search

compomics / DeepLC

DeepLC: Retention time prediction for (modified) peptides using Deep Learning.
https://iomics.ugent.be/deeplc
Apache License 2.0
52 stars 18 forks source link

Training new models #63

Closed KarlClauser closed 11 months ago

KarlClauser commented 11 months ago

Issue #12 from May 2020 response was:

yes it is possible to train new models. The python scripts to retrain are available in "figures_without_models.zip" here: >>https://doi.org/10.5281/zenodo.3706875 In that zip file you can use "run_full_mod.py" to retrain. It should be documented, but if things are unclear please do not hesitate to ask.

I downloaded that zip and can not find "run_full_mod.py" perhaps you mean "run.py"?

Is there a more current version of the new model training capability? The current release contains a file called deeplc/trainl3.py

RobbinBouwmeester commented 11 months ago

Dear Karl,

It should be in the zip file, going to check now. There are multiple zip files, which one did you check?

About training your own models, you can now also have a look at this tutorial that uses a new package (retrainer) to apply either transfer learning or train new models from scratch:

https://proteomicsml.org/tutorials/retentiontime/deeplc-transfer-learning.html

Also looking into the other issue you posted will come back to that soon.

Kind regards,

Robbin

KarlClauser commented 11 months ago

Hi Robbin,

I checked the one indicated: figures_without_models.zip proteomicsml.org looks to be a great resource, I will work from that instead.

RobbinBouwmeester commented 11 months ago

Ah, it seems I posted the wrong zip in the answer... It should be "train_eval.zip".

For now I will close this issue, feel free to reopen it.