automl / TabPFN

Official implementation of the TabPFN paper (https://arxiv.org/abs/2207.01848) and the tabpfn package.
http://priorlabs.ai
Apache License 2.0
1.22k stars 109 forks source link

what are the inputs to the encoder and decoder of the Transformer model during pre-training #56

Closed Sakura5540 closed 1 year ago

Sakura5540 commented 1 year ago

Hello team

I'm sorry I wasn't clear in my last question.My question is, what are the inputs to the encoder and decoder of the Transformer model during pre-training, respectively? I guess, the encoder's input is the features of the dataset, but I don't know what the decoder's input will be?

Thank you~

noahho commented 1 year ago

The encoders inputs are the synthetic datasets (see the synthetic datasets here: https://github.com/automl/TabPFN/blob/main/tabpfn/PriorFittingCustomPrior.ipynb), the decoder inputs are the encoded datasets after they were encoded by the transformer. The loss is applied to the target values of the synthetic datasets (see same notebook).

All the best