demo.py

import pandas as pd
import numpy as np
from tqdm.auto import tqdm
import plotly.express as px
import gradio as gr
from gradio_client import Client

import os
import re

from translate import translate_pa_outcome, translate_pitch_outcome, jp_pitch_to_en_pitch, jp_pitch_to_pitch_code, translate_pitch_outcome

# load game data
game_df = pd.read_csv('game.csv').drop_duplicates()
assert len(game_df) == len(game_df['game_pk'].unique())

# load pa data
pa_df = []
for game_pk in tqdm(game_df['game_pk']):
  pa_df.append(pd.read_csv(os.path.join('pa', f'{game_pk}.csv'), dtype={'pa_pk': str}))
pa_df = pd.concat(pa_df, axis='rows')

# load pitch data
pitch_df = []
for game_pk in tqdm(game_df['game_pk']):
  pitch_df.append(pd.read_csv(os.path.join('pitch', f'{game_pk}.csv'), dtype={'pa_pk': str}))
pitch_df = pd.concat(pitch_df, axis='rows')
pitch_df

# load player data
player_df = pd.read_csv('player.csv')
player_df

# translate pa data
pa_df['_des'] = pa_df['des'].str.strip()
pa_df['des'] = pa_df['des'].str.strip()
pa_df['des_more'] = pa_df['des_more'].str.strip()
pa_df.loc[pa_df['des'].isna(), 'des'] = pa_df[pa_df['des'].isna()]['des_more']
pa_df.loc[:, 'des'] = pa_df['des'].apply(lambda item: item.split()[0] if (len(item.split()) > 1 and re.search(r'＋\d+点', item)) else item)
non_home_plate_outcome = (pa_df['des'].isin(['ボール', '見逃し', '空振り'])) | (pa_df['des'].str.endswith('塁けん制'))
pa_df.loc[non_home_plate_outcome, 'des'] = pa_df.loc[non_home_plate_outcome, 'des_more']
pa_df['des'] = pa_df['des'].apply(translate_pa_outcome)

# translate pitch data
pitch_df = pitch_df[~pitch_df['pitch_name'].isna()]
pitch_df.loc[:, 'jp_pitch_name'] = pitch_df['pitch_name']
pitch_df.loc[:, 'pitch_name'] = pitch_df['jp_pitch_name'].apply(lambda pitch_name: jp_pitch_to_en_pitch[pitch_name])
pitch_df.loc[:, 'pitch_type'] = pitch_df['jp_pitch_name'].apply(lambda pitch_name: jp_pitch_to_pitch_code[pitch_name])
pitch_df.loc[:, 'description'] = pitch_df['description'].apply(lambda item: item.split()[0] if len(item.split()) > 1 else item)
pitch_df.loc[:, 'description'] = pitch_df['description'].apply(translate_pitch_outcome)

# translate player data
client = Client("Ramos-Ramos/npb_name_translator")
en_names = client.predict(
		jp_names='\n'.join(player_df.name.tolist()),
		api_name="/predict"
)
player_df['jp_name'] = player_df['name']
player_df['name'] = [name if name != 'nan' else np.nan for name in en_names.splitlines()]

# merge pitch and pa data
df = pd.merge(pitch_df, pa_df, 'inner', on=['game_pk', 'pa_pk'])
df = pd.merge(df, player_df.rename(columns={'player_id': 'pitcher'}), 'inner', on='pitcher')
df['whiff'] = df['description'].isin(['SS', 'K'])
df['swing'] = ~df['description'].isin(['B', 'BB', 'LS', 'inv_K', 'bunt_K', 'HBP', 'SH', 'SH E', 'SH FC', 'obstruction', 'illegal_pitch', 'defensive_interference'])

# gradio function(s)
def get_usage(player):
  _df = df.set_index('name').loc[player, 'pitch_name'].value_counts()
  fig = px.pie(df.set_index('name').loc[player, 'pitch_name'], names='pitch_name')
  fig.update_traces(texttemplate='%{percent:.1%}', hovertemplate=f'<b>{player}</b><br>' + 'threw a <b>%{label}</b><br><b>%{percent:.1%}</b> of the time<br>(<b>%{value}</b> pitches)')
  return fig

# demo
with gr.Blocks(fill_height=True) as demo:
  gr.Markdown('''
  # NPB data visualization demo
  [Data from SportsNavi](https://sports.yahoo.co.jp/)
  ''')
  player = gr.Dropdown(choices=sorted(player_df['name'].dropna().tolist()), label='Player')
  player_info = gr.Markdown()
  usage = gr.Plot(label='Pitch Distribution')

  player.input(get_usage, inputs=player, outputs=usage)
  player.input(lambda player: f'# {player}', inputs=player, outputs=player_info)

demo.launch(share=True)