David Nicholson
David Nicholson
to DATA section of config? that defaults to True, since so far we are almost always dealing with annotations that have silent gaps pros: - no need to determine programatically...
as a possible fix for #301 where @SamuelBrudner seems to be hitting an OOM error, and #257 currently we just feed the entire "spectrogram windows batch" into the network https://github.com/NickleDave/vak/blob/e01cd2cf1b85106390fd74c6675f93c97061c9a5/src/vak/engine/model.py#L309...
from this blog post seems to me like we could frame problem of splitting dataset into splits of specified durations, all containing same set of labels ,as a subset cover...
instead of cryptic TOML error about escape characters
@yardencsGitHub reports that the start of `learncurve` and `train` is very slow during start-up for him We noticed it is right after this log statement: ``` f'Size of each timebin...
if `data_dir` option in [PREP] section of config.toml ends in `/**`, treat this as "recursive glob", and search recursively in `data_dir` for audio / spectrogram / annotation files when `vak...
have multiple configs for all commands: train, eval, predict should cover different combinations of training data formats: * any type of audio (wav, cbin) * any type of annotation format...