pytorch-frame icon indicating copy to clipboard operation
pytorch-frame copied to clipboard

Support for additional hyperparameters

Open qychen2001 opened this issue 1 year ago • 0 comments

Very good work! But when I wanted to use hyperparameter search for thesis result reproduction, I realized that the implementation of this repository ignores a lot of hyperparameters. For example dropout etc. The absence of these parameters will hinder the further use of this work in scientific comparisons. So is it considered to add additional hyperparameters? Below is an example of FT-Transformer showing the hyperparameter space in the original paper and the version implemented in this repository.

image image

For a more detailed comparison check out the original repository implementation.

qychen2001 avatar Jul 08 '24 02:07 qychen2001