Hi, I would like to know to what extent will Zindi/the judging committee try to reproduce the submitted results.
Is it just on the model inference part or we are talking about a full end-to-end run that can reproduce both the fine-tuned model (i.e. model training) as well as the prediction from the trained model?
Hi, I have the similar questions as this post. I would like to know whether the reproducibility requirement also covers the data augmentation process. For example, will the committee attempt to reproduce any augmented or derived datasets used during training, or is it sufficient to provide the final processed data and trained model artifacts?
For sure, you will need to present the full pipeline, and make it reproducible. We target to analyse your work end to end. Thanks