MS2 - Baseline Models (15%)

[TimkLee]

• [x] Create training and validation set
• [x] Train a logistic regression with only the distance feature (default settings)
• [x] Produce 4 figures
• [x] Train logistic regression with angle feature, produce same curves as before
• [x] Train logistic regression with angle and distance feature, produce same curves as before
• [x] Include link to comet.ml next to the figures.

[TimkLee]

Notes:

• interested in the notion of expected goals (predict_proba(X))
• Receiver Operating Characteristic (ROC) curves and the AUC metric of the ROC curve.
• The goal rate (#goals / (#no_goals + #goals)) as a function of the shot probability model percentile
• The cumulative proportion of goals (not shots) as a function of the shot probability model percentile.
• The reliability diagram (calibration curve).
• 4 lines on each figure:
  • Logistic Regression, trained on distance only (already done above)
  • Logistic Regression, trained on angle only
  • Logistic Regression, trained on both distance and angle
  • Random baseline: predicted probability is sampled from a uniform distribution