1edv
commented
2 years ago Hi,
The "Generate standard scaler for positive and negative selection experiments using the extreme cases of the designed sequences (Done together since designed sequences are from the same file )" section in the https://github.com/1edv/evolution/blob/master/manuscript_code/model/tpu_model/data_processing.ipynb file describes how the scaler was trained. (The scaler can also be simply trained on data from the whole training data Sort-seq experiment as well - instead of the specific dataset we used here.)
Best, Eeshit
PS - Note that the scaler isn't strictly necessary here - you should be able to get equivalent results even without the scaler.
Hi, in your code you seem to be scaling the target values before training, however I am struggling to find reference to this procedure in the manuscript? The only mention of scaling I can find appears to be for purely visualisation purposes (section titled Visualizing promoter fitness landscapes). Would you mind pointing me to where I can find more detail about the procedure and to how the scaler was trained?
Thank you!