cannot get the results that present in the article

[wanyueli]

Hi! I have run your code and using the datasets you mentioned in the article and all the steps were followed with that you said in the article, however, I can't get the results as good as presented in the article, we only get 1.42 of average RMSE on the core set. I guess the difference may cause by the data preprocessing step, so can you share the preprocessed data you used to train the network? (only the .tfrecords files is enough) Thanks a lot!

[hhassan1]

Hi, could you send an email to the corresponding author? I will try to get the data sent to you, as I cannot publish it in Github (size constraints)

[hhassan1]

Hi again. Thank you for your input. I found a bug on the code that made it use only the last feature in the selected set of features. If using the paper's feature selection string given in the README, this would only use Rosetta features, which matches the result you report. I will write a patch as soon as possible.

Unfortunately, this means you will have to retrain your models from scratch, but the processed data should still be usable without any change.

[zengjianyou]

@hhassan1 @wanyueli Which version of Rosetta you downloaded, can you help me?