Documentation [Help needed]

The documentation in fastai v2 is not there yet and as we have no implemented every main feature from v1, this is now our main area of focus (it should even have been finished for the new course but confinement with (sometimes crazy) kids happened…).

The main help we would love to get from the community is not to actually write the documentation, but look at the notebooks we have done so far and check everything looks nicely explained, spot typos, and report either in this topic or GitHub issues that is still unclear.

The documentation in itself is here and as you may know, it’s written directly form the notebooks in the main repo. If you want to make a PR, please only include your changes to the notebook but don’t build the documentation as it makes for a large diff to review.

You can check in the list below which notebooks have been treated by us and on which you can focus (they are in the order we expect to treat them):

Notebook name Status :white_check_mark:
38_tutorial.text :white_check_mark:
44_tutorial.tabular :white_check_mark: :white_check_mark:
21_vision.learner :white_check_mark: :white_check_mark:
37_text.learner :white_check_mark: :white_check_mark:
43_tabular.learner :white_check_mark:
45_collab :white_check_mark:
46_tutorial.collab :white_check_mark:
13a_learner :white_check_mark:
50_tutorial.datablock :white_check_mark:
10_tutorial.pets :white_check_mark:
22_tutorial.imagenette :white_check_mark:
35_tutorial.wikitext :x:
24_tutorial.siamese :x:
09_vision.augment :x:
13b_metrics :x:
20_interpret :x:
06_data.block :x:
05_data.transforms :x:
04_data.external :x:
12_optimizer :x:
13_callback.core :x:
14_callback.schedule :x: :x:
15_callback.hook :x:
16_callback.progress :x:
17_callback.tracker :x:
18_callback.fp16 :x: :x:
19_callback.mixup :x:
34_callback.rnn :x:
07_vision.core :x:
30_text.core :x:
42_tabular.model :x:
40_tabular.core :x:
03_data.core :x:
02_data.load :x:
01_layers :x:
11_vision.models.xresnet :x:
15a_vision.models.unet :x:
32_text.models.awdlstm :x:
33_text.models.core :x:
00_torch_core :x:
09b_vision.utils :x:
20a_distributed :x:
36_text.models.qrnn :x:
09c_vision.widgets :x:
61_tutorial.medical_imaging :x:
60_medical.imaging :x:
65_medical.text :x:
70_callback.wandb :x:
71_callback.tensorboard :x:
72_callback.neptune :x:

If I check a notebook, and everything seems okay, how should I report that?

By not doing anything :wink:


Hello Sylvain, here is what I have read so far. My comments are maybe not relevant since I am begining to go from fastai v1 to fastai v2 (so if someone wants to double check the doc I think it’s a good idea) , and also I am not a native english speaker.

Tabular : :

In it could be maybe interesting to say what are the ‘procs’ (it’s the only definition you that we don’t see).

And also you have put 2 times : show_doc(TabularDataLoaders.from_df) , maybe it’s intentional.

42_tabular.model : looks good

43_tabular.learner :looks good

45_collab : looks good

Is my feedback OK or do you want something else?

I believe he wants us to write in the notebooks themselves then make a PR to the fastai2 repo (but don’t build the docs) :wink:


Some docs are missing from, such as callback.tensorboard and callback.neptune. The corresponding html files are present at the docs/ folder in the fastai2 repository, but the sidebar.json was not updated.

I am looking at the callbacks notebooks and it is all over the space. @sgugger, can you provide a template on how these notebooks should be documented? so we can modify them to a consistent format and new callbacks could be written with this specific format.