Néstor Nápoles López
Néstor Nápoles López
Currently, a lot of things show up in the console: - `tf`-related logs - Intermediate rntxt files - Other logging information Ideally, everything the standalone scripts provide to the user...
They used to be always `0.0`, which means the chordtones-nonchordtones, bass, and alignment are always consistent between the content of the synthetic scores and the annotations (as they should). Recently,...
It would be nice if one could include only one dataset and the syntax was equivalent for either enabling a single dataset or the entire aggregated dataset. It'd also be...
Currently, it is hardcoded. In that part of the workflow, it is irrelevant what the `cli` defaults say, the sequence length is the shape of the input in the model.
All csv annotations include any padding done to the input. This is misleading because the padding shows usually as a `Cb` key with some other nonsensical information. I know this...
Make sure that every part of the workflow (tsv generation, npz, and training) can reliably be initialized with a random seed. Ensuring reproducible experiments will help with network debugging, paper...
The quality metrics would benefit from specific unit tests. For example, different instances of *wrong inversion* or *non-chord tones*. I imagine some of these are already present in the Haydn...
None of these have. The difficult thing is to save those numpy vectors for ground truth.
I was wondering if you are familiar with the RNA syntax from [David Huron](https://www.humdrum.org/rep/harm/) as an alternative for the spelling convention. I am not quite literate in fonts and I...
@JoyfulGen mentioned an issue about an `Invalid MEI` output in `rodan-staging` Neon. This error/warning disappeared after doing any trivial change in the editor (e.g., slightly moving the position of an...