Closed venondev closed 5 months ago
We have to decide which dataset we want to publish and use for the training. Currently we just train models, xai methods etc. for both types of datasets.
We decided to use the no-subsampling dataset due to better performance of models on the test set.
We have to decide which dataset we want to publish and use for the training. Currently we just train models, xai methods etc. for both types of datasets.