Conversation
I fix bugs and now it works perfectly
Add fine tuning class, dataset class, and make changes to doc string.
Remove TensorFlow and add PyTorch.
Remove redundancy.
PyTorch version of DeepLC
Cleanup and preliminary refactoring of feature extractor
- Remove ensemble-model mode (different kernel sizes). - Split of model loading to separate function. - Make dataset take peptidoforms instead of PSMs. - Get unique peptidoforms before predicting and keep inverse index for mapping back predictions to input PSM list.
…into feature/new-api
Initial work for new 4.0 API
There was a problem hiding this comment.
TODO - meant to take in a PSM list and calibrate separate runs to a single run
There was a problem hiding this comment.
Skip GUI for now and focus mostly on streamlit/webserver?
There was a problem hiding this comment.
Whowa, pytorch models are so much smaller
There was a problem hiding this comment.
It seems that some functionality is still only available through "combined" functions. Such as "predict_and_calibrate" shouldn't there also be a "calibrate" functionality?
There was a problem hiding this comment.
Add CCS feature logic can be removed, check if n-terminal (or c-terminal I guess) modifications are now handled correctly. We used to put them on either the first or last AA without regards to where they were located (n-terminal or residual). I think it is also fine to add an additional feature field in the matrix that allocates them specifically on the termini, but this would require retraining, so lets keep it as is for now...
| raise CalibrationError("Source values have zero or invalid range; cannot calibrate.") | ||
|
|
||
| boundaries = np.linspace(cal_min, cal_max, self.number_of_splits + 1, dtype=np.float32) | ||
| starts: np.ndarray = np.searchsorted(source, boundaries[:-1], side="left") # type: ignore[var-annotated] |
There was a problem hiding this comment.
What is this type?
There was a problem hiding this comment.
CCS feature logic can be removed
4 participants
No description provided.