bakunowski / listen_like_a_bat Goto Github PK

Currently getting an item from the dataset returns whole .csv file - each containing 101 impulse responses. The IR is chosen randomly.

We want each item to be one impulse response. The fix would be to somehow add all impulse responses to the .csv file containing names of files with IRs and labels.

It would be a good idea to find a more elegant solution for this...

One giant .csv file dependent on local directory structure is far from perfect.

Add one hot encoding for classes of each plant as not much can be done with classes 1-9 etc