app.py

from typing import Optional, Tuple

import gradio as gr
import pandas as pd
from pathlib import Path
import seaborn as sns
import matplotlib.pyplot as plt
from wordcloud import WordCloud
import numpy as np

def _rename_columns(df: pd.DataFrame, is_tournament: bool) -> pd.DataFrame:
    columns = {
        "Rating": "rating",
        "Result": "result",
        "Scores": "scores",
        "Opponent": "opponent",
        "OpponentRating": "opponent_rating",
    }

    if is_tournament:
        columns.update({
            "TournamentStartDate": "tournament_start_date",
            "TournamentEndDate": "tournament_end_date",
            " Touranament": "tournament",
        })
    else:
        columns.update({
            "EventDate": "event_date",
            "LeagueName": "league_name"
        })

    return df.rename(columns=columns)


def _fix_dtypes(df: pd.DataFrame, is_tournament: bool) -> pd.DataFrame:
    if is_tournament:
        df["tournament_start_date"] = pd.to_datetime(df["tournament_start_date"])
        df["tournament_end_date"] = pd.to_datetime(df["tournament_end_date"])
        df["tournament"] = df["tournament"].astype('category')
    else:
        df["event_date"] = pd.to_datetime(df["event_date"])
        df["league_name"] = df["league_name"].astype('string')

    df["rating"] = df["rating"].astype('int')
    df["result"] = df["result"].astype('category')
    df["scores"] = df["scores"].astype('string')
    df["opponent"] = df["opponent"].astype('category')
    df["opponent_rating"] = df["opponent_rating"].astype('int')

    return df


def _check_match_type(match_type: str) -> str:
    allowed_match_types = {"tournament", "league"}
    if match_type not in allowed_match_types:
        raise ValueError(
            f"The only supported match types are {allowed_match_types}. Found match type of '{match_type}'.")
    return match_type


def get_num_competitions_played(df: pd.DataFrame, is_tournament: bool) -> int:
    key_name = "tournament" if is_tournament else "event_date"
    return df[key_name].nunique()


def get_matches_per_competition_fig(df: pd.DataFrame, is_tournament: bool):
    fig = plt.figure()
    plt.title('Matches per competition')
    sns.histplot(df.groupby('tournament' if is_tournament else "event_date").size())
    plt.xlabel('Number of matches in competition')
    return fig


def get_competition_name_word_cloud_fig(df: pd.DataFrame, is_tournament: bool):
    fig = plt.figure()
    key_name = "tournament" if is_tournament else "league_name"
    wordcloud = WordCloud().generate(" ".join(df[key_name].values.tolist()))
    plt.imshow(wordcloud, interpolation='bilinear')
    plt.axis("off")
    return fig


def get_opponent_name_word_cloud_fig(df: pd.DataFrame):
    fig = plt.figure()
    wordcloud = WordCloud().generate(" ".join(df.opponent.values.tolist()))
    plt.imshow(wordcloud, interpolation='bilinear')
    plt.axis("off")
    return fig


def get_rating_over_time_fig(df: pd.DataFrame, is_tournament: bool):
    fig = plt.figure()
    plt.title('Rating over time')
    sns.lineplot(data=df,
                 x="tournament_end_date" if is_tournament else "event_date",
                 y="rating",
                 marker='.',
                 markersize=10)
    plt.xlabel('Competition date')
    plt.ylabel('Rating')
    return fig


def get_max_int(int_csv_str: str) -> int:
    """Get the max int from an int CSV."""
    ints = [int(i.strip()) for i in int_csv_str.split(',') if i]
    return max(ints)


def get_match_with_longest_game(df: pd.DataFrame, is_tournament: bool) -> Optional[pd.DataFrame]:
    if not is_tournament:
        return None
    return df.loc[[np.argmax(df.scores.apply(get_max_int))]]


