Model Retraining Schedule
Retraining a model is scheduled every 3 months or following a significant change in market dynamics. The current retraining schedule for both PD and P7DAY follows:
Status | Retrained model release dates | Retraining process and model changes |
---|---|---|
Future | EO April 2025 |
|
Completed | 28th January 2025 | Retrained model captures recent market dynamics. Hyperparameter settings and feature are the same as the previous model. The training period for this model has been shortened to prevent overfitting to old data primarily before the influx of BDUs. |
Completed | 12th November 2024 | Retrained model captures recent market dynamics. Hyperparameter settings and feature are the same as the previous model. The training period for this model includes the most recent market data otherwise the training periods are the same as the previous model. |
Completed | 26th July 2024 | Retrained model captures recent market dynamics. Hyperparameter settings and feature are the same as the previous model. The training period for this model includes the most recent market data otherwise the training periods are the same as the previous model. |
Completed | 6th May 2024 | Retrained model captures recent market dynamics. Hyperparameter settings and feature are the same as the previous model. The training period for this model includes the most recent market data otherwise the training periods are the same as the previous model. |
Completed | 5th February 2024 | Retrained model captures recent market dynamics. Overall approximately 5% improvement compared to old model. |
Completed | 9th November 2023 | Retrained model resulted in 5% and 10%-15% improvement compared to the old predispatch and 7day models respectively. No changes to the feature list nor the hyper parameters. |
Completed | 8th August 2023 | Significant improvement compared to old model. Refer to current model description page to see error metric improvements. |
Completed | 6th June 2023 | No material improvements to the model could be made, hence the model has not changed. |