Upload 7 files

stuff_model/feature_engineering.py

@@ -0,0 +1,133 @@
+import polars as pl
+import numpy as np
+
+def feature_engineering(df: pl.DataFrame) -> pl.DataFrame:
+    # Extract the year from the game_date column
+    df = df.with_columns(
+        pl.col('game_date').str.slice(0, 4).alias('year')
+    )
+
+    df = df.with_columns([
+        
+        (-(pl.col('vy0')**2 - (2 * pl.col('ay') * (pl.col('y0') - 17/12)))**0.5).alias('vy_f'),
+        ])
+
+    df = df.with_columns([
+        ((pl.col('vy_f') - pl.col('vy0')) / pl.col('ay')).alias('t'),
+        ])
+
+    df = df.with_columns([
+        (pl.col('vz0') + (pl.col('az') * pl.col('t'))).alias('vz_f'),
+        (pl.col('vx0') + (pl.col('ax') * pl.col('t'))).alias('vx_f')
+        ])
+
+    df = df.with_columns([
+            (-np.arctan(pl.col('vz_f') / pl.col('vy_f')) * (180 / np.pi)).alias('vaa'),
+            (-np.arctan(pl.col('vx_f') / pl.col('vy_f')) * (180 / np.pi)).alias('haa')
+        ])
+
+    # Mirror horizontal break for left-handed pitchers
+    df = df.with_columns(
+        pl.when(pl.col('pitcher_hand') == 'L')
+        .then(-pl.col('ax'))
+        .otherwise(pl.col('ax'))
+        .alias('ax')
+    )
+
+    # Mirror horizontal break for left-handed pitchers
+    df = df.with_columns(
+        pl.when(pl.col('pitcher_hand') == 'L')
+        .then(-pl.col('hb'))
+        .otherwise(pl.col('hb'))
+        .alias('hb')
+    )
+
+    # Mirror horizontal release point for left-handed pitchers
+    df = df.with_columns(
+        pl.when(pl.col('pitcher_hand') == 'L')
+        .then(pl.col('x0'))
+        .otherwise(-pl.col('x0'))
+        .alias('x0')
+    )
+
+    # Define the pitch types to be considered
+    pitch_types = ['SI', 'FF', 'FC']
+
+    # Filter the DataFrame to include only the specified pitch types
+    df_filtered = df.filter(pl.col('pitch_type').is_in(pitch_types))
+
+    # Group by pitcher_id and year, then aggregate to calculate average speed and usage percentage
+    df_agg = df_filtered.group_by(['pitcher_id', 'year', 'pitch_type']).agg([
+        pl.col('start_speed').mean().alias('avg_fastball_speed'),
+        pl.col('az').mean().alias('avg_fastball_az'),
+        pl.col('ax').mean().alias('avg_fastball_ax'),
+        pl.len().alias('count')
+    ])
+
+    # Sort the aggregated data by count and average fastball speed
+    df_agg = df_agg.sort(['count', 'avg_fastball_speed'], descending=[True, True])
+    df_agg = df_agg.unique(subset=['pitcher_id', 'year'], keep='first')
+
+    # Join the aggregated data with the main DataFrame
+    df = df.join(df_agg, on=['pitcher_id', 'year'])
+
+    # If no fastball, use the fastest pitch for avg_fastball_speed
+    df = df.with_columns(
+        pl.when(pl.col('avg_fastball_speed').is_null())
+        .then(pl.col('start_speed').max().over('pitcher_id'))
+        .otherwise(pl.col('avg_fastball_speed'))
+        .alias('avg_fastball_speed')
+    )
+
+    # If no fastball, use the fastest pitch for avg_fastball_az
+    df = df.with_columns(
+        pl.when(pl.col('avg_fastball_az').is_null())
+        .then(pl.col('az').max().over('pitcher_id'))
+        .otherwise(pl.col('avg_fastball_az'))
+        .alias('avg_fastball_az')
+    )
+
+    # If no fastball, use the fastest pitch for avg_fastball_ax
+    df = df.with_columns(
+        pl.when(pl.col('avg_fastball_ax').is_null())
+        .then(pl.col('ax').max().over('ax'))
+        .otherwise(pl.col('avg_fastball_ax'))
+        .alias('avg_fastball_ax')
+    )
+
+    # Calculate pitch differentials
+    df = df.with_columns(
+        (pl.col('start_speed') - pl.col('avg_fastball_speed')).alias('speed_diff'),
+        (pl.col('az') - pl.col('avg_fastball_az')).alias('az_diff'),
+        (pl.col('ax') - pl.col('avg_fastball_ax')).abs().alias('ax_diff')
+    )
+
+    # Cast the year column to integer type
+    df = df.with_columns(
+        pl.col('year').cast(pl.Int64)
+    )
+
+
+    
+    df = df.with_columns([
+        pl.lit('All').alias('all')
+    ])
+
+    # Calculate mound_to_release as 60.5 - extension
+    df = df.with_columns([
+        (60.5 - df["extension"]).alias("release_pos_y")
+    ])
+    
+    # Calculate delta time (Δt)
+    delta_t = (df["release_pos_y"] - df["y0"]) / df["vy0"]
+
+    # Corrected back-calculation of release_pos_x and release_pos_z
+    df = df.with_columns([
+        (df["x0"] + df["vx0"] * delta_t + 0.5 * df["ax"] * delta_t ** 2).alias("release_pos_x"),
+        (df["z0"] + df["vz0"] * delta_t + 0.5 * df["az"] * delta_t ** 2).alias("release_pos_z")
+    ])
+
+    
+
+    
+    return df