def get_win_loss_record_str(group_df) -> str:
    if len(group_df) > 0:
        win_loss_counts = group_df.value_counts()
        n_wins = win_loss_counts.Won if hasattr(win_loss_counts, "Won") else 0
        n_losses = win_loss_counts.Lost if hasattr(win_loss_counts, "Lost") else 0
    else:
        n_wins = 0
        n_losses = 0

    return f"{n_wins}, {n_losses}"


def get_most_frequent_opponents(df: pd.DataFrame, top_n: int = 5) -> pd.DataFrame:
    df_with_opponents = df.loc[df.opponent != "-, -"]

    most_common_opponents_df = df_with_opponents.groupby('opponent').agg({"result": [get_win_loss_record_str, "size"]})
    most_common_opponents_df.columns = most_common_opponents_df.columns.get_level_values(1)
    most_common_opponents_df.rename({"get_win_loss_record_str": "Win/loss record", "size": "Number of matches"}, axis=1,
                                    inplace=True)
    most_common_opponents_df["Opponent"] = most_common_opponents_df.index
    return most_common_opponents_df.sort_values("Number of matches", ascending=False)[
        ["Opponent", "Number of matches", "Win/loss record"]].head(top_n)


def get_best_wins(df: pd.DataFrame, top_n: int = 5) -> pd.DataFrame:
    """Get the top-n wins sorted by opponent rating."""
    return df.loc[df.result == 'Won'].sort_values("opponent_rating", ascending=False).head(top_n)


def get_biggest_upsets(df: pd.DataFrame, top_n: int = 5) -> pd.DataFrame:
    """Get the top-n wins sorted by rating difference."""
    df['rating_difference'] = df['opponent_rating'] - df['rating']
    return df.loc[df.result == 'Won'].sort_values("rating_difference", ascending=False).head(top_n)


def get_highest_rated_opponent(df: pd.DataFrame) -> pd.DataFrame:
    return df.iloc[df.opponent_rating.idxmax()].to_frame().transpose()


def get_opponent_rating_distr_fig(df: pd.DataFrame):
    fig = plt.figure()
    plt.title('Opponent rating distribution')
    sns.histplot(data=df, x="opponent_rating", hue='result')
    plt.xlabel('Opponent rating')
    return fig


def get_opponent_rating_dist_over_time_fig(df: pd.DataFrame, is_tournament: bool):
    fig, ax = plt.subplots(figsize=(12, 8))
    plt.title(f'Opponent rating distribution over time')
    x_key_name = "tournament_end_date" if is_tournament else "event_date"
    sns.violinplot(data=df,
                   x=df[x_key_name].dt.year,
                   y="opponent_rating",
                   hue="result",
                   split=True,
                   inner='points',
                   cut=1,
                   ax=ax)
    plt.xlabel('Competition year')
    plt.ylabel('Opponent rating')
    return fig


def load_match_df(file_path: Path) -> Tuple[pd.DataFrame, bool]:
    match_type = _check_match_type(file_path.name.split('_')[0])
    is_tournament = match_type == "tournament"

    df = pd.read_csv(file_path)
    df = _rename_columns(df, is_tournament)
    df = _fix_dtypes(df, is_tournament)

    return df, is_tournament


def usatt_rating_analyzer(file_obj):
    # Load data.
    df, is_tournament = load_match_df(Path(file_obj.name))

    # Create outputs.
    n_competitions_played = get_num_competitions_played(df, is_tournament)
    n_matches_played = len(df)
    matches_per_competition_fig = get_matches_per_competition_fig(df, is_tournament)
    opponent_name_word_cloud_fig = get_opponent_name_word_cloud_fig(df)
    competition_name_word_cloud_fig = get_competition_name_word_cloud_fig(df, is_tournament)
    most_frequent_opponents = get_most_frequent_opponents(df)
    best_wins = get_best_wins(df)
    biggest_upsets = get_biggest_upsets(df)
    highest_rated_opponent = get_highest_rated_opponent(df)
    rating_over_time_fig = get_rating_over_time_fig(df, is_tournament)
    match_with_longest_game = get_match_with_longest_game(df, is_tournament)
    opponent_rating_distr_fig = get_opponent_rating_distr_fig(df)
    opponent_rating_dist_over_time_fig = get_opponent_rating_dist_over_time_fig(df, is_tournament)

