights - players with fewer matchups get higher weights
    #         # Add 1 to avoid division by zero, then invert so fewer matchups = higher weight
    #         max_matchups = df["matches"].max() if df["matches"].max() > 0 else 1
    #         weights = (max_matchups + 1) - df["matches"]
            
    #         # Convert to probability weights
    #         total_weight = weights.sum()
    #         probabilities = weights / total_weight
            
    #         # Select first player based on weights
    #         p1 = np.random.choice(df["name"].tolist(), p=probabilities.tolist())
            
    #         # For second player, exclude the first player and recalculate weights
    #         remaining_df = df[df["name"] != p1].copy()
    #         if len(remaining_df) > 0:
    #             remaining_weights = (max_matchups + 1) - remaining_df["matches"]
    #             remaining_total = remaining_weights.sum()
    #             remaining_probs = remaining_weights / remaining_total
                
    #             p2 = np.random.choice(remaining_df["name"].tolist(), p=remaining_probs.tolist())
    #         else:
    #             # Fallback if somehow no other players available
    #             p2 = df[df["name"] != p1]["name"].iloc[0]
            
    #         current_player1.set(p1)
    #         current_player2.set(p2)
    
    # # Generate initial matchup when app starts
    # @reactive.effect
    # def _():
    #     if current_player1() == "" and current_player2() == "":
    #         generate_random_matchup()

    def generate_random_matchup():
        """Generate a random matchup between two players, prioritizing those with more wins"""
        df = players_data()
        if len(df) >= 2:
            df_small = df.copy()
            # Calculate weights - players with more wins get higher weights
            # Add 1 to give all players a base chance, then add wins for scaling
            weights = 1 + (df_small["wins"] ** 1)
            
            # Convert to probability weights
            total_weight = weights.sum()
            probabilities = weights / total_weight
            
            # Select first player based on weights
            p1 = np.random.choice(df_small["name"].tolist(), p=probabilities.tolist())
            
            # For second player, exclude the first player and recalculate weights
            remaining_df = df[(df["name"] != p1)].copy()
            if len(remaining_df) > 0:
                remaining_weights = 1 + remaining_df["wins"] ** 1
                remaining_total = remaining_weights.sum()
                remaining_probs = remaining_weights / remaining_total
                
                p2 = np.random.choice(remaining_df["name"].tolist(), p=remaining_probs.tolist())
            else:
                # Fallback if somehow no other players available
                p2 = df[df["name"] != p1]["name"].iloc[0]
            
            current_player1.set(p1)
            current_player2.set(p2)

    # Generate initial matchup when app starts
    @reactive.effect
    def _():
        if current_player1() == "" and current_player2() == "":
            generate_random_matchup()

    @reactive.effect
    @reactive.event(input.new_matchup)
    def generate_new_matchup():
        """Generate a new random matchup"""
        generate_random_matchup()
    
    @reactive.effect
    @reactive.event(input.update_rankings)
    def trigger_rankings_update():
        """Manually trigger rankings table update"""
        rankings_update_trigger.set(rankings_update_trigger() + 1)
        ui.notification_show("Rankings table updated!", type="message")
    
    @reactive.effect
    @reactive.event(input.upload_rankings)
    def upload_to_huggingface():
        """Upload current rankings to Hugging Face"""
        try:
            df = players_data()
            
            # Sort by ELO rating (descending)
            df_sorted = df.sort_values("elo", ascending=False).reset_index(drop=True)
            
            # Add rank column
            df_sorted["rank"] = range(1, len(df_sorted) + 1)
            
            # Calculate win percentage
            df_sorted["win_pct"] = (df_sorted["wins"] / df_sorted["matches"] * 100).round(1)
            df_sorted.loc[df_sorted["matches"] == 0, "win_pct"] = 0
            
            # Format columns for upload
            upload_df = df_sorted[["rank", "name", "elo", "matches", "wins", "win_pct"]].copy()
            upload_df["elo"] = upload_df["elo"].round().astype(int)
            upload_df.columns = ["Rank", "Player", "ELO", "Matches", "Wins", "Win %"]
            
            # Upload updated CSV to Hugging Face
            from huggingface_hub import HfApi
            csv_buffer = io.BytesIO()
            upload_df.to_csv(csv_buffer, index=False, quoting=csv.QUOTE_MINIMAL, escapechar="\\")
            csv_buffer.seek(0)  # Reset buffer position
            
            api = HfApi()
            api.upload_file(
                path_or_fileobj=csv_buffer,
                path_in_repo="elo_rankings.csv",
                repo_id="TJStatsApps/scouting_reports",
                repo_type="dataset"
            )
            
            ui.notification_show("Rankings successfully uploaded to Hugging Face!", type="message")
            
        except Exception as e:
            ui.notification_show(f"Upload failed: {str(e)}", type="error")
    
    @render.ui
    def current_matchup():
        """Display the current matchup with winner buttons"""
        p1 = current_player1()
        p2 = current_player2()
        
        if not p1 or not p2:
            return ui.div("Generating matchup...")
        
