Machine Project: Kobe Bryant Shot Selection

does exact location hurt or not, is shot zone good enough?

What is test set

1 - Random 2 - train on regular predict on playoff 3 - train on data against 1 team in 1 season and then predict the match results against the same team in the next season. (Suns or the clippers) Divide the suns's game by each season.

Things to discuss

Do we need "game_event_id"? maybe it is useful maybe not.

Project Structure

src/ ├── 1-EDA.ipynb ├── 2-feature-engineering.ipynb |-- feature_engineering.py (In feature_engineering.py we are mainting copy of some funtions from the 2-feature-engineering.ipynb ) ├── 4-models.ipynb |-- 5-random-forrest.ipynb ├── ml03-linear-regression.ipynb ├── modeling_prep.ipynb └── shots_made_and_missed.png

Important Note If you make changes to the functions in 2-feature-engineering.ipynb make sure to sync that with feature_engineering.py

References

Link to dataset: https://www.kaggle.com/competitions/kobe-bryant-shot-selection/overview

ToDos Next tasks

• Complete the feature engineering in the same notebook. (convert categorical variables to numeric) (filip)
• Split the dataset as discussed above

ToDos

• More statistics, by Filip ( 2-feature-engineering )
• Move "removing redundant columns section to notebook 2"
• p Analysis and remove redundant rows.
• Ask Prof:
• Two styles of train/test split 1- Train: S1-regular, Test: S1-playoff. If we average shots made for S1-regular and then train and test on testset would it be okay. Techinically there is data leakage in training data but not in the test data. We'll be using season average from S1-regular in test rows. 2- If we take MV of past 5 games, would it be a problem in regular season trining data. if we do that, should we do the same for the testing data? in real life we would know data of all the games before 3- Do we remove duplicates usually? If yes why and happens if we don't?
• Does keeping "date" column makes sense? what effect does dropping have if any. Given no 2 games will have same date.
• Move Section 2 remove redundant columns to Notebook-3

ToDos that are Done

• Ask Prof Question) If this the distribution of a column (shot_distance) , can we normalize it to improve the results of linear regression on it? (done)
• How do outlier detection on categorical data (done)
• We have a categorical column wiht 2 different values "away" & "home". Question is, is it better apply one-hot encoding to it? meaning we will get 2 different columns (match_away, match_home) and in each column "1" will denote true value. Or we use a single column and have 1 denote "away" and 0 denote "home"? (done)

EDA on Columns Done: (22/22)

Eventual status of columns game_event_id - drop game_id - drop lat, lon, - drop loc_x, loc_y minutes_remaining, seconds_remaining - 8 shot_distance period playoffs season matchup shot_type shot_zone_area shot_zone_basic shot_zone_range combined_shot_type - 18 action_type opponent game_date shot_made_flag