Skip to content

v0.7.0 — ML Foundation: Lap Time & Tire Degradation

Choose a tag to compare

View all tags
@VforVitorio VforVitorio released this 05 Mar 13:09
· 922 commits to main since this release
v0.7.0
222c7c8

Closes out the ML foundation phase. Two predictive models fully trained, validated on 2025 held-out data, and exported to data/models/.

Lap Time Predictor (N06)

XGBoost delta-lap-time model with circuit clustering features. Trained on 2023–2024, tested on 2025.

  • MAE 0.392s on 2025 test data
  • Features include fuel-corrected lap times, tyre life, compound, circuit cluster, race phase

Tire Degradation Predictor (N07–N10)

Temporal Convolutional Network (TCN) in PyTorch with per-compound fine-tuning and MC Dropout for uncertainty quantification.

  • Architecture: TCN → per-compound heads (SOFT / MEDIUM / HARD)
  • MC Dropout: N=50 forward passes at inference time
  • Calibration JSON exported alongside model weights
  • Exported to data/models/tire_degradation/

Notes

  • src/ module integration deferred to v0.9.0 (post-notebooks phase)
  • Tire compound mapping (C1–C5) identified as future enhancement — current data only provides relative names (SOFT/MEDIUM/HARD)

Next: v0.8.0 — Additional Predictors (Overtake its done, Safety Car probability in progress)

Assets 2
Loading