[x] Get rid of suffix, instead use the data_lying or data_sitting namespace, and use all files in the folder.
[x] Make sure all identifiers are explicit, i.e. video_id, patient_id
[x] Reorganize code into data generator classes (see this tutorial, and these examples ) We iterate over the (or a selection of) rows in the scores file, read in the corresponding video file, then generate 1 training example at a time. This makes the process from reading in data to feeding it to the model more explicit. We can create multiple generator classes for different inputs (i.e. raw & hand-crafted features). And use inheritance for code shared across data generators.
[x] Create a folder called notebooks, in it create notebooks for different experiments (think of it as different chapters of the results section in a papr), i.e. 01-deep-learning-multiple-outputs-one-model . The shared code should be factored out as much as possible. This will be simpler to understand than having one notebook with everything we ever tried in it (and actually a lot that we tried is not in it anymore).
suffix
, instead use thedata_lying
ordata_sitting
namespace, and use all files in the folder.video_id
,patient_id