ms2deepscore
ms2deepscore copied to clipboard
matchms
Deep learning similarity measure for comparing MS/MS spectra with respect to their chemical similarity
I am running the new multi_inputs pipeline, when running MS2Deepscore, you require to add multi_inputs=True. The error you get when forgetting to set this to True is confusing: /ms2deepscore/MS2DeepScore.py", line...
We should provide our most recent/performant model on zenodo with a link on the readme in this repo. Then, we should add a GitHub workflow to run a test using...
Currently, the data generator has to pick random pairs of spectra during training (and testing) which fall into specific Tanimoto score bins. This is very expensive computationally because for each...
The current version of ms2deepscore has tensorflow in the install_requires which prevents successful installation on m1 macbooks. I've managed to get ms2deepscore to install and run on my M1 MacBook...
Currently, a large array of reference scores (e.g. Tanimoto) is used during training. The data format can be downscaled to float16 to save memory (now: float32 or float64?).
I ran a model training with early stopping and than got unexpected behavior of `model.save()` since that will now save the final model after training, hence ignoring any `save_best_only=True` setting.
Other ideas that came up for additional discussion or next steps: - Experiment with binning → “smart” binning based on peak importance (e.g. something like tfidf?). - We played around...
We are currently working our way towards matchms 1.0 which is likely to come with a few API-breaking changes. This issue is thus a TODO note. In addition, we will...
Now we used fixed size bins, but other choices might work even better. For instance peak sizes that increase with increasing m/z (to reflect higher ppm precision for low m/z...