You can find the implementation for DKT (regression) in the repository at this address. The code for the DKT classification is available here instead. Give a look at the DKT class in both file, it is pretty simple to define a DKT model. Note that, there is also an associated train and test look under the same class.
For DKT regression, the only object you need to provide is the backbone (as you can see at this line). The backbone can be your own (e.g. ResNet) or one of the backbones that you can find here in our repo. We have used a Conv4 in our experiments.
Moreover, if you are using DKT for regression I suggest to give a look at the code for the sine-waves experiments that you can find here. This code is self-contained, and you can see how to create a DKT model from scratch by using the ExactGP model from GPytorch.
Hi @Shantan243 .
You can find the implementation for DKT (regression) in the repository at this address. The code for the DKT classification is available here instead. Give a look at the
DKTclass in both file, it is pretty simple to define a DKT model. Note that, there is also an associated train and test look under the same class.For DKT regression, the only object you need to provide is the backbone (as you can see at this line). The backbone can be your own (e.g. ResNet) or one of the backbones that you can find here in our repo. We have used a Conv4 in our experiments.
Moreover, if you are using DKT for regression I suggest to give a look at the code for the sine-waves experiments that you can find here. This code is self-contained, and you can see how to create a DKT model from scratch by using the ExactGP model from GPytorch.