vignesh-99/nfl

1 2from collections import Counter3import numpy as np4import seaborn as sns5import matplotlib.pyplot as plt6import pandas as pd7from concurrent.futures import ThreadPoolExecutor8import os9 10 11TRAIN_INPUT_FILE_PATH = 'C:/Users/vigne/nfl/train_input'12TRAIN_OUTPUT_FILE_PATH = 'C:/Users/vigne/nfl/train_output'13 14 15def get_train_file_paths():16        17    def get_output_file(input_filename):18        return input_filename.replace('input', 'output')19    20    input_file_paths = []21    output_file_paths = []22    input_files_dir = TRAIN_INPUT_FILE_PATH23    for w in range(1, 19):24        input_filename = f'input_2023_w{w:02d}.csv'25        if  os.path.isfile(f'{input_files_dir}/{input_filename}'):26            output_filename = get_output_file(input_filename)27            input_file_path = os.path.join(TRAIN_INPUT_FILE_PATH, input_filename)28            output_file_path = os.path.join(TRAIN_OUTPUT_FILE_PATH, output_filename)29            input_file_paths.append(input_file_path)30            output_file_paths.append(output_file_path)31        else:32            raise Exception(f'input file for week {w} does not exist')33            34    return (input_file_paths, output_file_paths)35 36def load_file(file_path):37    return pd.read_csv(file_path)38    39def get_input_output_df():40    input_file_paths, output_file_paths = get_train_file_paths()41   42    with ThreadPoolExecutor(max_workers = 8) as executor:43        input_dfs = executor.map(load_file, input_file_paths)44        input_df = pd.concat(input_dfs, axis=0)45 46    with ThreadPoolExecutor(max_workers = 8) as executor:47        output_dfs = executor.map(load_file, output_file_paths)48        output_df = pd.concat(output_dfs, axis=0)49 50    return input_df.reset_index(drop=True), output_df.reset_index(drop=True)51 52def plot_distribution_of_features(input_df, output_df):53    predict_players_position = Counter()54    predict_players_role = Counter()55    predict_players_side = Counter()56    57    num_frames_to_predict = Counter()58    no_players_prediction_in_a_play = Counter()59    60 61    plays = input_df.groupby(['game_id', 'play_id'], as_index = False)62    63    per_play_change_in_dists = []64    per_frame_change_in_dists = []65    per_frame_change_in_x_dists = []66    per_frame_change_in_y_dists = []67    for _, play in plays:68 69        predict_players = play[play['player_to_predict']]70        71        no_players_prediction_in_a_play[predict_players['nfl_id'].nunique()]+=172 73        74        num_frames_output = play['num_frames_output'].iloc[0].item()75        num_frames_to_predict[num_frames_output]+=176        77        predict_players_last_frame = predict_players.groupby(['nfl_id'], as_index=False).last()78        for index, p_l in predict_players_last_frame.iterrows():79            game_id = p_l['game_id']80            play_id = p_l['play_id']81            p_nfl_id = p_l['nfl_id']82            p_output = output_df[(output_df['game_id'] == game_id) & (output_df['play_id'] == play_id) & (output_df['nfl_id'] == p_nfl_id)]83            84            85            s = np.array([p_l['x'], p_l['y']])86            87            total_dis = 088            for _,p_o in p_output.iterrows():89                e = np.array([p_o['x'].item(), p_o['y'].item()])90                current_dis = np.linalg.norm(e - s)91                per_frame_change_in_dists.append(current_dis)92                per_frame_change_in_x_dists.append(np.abs(e[0] - s[0]))93                per_frame_change_in_y_dists.append(np.abs(e[1] - s[1]))94            95                total_dis+= current_dis96                s = e97    98            per_play_change_in_dists.append(total_dis)99    100    101            position = p_l['player_position']102            role = p_l['player_role']103            side = p_l['player_side']104            105            predict_players_position[position]+=1106            predict_players_role[role]+=1107            predict_players_side[side]+=1108            109    def plot_bargraph(dict_items, name):110        plt.figure(figsize=(10, 6))111        df = pd.DataFrame(list(dict_items), columns = [name, 'count'])112        df = df.sort_values(by='count', ascending=False)113        sns.barplot(data=df, x=name, y='count')114        plt.show()115 116    117    plot_bargraph(predict_players_position.items(), 'predict_player position')118    