stuff_model/lgbm_model_2020_2023.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41001a1acf6ce7dbe247f1b8b7e68a1bb1b112f39d080b7e95a83479e56cb7c1
+size 3092328

stuff_model/stuff_apply.py

@@ -0,0 +1,57 @@
+import polars as pl
+import joblib
+
+model = joblib.load('stuff_model/lgbm_model_2020_2023.joblib')
+# Read the values from the text file
+with open('stuff_model/target_stats.txt', 'r') as file:
+    lines = file.readlines()
+    target_mean = float(lines[0].strip())
+    target_std = float(lines[1].strip())
+    
+# Define the features to be used for training
+features = ['start_speed',
+            'spin_rate',
+            'extension',
+            'az',
+            'ax',
+            'x0',
+            'z0',
+            'speed_diff',
+            'az_diff',
+            'ax_diff']
+
+
+def stuff_apply(df:pl.DataFrame) -> pl.DataFrame:
+    # Filter the dataframe to include only the rows for the year 2024 and drop rows with null values in the specified features and target column
+    # df_test = df.drop_nulls(subset=features)
+    df_test = df.clone()
+
+    # Predict the target values for the 2024 data using the trained model
+    df_test = df_test.with_columns(
+        pl.Series(name="target", values=model.predict(df_test[features].to_numpy()))
+    )
+    # Standardize the target column to create a z-score
+    df_test = df_test.with_columns(
+        ((pl.col('target') - target_mean) / target_std).alias('target_zscore')
+    )
+
+    # Convert the z-score to tj_stuff_plus
+    df_test = df_test.with_columns(
+        (100 - (pl.col('target_zscore') * 10)).alias('tj_stuff_plus')
+    )
+
+    df_pitch_types = pl.read_csv('stuff_model/tj_stuff_plus_pitch.csv')
+
+    # Join the pitch type statistics with the main DataFrame based on pitch_type
+    df_pitch_all = df_test.join(df_pitch_types, left_on='pitch_type', right_on='pitch_type')
+
+    # Normalize pitch_grade values to a range between -0.5 and 0.5 based on the percentiles
+    df_pitch_all = df_pitch_all.with_columns(
+        ((pl.col('tj_stuff_plus') - pl.col('mean')) / pl.col('std')).alias('pitch_grade')
+    )
+
+    # Scale the pitch_grade values to a range between 20 and 80
+    df_pitch_all = df_pitch_all.with_columns(
+        (pl.col('pitch_grade') * 10 + 50).clip(20, 80)
+    )
+    return df_pitch_all

stuff_model/target_stats.txt

@@ -0,0 +1,2 @@
+0.0034732498406374636
+0.006846752748626548

stuff_model/tj_stuff_plus_pitch.csv

@@ -0,0 +1,16 @@
+pitch_type,mean,std,median,min,max,percentile_1,percentile_99
+ST,106.44784631565936,5.593943599731136,106.24878922952112,91.18894850636659,125.29541262167034,91.69322149368426,125.25688309207108
+SV,103.73183202363764,3.001226780758946,103.50047554089315,93.3173875900245,111.34757479687066,93.32953434698274,111.33689503153641
+SL,103.49296290610897,5.265572779780409,103.19144262214559,88.84957017284297,121.88798777026031,89.76670287371176,121.36013955239422
+KC,101.8993919341341,4.271694896723436,100.79211889194949,93.69754063161618,119.4933202093256,93.75149298057133,119.38166236091195
+All,99.9275100894791,5.01699442232884,99.65265124489378,84.73033633038408,116.94934527087541,86.65905811630736,116.7610246502804
+CU,99.88832068607897,4.615228571103906,99.08993373693156,89.84495168337246,119.90089262632986,90.20429983334718,117.89567125997061
+FC,98.83449547008738,5.811964883678063,98.54483029899575,83.20928731685326,119.78700324933075,83.34007602984008,118.21186533190846
+FS,98.25541635267653,6.898952096824192,98.46204303842217,72.25450024197754,114.88400714657823,73.39595959354874,114.78967217449389
+FO,98.15224613640243,1.081819065809178,99.94816563615653,94.0023252668585,100.50624750619224,94.0142169475971,100.50513134245217
+FF,97.29024735737988,6.078459125845886,97.09670890504734,81.2230917971995,118.10419744965911,81.32311771953398,117.7938724746093
+SC,97.27958020025409,1.2452898498180456,97.27958020025409,93.536223938276,101.02293646223218,93.54371065079995,101.01544974970822
+CH,96.35866365133434,6.178939251378385,95.80884625564597,81.28802319264824,121.14136334013493,82.02275793969746,119.09639344796777
+SI,95.14161603816645,4.9734372581529955,95.11657827702109,82.5850956341191,112.99618112461533,82.8856383780296,112.72626192694757
+CS,93.97853627048322,0.0,93.97853627048322,93.97853627048322,93.97853627048322,93.97853627048322,93.97853627048322
+KN,93.41890096234394,0.0,93.41890096234394,93.41890096234394,93.41890096234394,93.41890096234394,93.41890096234394