Multi-label multi-class task

Part 1 (2019)
1

Are there any ideas how to use fast.ai for multi-label multi-classification task?
In my DataFrame I have labels as Column names and classes as values:
| |image|digit1|digit2|digit3|digit4|digit5|
|0|data/train/24689|2|5|10|10|10|
|1|data/train/13196|1|7|9|10|10|
|2|data/train/16036|6|3|10|10|10|
|3|data/train/18151|5|0|10|10|10|
|4|data/train/18513|9|1|10|10|10|

  1. How to load this into ImageDataBunch?
    I try this but no success:
    src = (ImageList.from_df(df, path, suffix='.png')
    .split_by_rand_pct(0.2)
    .label_from_df(cols=['digit1','digit2','digit3','digit4', 'digit5'], label_cls=MultiCategoryList, one_hot=True))
  2. What loss functions are best for this case?
2

Did you end up solving this? I am having a similar issue

3

No I didn’t. Still trying to understand what are the possibilities.