-
-
Notifications
You must be signed in to change notification settings - Fork 149
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Diffused transformations in docs to preprocessing, DA, and utilities #1061
base: master
Are you sure you want to change the base?
Conversation
Would it make sense to have a subsection as transformations with subsubsections as (i) "Preprocessing", (ii) "Data Augmentations", and (iii) "Utilities" as shown below: (Why?) ivadomed at its core uses PyTorch and this aligns with |
@kanishk16 It would definitely make sense, thanks for your suggestion! Another supporting argument for your idea is that the key used in the config file is A counter-argument for this idea is non-consistency with "Preprocessing" heading as @jcohenadad noted here, but I would be in favor of your idea because it solves more problems. |
@uzaymacar Thanks for highlighting this, I completely missed this. Apart from this trade-off, another plausible solution could be to have a different key ( |
@kanishk16 could you please show us in a 'tree'-like display what your TOC suggestion would look like? |
Thanks, @uzaymacar for restructuring the docs! It is much required! I am in favor of keeping the name "Transformations" as it is currently in the TOC and it also corresponds well with the keys that we have in the config file. However, as "Preprocessing" shouldn't technically be under "Transformations", I agree with @kanishk16's comment on creating new key-value pairs for preprocessing in the config file and having another subsection (I think preprocessing keys are important enough to justify a new subsection!). At the risk of extending the discussion, here are my 2 more cents on this:
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would personally rename "Configuration File" into "Usage"
currently, "Usage" is pretty light and only has the CLI info, whereas Configuration file should be part of Usage (moreover, some of the params could also be called via the CLI, right?).
But maybe that should be part of another PR?
Thanks all for the excellent feedback! I would like to pick a few points to handle in separate issues + PRs if possible:
Moving onto this PR, IIUC currently we have two proposals on top of the current state of the branch:
I was in favor of first option because all of preprocessing, data augmentations, and utility are implemented as and are transformations: (a) being implemented in How can we fuse the two? LMK if I have missed anything! Perhaps first taking care of the tangential but excellent points mentioned at the beginning might shine some light on this discussion. |
Very interesting discussion! Regarding Usage, Configuration File and Scripts:
I would be in favor of keeping the term and section "configuration file" but reorganizing the "Usage" section as: About NumpyToTensor:
As per #818, #826 and #830, the NumpyToTensor key is not necessary in the config file and is applied directly in the code here (for transforms that are not preprocessing). The code is retro-compatible for config files that contains the key here (the key is ignored). About preprocessing:
I'm less in favor of this one because preprocessing transforms ARE transforms (same arguments as Uzay for option 1). About the doc hierarchy:
Because of my previous point, I will suggest a third option.
|
Latest commit 3035fa5 implements |
Pull Request Test Coverage Report for Build 1815266603
💛 - Coveralls |
from @mariehbourget
👍 from @uzaymacar
where would 'postprocessing' be located? I would suggest using this more intuitive display to get the discussion going:
|
Just to clarify: "Transformations" is nested under "Configuration file" with all the other config parameters (including Postprocessing), see updated structure below. As for "Postprocessing", I don't think it should be inside "Transformations" because:
|
from #1061 (comment), i feel that an h3 nesting for 'Transformations' (which is a pretty important section) will likely be missed. Maybe we should reconsider some of the proposed changes about 'transformations'? And if we wish to merge 'scripts' with usage, if we consider that
to sum up the proposed changes are:
|
Checklist
GitHub
PR contents
Description
This PR diffuses the "Transformations" heading into (i) "Preprocessing", (ii) "Data Augmentations", and (iii) "Utility Transformations" headings at the same level.
In doing so, I also alphabetized the transformations under the new "Data Augmentations" as previously mentioned here.
In the reviewing of this PR, it would be great to get some feedback for
"applied_to"
,"dataset_type"
, etc.) are repeated in all three headings which is not idealLinked issues
Resolves #1059