Thanks for letting me know. That’s fixed now.
Just a small note for others who may want to try to use google colaboratory as a free alternative to trying the library out. It will not work! Colab does not support python 3.7 as of writing this. I will look into other free options for those of us who cannot afford scraping money atm.
It looks like paperspaces free tiers begin in March
1 Like
Came across two more errors when I ran the build_docs.py.
First one was after
converting: 19_callback_mixup.ipynb => docs/callback.mixup.html
converting: 92_notebook_showdoc.ipynb => docs/notebook.showdoc.html
An error occurred while executing the following cell:
------------------
show_doc(show_doc, default_cls_level=2)
------------------
---------------------------------------------------------------------------
NameError Traceback (most recent call last)
<ipython-input-30-f55211237a06> in <module>
----> 1 show_doc(show_doc, default_cls_level=2)
<ipython-input-29-2f3af20125ce> in show_doc(elt, doc_string, name, title_level, disp, default_cls_level)
14 doc = f'<h{title_level} id="{name}" class="doc_header">{name}{source_link}</h{title_level}>'
15 doc += f'\n\n> {args}\n\n' if len(args) > 0 else '\n\n'
---> 16 if doc_string and inspect.getdoc(elt): doc += add_doc_links(inspect.getdoc(elt))
17 if disp: display(Markdown(doc))
18 else: return doc
<ipython-input-13-3a2dc205f4b1> in add_doc_links(text)
3 "Search for doc links for any item between backticks in `text`."
4 def _replace_link(m): return doc_link(m.group(1) or m.group(2))
----> 5 return _re_backticks.sub(_replace_link, text)
<ipython-input-13-3a2dc205f4b1> in _replace_link(m)
2 def add_doc_links(text):
3 "Search for doc links for any item between backticks in `text`."
----> 4 def _replace_link(m): return doc_link(m.group(1) or m.group(2))
5 return _re_backticks.sub(_replace_link, text)
<ipython-input-10-50cf3cb77d8f> in doc_link(name, include_bt)
4 cname = f'`{name}`' if include_bt else name
5 #Link to modules
----> 6 if is_fastai_module(name): return f'[{cname}]({FASTAI_DOCS}/{name}.html)'
7 #Link to fastai functions
8 try_fastai = source_nb(name, is_name=True)
<ipython-input-7-3edfbf61a3cc> in is_fastai_module(name)
3 "Test if `name` is a fastai module."
4 dir_name = os.path.sep.join(name.split('.'))
----> 5 return (Path(_file_).parent.parent/f"{dir_name}.py").exists()
NameError: name '_file_' is not defined
NameError: name '_file_' is not defined
Second one was after
converting: 05_data_core.ipynb => docs/data.core.html
converting: 15_callback_hook.ipynb => docs/callback.hook.html
Traceback (most recent call last):
File "build_docs.py", line 43, in <module>
force_all:Param("Rebuild even notebooks that haven't changed", bool)=False
File "/Users/i077725/Documents/GitHub/fastai_dev/dev/local/script.py", line 34, in call_parse
func(**args.__dict__)
File "build_docs.py", line 46, in main
_make_sidebar()
File "build_docs.py", line 39, in _make_sidebar
open('docs/_data/sidebars/home_sidebar.yml', 'w').write(res_s)
FileNotFoundError: [Errno 2] No such file or directory: 'docs/_data/sidebars/home_sidebar.yml'Yup I found them too as I was getting the docs up and running - both fixed now.
Perhaps we should try to support py36 for now, since py37 isn’t that widely released yet.
1 Like
I’ve now made the v2 docs site available at: http://dev.fast.ai
For instance, here’s a tutorial on the new lower-level data APIs: http://dev.fast.ai/pets.tutorial.html
I suspect the notebooks are still the best way to learn what’s in the library so far - but the docs site might be useful too due to the hyperlinks.
8 Likes
I’ll wait patiently then 
ndim was added to pytorch since 1.2.0 torch.Tensor — PyTorch 2.4 documentation
Therefore
Tensor.ndim = property(lambda x: x.dim())
monkey patching is no longer required: pytorch>=1.2.0 is in environment.yml
PL submitted: Update imports.py by kdorichev · Pull Request #158 · fastai/fastai_dev · GitHub
1 Like
Good point. Guess who suggested that addition to PyTorch 
9 Likes
@jeremy how much overhead would need to be done for 3.6 capability? Or what specific python 3.7 functionalities are being used? (If that’s a simple answer even)
I can try to help out with that.
I have no idea. Let’s try to find out!
1 Like
I’ll go ahead and make a new forum post for where differences are and how to fix them in hopes of a team effort with this 
I’m going to take a quick look at it now. It might be an easy fix.
1 Like
Looks to be just the MethodWrapperType
1 Like
external.py is not generating a config.yml when we run untar_data, looking into it now. Temporary fix is copying the Config() from v1
OK all works with py36 now. Also added some basic install and test-running steps to the readme.
There’s a few failing tests still, which I’ll look at.
1 Like
Thanks for the 3.6! Did you intend to use the same Config() class from v1? Or were you wanting to switch them. I noticed if it does not exist then config if just blank. Eg:
config_path = Path(os.getenv('FASTAI_HOME', '~/.fastai')).expanduser()
config_file = config_path/'config.yml'
if config_file.exists():
with open(config_file, 'r') as yaml_file:
config = yaml.safe_load(yaml_file)
if 'version' in config and config['version'] == 1: return config
else: config = {} <- Here we are not creating the config file
I assume we’re hoping to use the same config.yml that’s in v1, unless that’s incompatible in some way.
And yes, if it’s missing, we should create it!
1 Like
I’ll put in the PR! It works
1 Like
Any thought to optional labeled test datasets? I know this was a request for v1. If it were to work how you would like I believe it would fall along the following: defaultly label them flat unless explicitly stated. As the unlabeled test set helps for Kaggle, a labeled test set helps for research.
(something of which I have been doing quite often)
And PR posted