Steve Schuster (@growcasting) header
Steve Schuster (@growcasting) profile picture

Catch someone doing something right.

Joined January 2017
Took six years, but finally joined the four plate club. Had to be for two reps and had to be wearing the Driveline shirt. Backed it up for another double. Retiring from lifting and need a new hobby.
2
19
3,026
A love letter to indy ball enthusiasts: steve-schuster.shinyapps.io/… An all-time register you can aggregate by team/franchise/league/player/career, plus a map of every player where their birthplace is on record.
6
6
3
37
15,943
This Reggie moment has always stood above any of the statistical achievements. At the end of the 2017 championship celebration, Reggie went out of his way to give props to our closer who had a tough finish to the season.
2
1
34
2,654
B-Mets win followed by Goldeyes rubber game.
2
15
1,816
Suggesting a negative stride length clears the hips into foot plant, ties into creating more hip/shoulder separation. However, it was surprising to see that HSS had little correlation to ball velocity across the four playing levels.
2
1
1
1,063
One variable that stood out, but was not part of the VIP group was stride direction. Positive values meant the athlete strode arm side (RHP towards 3B line) across body, whereas negative values meant the athlete strode more glove side ("swingy" lead leg).
1
2
707
- Arm Slot - Horizontal Abduction/Scap Retraction at Foot Plant - Torso Rotational Velocity
1
1
646
- Athlete's Age - Energy Transfer: Torso to Shoulder - Maximum External Rotation (layback) - Maximum Shoulder Horizontal Abduction (scap retraction)
1
1
653
- Energy Transfer Across Elbow - Energy Transfer Across Shoulder - Maximum shoulder internal rotation velocity - Peak shoulder internal rotation torque
1
2
674
Was interested to see how a more specific set of predictors would perform, and ran another xgboost model using the 11 VIP variables from the five blender trials. The r-squared and rmse actually performed slightly worse (though, only ran one trial here).
1
620
Here are the predictors of most importance for the four best xgboost models. You'll see a lot of consistency between these four. Although, they differ somewhat from the VIP chart for the linear model.
1
1
679
Five xgboost "blender" models were then run using different seed numbers, four of which performed extremely well on their respective test data sets (r-squared values over 0.92, and rmse values under 1.4 mph).
1
1
612