Recently we've added a possibility to handle custom events, so we need to keep "Concepts" documentation updated.
There is also a bad docstring rendering of parameters on methods
@vfdev-5 : also there is a minor bug in the ModelCheckpoint docstring, the save_as_state_dict parameter is not in the list.
Please can you give me your policy how I should proceed and then I will correct it (I do not have much experience in contributing to other's work). Thanks in advance!
@rpatrik96 welcome on board of open-source :) Thanks for pointing out other undocumented parts.
In brief how to fix a docstrings:
cd docs/build/ && python3 -m http.server 8080 to display built docs locally. Like this you can check if fixes are working.And for a general context take a look also at CONTRIBUTING
Do not hesitate to ask details on the points if you have other question.
HTH
@vfdev-5 thanks for the detailed help!
I have fixed the missing argument, and as a warmup I have created my first pull request :).
EDIT: travis has been failed on my PR, the only thing which has failed was the _mnist_with_visdom_ example. Could you help me how I can fix this issue?
A small future reference for other newbies regarding sphinx: if you have Python/Anaconda/whatever in your PATH, then sphinx _will find the original and not the forked files_. As a fix, read https://github.com/sphinx-doc/sphinx/issues/4317.
Summary: modify the docs/source/conf.py by uncommenting the few first lines, thus getting:
sys.path.insert(0, os.path.abspath("/insert/the/path/to/the/cloned-forked/repo"))
Sorry for being hyperactive, but I have noticed something regarding _Events_:
If I have the following callback:
trainer.add_event_handler(Events.ITERATION_COMPLETED, my_callback, model)
Then model.training will be False. I assume that this is with intention (but for me, at first it was not trivial), so the question comes: where can I find this description in the docs? If it is not included, I will happily add it to the docs.
@rpatrik96 thanks for your help.
I'll review your PR in the coming time, there is a priority on fixing failing travis. There is a problem with Visdom, see #241
sphinx will find the original and not the forked files
I did not get what you call original and forked files. However, I think better approach not to modify docs/source/conf.py file by commenting otherwise beginner developer can forget to revert such changes...
Concerning model.training=False I need to check and analyse this, thanks for pointing out :)
Hi @rpatrik96 can you provide a small snippet of code to demonstrate the model.training==False issue. Could you start a new github issue with this so we can track it there? Thanks
Hi @alykhantejani I did some refactoring after facing the issue and it cannot be reproduced, so I assume I have made a bug, there is no problem at all :).
I reopen this because I did not yet fixed the problems cited in the issue comment.
Fixed in #247
Most helpful comment
@vfdev-5 : also there is a minor bug in the ModelCheckpoint docstring, the save_as_state_dict parameter is not in the list.
Please can you give me your policy how I should proceed and then I will correct it (I do not have much experience in contributing to other's work). Thanks in advance!