Performance Evaluation: 12-Layer vs. 6-Layer Geneformer
#314
by
Qianaqian
- opened
Hi there,
Thanks for your amazing work ! I was curious,Does the 12 layer geneformer model performered better than 6 layer geneformer in finetune-task？
Thank you for your interest in Geneformer! To best compare the two models, we would suggest hyperparameter tuning both for your task with the same size of ranges for each hyperparameter and the same number of total trials. The max and min of the range may need to be different for each model (for example learning rate) but the size of the range should be same for a fair comparison, or you could choose a very large range to cover both models. We would suggest comparing performance on a held-out test set, separate from the held-out validation set used for optimizing hyperparameters.
ctheodoris
changed discussion status to
closed