        # Get current ELO ratings for display
        df = players_data()
        p1_elo = int(df[df["name"] == p1]["elo"].iloc[0])
        p2_elo = int(df[df["name"] == p2]["elo"].iloc[0])

        
        return ui.div(
            ui.h4("Who wins?", style="text-align: center; margin-bottom: 20px;"),
            ui.div(
                ui.input_action_button(
                    "winner_p1", 
                    f"{p1}\n(ELO: {p1_elo})",
                    class_="btn-success btn-lg",
                    style="width: 100%; margin-bottom: 10px; white-space: pre-line;"
                ),
                ui.h5("VS", style="text-align: center; margin: 10px 0;"),
                ui.input_action_button(
                    "winner_p2", 
                    f"{p2}\n(ELO: {p2_elo})",
                    class_="btn-success btn-lg",
                    style="width: 100%; white-space: pre-line;"
                ),
                style="text-align: center;"
            )
        )
    
    @reactive.effect
    @reactive.event(input.winner_p1)
    def record_p1_win():
        """Record player 1 as winner"""
        record_match_result(current_player1(), current_player2(), current_player1())
    
    @reactive.effect
    @reactive.event(input.winner_p2)
    def record_p2_win():
        """Record player 2 as winner"""
        record_match_result(current_player1(), current_player2(), current_player2())


    def record_match_result(p1, p2, winner):
        """Record a match result and update ELO ratings"""
        if not p1 or not p2:
            return
        
        df = players_data()
        matches_df = matches_data()
        
        # Get current ELO ratings
        p1_elo = df[df["name"] == p1]["elo"].iloc[0]
        p2_elo = df[df["name"] == p2]["elo"].iloc[0]
        
        # Calculate new ELO ratings
        if winner == p1:
            p1_score = 1
            p2_score = 0
        else:
            p1_score = 0
            p2_score = 1
        
        p1_elo_change = calculate_elo_change(p1_elo, p2_elo, p1_score)
        p2_elo_change = calculate_elo_change(p2_elo, p1_elo, p2_score)
        
        new_p1_elo = p1_elo + p1_elo_change
        new_p2_elo = p2_elo + p2_elo_change
        
        # Update player data
        df.loc[df["name"] == p1, "elo"] = new_p1_elo
        df.loc[df["name"] == p2, "elo"] = new_p2_elo
        df.loc[df["name"] == p1, "matches"] += 1
        df.loc[df["name"] == p2, "matches"] += 1
        df.loc[df["name"] == winner, "wins"] += 1
        
        players_data.set(df)
        
        # Record match history
        new_match = pd.DataFrame({
            "player1": [p1],
            "player2": [p2],
            "winner": [winner],
            "player1_elo_before": [round(p1_elo)],
            "player2_elo_before": [round(p2_elo)],
            "player1_elo_after": [round(new_p1_elo)],
            "player2_elo_after": [round(new_p2_elo)]
        })
        
        updated_matches_df = pd.concat([new_match, matches_df], ignore_index=True)
        matches_data.set(updated_matches_df)
        
        # Show notification and generate new matchup
        loser = p1 if winner == p2 else p2
        elo_change = abs(round(p1_elo_change)) if winner == p1 else abs(round(p2_elo_change))
        ui.notification_show(f"{winner} defeats {loser}! (+{elo_change} ELO)", type="message")
        
        # Auto-generate next matchup
        generate_random_matchup()
    
    @reactive.effect
    @reactive.event(input.reset_all)
    def reset_all_data():
        """Reset all player and match data"""
        initial_df = pd.DataFrame({
            "name": DEFAULT_PLAYERS,
            "elo": [INITIAL_ELO] * len(DEFAULT_PLAYERS),
            "matches": [0] * len(DEFAULT_PLAYERS),
            "wins": [0] * len(DEFAULT_PLAYERS)
        })
        players_data.set(initial_df)
        matches_data.set(pd.DataFrame(columns=["player1", "player2", "winner", "player1_elo_before", "player2_elo_before", "player1_elo_after", "player2_elo_after"]))
        generate_random_matchup()
        ui.notification_show("All data has been reset!", type="message")
    
    @render.data_frame
    def rankings_table():
        """Display current ELO rankings"""
        # Include the update trigger to force refresh when button is clicked
        rankings_update_trigger()
        
        df = players_data()
        
        # Sort by ELO rating (descending)
        df_sorted = df.sort_values("elo", ascending=False).reset_index(drop=True)
        
        # Add rank column
        df_sorted["rank"] = range(1, len(df_sorted) + 1)
        
        # Calculate win percentage
        df_sorted["win_pct"] = (df_sorted["wins"] / df_sorted["matches"] * 100).round(1)
        df_sorted.loc[df_sorted["matches"] == 0, "win_pct"] = 0
        
        # Format and reorder columns
        display_df = df_sorted[["rank", "name", "elo", "matches", "wins", "win_pct"]].copy()
        display_df["elo"] = display_df["elo"].round().astype(int)
        display_df.columns = ["Rank", "Player", "ELO", "Matches", "Wins", "Win %"]
        
        return display_df
    
    @render.data_frame
    def match_history():
        """Display match history"""
        df = matches_data()
        if df.empty:
            return pd.DataFrame({"Message": ["No matches recorded yet. Click on a winner to start!"]})
        
        # Format display
        display_df = df.copy()
        display_df.columns = ["Player 1", "Player 2", "Winner", "P1 ELO Before", "P2 ELO Before", "P1 ELO After", "P2 ELO After"]
        
        return display_df

app = App(app_ui, server)