Bugs in design_bench.oracles.tensorflow.transformer_oracle.py

[trxcc]

It is a minor typo. In line 335, when we conduct protected_predcict, training is not defined thus it may raise a fault. A simple fix is to add a flag when initializing the oracle. Transformer oracle is required when considering GFP task, thus I think it is necessary to fix this bug.

[brandontrabucco]

Hi trxcc,

Thanks for your interest in our benchmark!

Could you share a snippet of how you are using the transformer_oracle and how you are loading the corresponding design_bench task so I can reproduce this error?

Thanks!

-Brandon

[trxcc]

When I attempted to reproduce the performance of the COMs for the task GFP, it was necessary to use an entry such as GFP-Transformer-v0. Additionally, a minor typo involving an undefined symbol training in line 335 triggered an error during the initialization of the oracle.

Thanks for your response!