search

brain-research / deep-molecular-massspec

Mass Spectrometry for Small Molecules using Deep Learning
Apache License 2.0
110 stars 41 forks source link

How to retrain #16

Closed yamaguchiraf closed 4 years ago

yamaguchiraf commented 4 years ago

Can I retrain with my own MS SDF?

jnwei-zz commented 4 years ago

Yes it should be possible

On Thu, Mar 5, 2020, 9:10 PM yamaguchiraf notifications@github.com wrote:

Can I retrain with my own MS SDF?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/brain-research/deep-molecular-massspec/issues/16?email_source=notifications&email_token=ADT3XUA246Y2GL3FV4WBCALRGBLRFA5CNFSM4LCW7KDKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IS67KOA, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADT3XUGWAVW25M2SFN5EFLDRGBLRFANCNFSM4LCW7KDA .

jnwei-zz commented 4 years ago

Let me know if you run into any issues; feel free to reopen this comment if so.

dchek001 commented 4 years ago

I have a somewhat related and perhaps rather naive question. My apologies - I am new to DNN. Retraining with your own MS data would not change the underlying NEIMS architecture and only re-optimize the weights, probably only in the output layer? Is that right? Thanks!

jnwei-zz commented 4 years ago

Right. You are also welcome to change the hyperparameters of the model by specifying hyperparameters with the hparams flag. For example, to change the number of hidden units in the DNN to 1500, you can use:

python molecule_estimator.py --hparams=num_hidden_units=1500

dchek001 commented 4 years ago

Thanks! Much appreciated.