plot_bargraph(predict_players_role.items(), 'predict_player role')119    plot_bargraph(predict_players_side.items(), 'predict_player side')120 121    plot_bargraph(no_players_prediction_in_a_play.items(), 'num of player to predict in a play')122    plot_bargraph(num_frames_to_predict.items(), 'num of frames to predict in a play')123 124 125    def plot_density_plot(data, title, x):126        plt.figure(figsize=(10, 6))127        sns.kdeplot(data, fill=True, color="dodgerblue")128        plt.title(title)129        plt.xlabel(x)130        plt.ylabel('Density')131        plt.show()132    133   134    plot_density_plot(per_play_change_in_dists, 'total distance moved by a player in a play', 'distance')135    plot_density_plot(per_frame_change_in_dists, 'distance moved by a player per frame', 'distance')136    plot_density_plot(per_frame_change_in_x_dists, 'distance moved by a player along x per frame', 'distance')137    plot_density_plot(per_frame_change_in_y_dists, 'distance moved by a player along y per frame', 'distance')138 139 140def get_last_frame(df):141    142    df_sorted = df.sort_values(['game_id', 'play_id', 'nfl_id', 'frame_id']).reset_index(drop=True)143    144    group_by_cols = ['game_id', 'play_id', 'nfl_id']145 146    feature_cols = ['x', 'y', 'o', 'dir', 's', 'a']147    148    df_sorted[[f'{c}_prev' for c in feature_cols]] = df_sorted.groupby(group_by_cols)[feature_cols].shift(1)149 150    #last() takes non none values from the last possible col151    #so even if last frame misses a feature , value is taken from the previous available one152    df_last_frame = df_sorted.groupby(group_by_cols, as_index=False).last()153 154    df_last_frame = df_last_frame.rename(columns={'x':'x_last', 'y':'y_last'})155    156    return df_last_frame157 158 159def predict_physics_baseline(input_df, output_df):160 161    def convert_to_radians(degrees):162        return degrees * np.pi / 180163 164    def sin(theta):165        return np.sin(convert_to_radians(theta))166    167    def cos(theta):168        return np.cos(convert_to_radians(theta))169    170 171    input_df = input_df.copy()172    173    output_df = output_df.copy()174    175    df_last_frame = get_last_frame(input_df)176 177    df_last_frame = df_last_frame[['game_id', 'play_id', 'nfl_id', 'x_last', 'y_last', 'o', 'dir', 's', 'a', 'num_frames_output']]178    179    df = output_df.merge(df_last_frame, on=['game_id', 'play_id', 'nfl_id'], how='left')180 181    sum_ = 0182    for _, group_df in df.groupby(['game_id', 'play_id', 'nfl_id'], as_index=False):183 184        group_df = group_df.sort_values('frame_id').reset_index(drop=True)185 186        prev = (group_df.iloc[0]['x'], group_df.iloc[0]['y'])187        for row in group_df.itertuples():188            dt = 0.1189 190            velocity_x = row.s * sin(row.dir)191            velocity_y_ = row.s * cos(row.dir)192            acc_x_ = row.a * sin(row.dir)193            acc_y_ = row.a * cos(row.dir)194 195            proj_x = prev[0] + velocity_x*dt + 0.5*acc_x_*(dt**2)196            proj_y = prev[1] + velocity_y_*dt + 0.5*acc_y_*(dt**2)197            198            sum_+= (row.x - proj_x)**2 + (row.y - proj_y)**2199            prev = (proj_x, proj_y)200       201    num_ele = df.shape[0]*2202    rmse = np.sqrt(sum_ / num_ele)203    print(f'RMSE of the simple physics based model is {rmse}')204 205 206input_df, output_df = get_input_output_df()207POSITION_MAPPING = [208    "FS --> Free Safety",209    "SS --> Strong Safety",210    "CB --> Cornerback",211    "MLB --> Middle Linebacker",212    "WR --> Wide Receiver",213    "TE --> Tight End",214    "QB --> Quarterback",215    "OLB --> Outside Linebacker",216    "ILB --> Inside Linebacker",217    "RB --> Running Back",218    "DE --> Defensive End",219    "FB --> Fullback",220    "NT --> Nose Tackle",221    "DT --> Defensive Tackle",222    "S --> Safety",223    "T --> Tackle",224    "LB --> Linebacker",225    "P --> Punter",226    "K --> Kicker"227]228 229PLAYER_ROLES = ['Defensive Coverage' 'Other Route Runner' 'Passer' 'Targeted Receiver']230 231PLAYER_SIDES = ['Defense', 'Offense']232 233print(f'player positions are {POSITION_MAPPING}')234print(f'player roles are {PLAYER_ROLES}')235print(f'player roles are {PLAYER_SIDES}')236 237# plot_distribution_of_features(input_df[:1_000_00], output_df)238 239predict_physics_baseline(input_df, output_df)