    return (n_competitions_played,
            n_matches_played,
            matches_per_competition_fig,
            opponent_name_word_cloud_fig,
            competition_name_word_cloud_fig,
            most_frequent_opponents,
            best_wins,
            biggest_upsets,
            highest_rated_opponent,
            rating_over_time_fig,
            match_with_longest_game,
            opponent_rating_distr_fig,
            opponent_rating_dist_over_time_fig,
            )


with gr.Blocks() as demo:
    analyze_btn_title = "Analyze"
    gr.Markdown(f"""# USATT rating analyzer
    Analyze [USA table tennis](https://www.teamusa.org/usa-table-tennis) tournament and league results. 

    ## Downloading match results
    1. Make sure you are [logged in](https://usatt.simplycompete.com/login/auth) to your USATT account.
    2. Find the *active* player you wish to analyze (e.g.,  [Kanak Jha](https://usatt.simplycompete.com/userAccount/up/3431)).
    3. Under 'Tournaments' or 'Leagues', click *Download Tournament/League Match History*.

    ## Usage
    1. Simply add your tournament/league match history CSV file and click the "{analyze_btn_title}" button.
    """)
    with gr.Row():
        with gr.Column():
            input_file = gr.File(label='USATT Results File', file_types=['file'])
            btn = gr.Button(analyze_btn_title)

    with gr.Group():
        with gr.Row():
            with gr.Column():
                num_comps_box = gr.Textbox(lines=1, label="Number of competitions (tournaments/leagues) played")
            with gr.Column():
                num_matches_box = gr.Textbox(lines=1, label="Number of matches played")
        
        with gr.Row():
            with gr.Column():
                rating_over_time_plot = gr.Plot(show_label=False)
            with gr.Column():
                matches_per_comp_plot = gr.Plot(show_label=False)
                
        with gr.Row():
            with gr.Column():
                opponent_names_plot = gr.Plot(label="Opponent names")
            with gr.Column():
                comp_names_plot = gr.Plot(label="Competition names")

        with gr.Row():
            with gr.Column():
                most_frequent_opponents_gdf = gr.Dataframe(label="Most frequent opponents", max_rows=5)
                best_wins_gdf = gr.Dataframe(label="Best wins (matches won sorted by opponent post-competition rating)", 
                                             max_rows=5)
                biggest_upsets_gdf = gr.Dataframe(label="Biggest upsets (matches won sorted by rating - opponent post-competition rating)",
                                                  max_rows=5)
                highest_rated_opponent_gdf = gr.Dataframe(label="Best opponent", max_rows=1)

        match_longest_game_gdf = gr.Dataframe(label="Match with longest game", max_rows=1)

        with gr.Row():
            with gr.Column():
                opponent_rating_dist_plot = gr.Plot(show_label=False)
            with gr.Column():
                opponent_rating_dist_over_time_plot = gr.Plot(show_label=False)

    inputs = [input_file]
    outputs = [
        num_comps_box,
        num_matches_box,
        matches_per_comp_plot,
        opponent_names_plot,
        comp_names_plot,
        most_frequent_opponents_gdf,
        best_wins_gdf,
        biggest_upsets_gdf,
        highest_rated_opponent_gdf,
        rating_over_time_plot,
        match_longest_game_gdf,
        opponent_rating_dist_plot,
        opponent_rating_dist_over_time_plot,
    ]


    btn.click(usatt_rating_analyzer, inputs=inputs, outputs=outputs)

if __name__ == "__main__":
    demo.launch()