igorsyl
(Igor Sylvester)
1
I’m going over https://www.fast.ai/2020/02/13/fastai-A-Layered-API-for-Deep-Learning/ and I’m unable to run the first vision code example:
>>> import fastai
>>> fastai.__version__
'1.0.60'
>>> from fastai.vision.all import *
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
<ipython-input-2-5635efd54484> in <module>
----> 1 from fastai.vision.all import *
ModuleNotFoundError: No module named 'fastai.vision.all'
muellerzr
(Zachary Mueller)
2
This should be from fastai2.vision.all
(as it’s on the new v2 version of the library)
@jeremy pinging you on this (should this be adjusted or is the end goal for it to be like this?)
igorsyl
(Igor Sylvester)
3
Thanks for your quick reply!
I did try fastai2 before posting and the code breaks here:
dls = ImageDataLoaders.from_name_re(
...
AssertionError: Failed to find "re.compile('/([^/]+)_\\d+.jpg$')" in "german_shorthaired_9.jpg"
After fixing the regex, then the code breaks here:
learn.fit_one_cycle(4)
...
~/miniconda3/envs/py37/lib/python3.7/site-packages/fastcore/foundation.py in __getattr__(self, k)
221 attr = getattr(self,self._default,None)
222 if attr is not None: return getattr(attr, k)
--> 223 raise AttributeError(k)
224 def __dir__(self): return custom_dir(self, self._dir() if self._xtra is None else self._dir())
225 # def __getstate__(self): return self.__dict__
AttributeError: _IterableDataset_len_called
muellerzr
(Zachary Mueller)
4
The _IterableDataset
issue is due to the most recent torchvision. It’s recommended to use torch 1.3.1
and torchvision 0.4.2
(along with Pillow 6.2.1
)
1 Like
igorsyl
(Igor Sylvester)
5
Those versions work, it trained in 25 secs/epoch!
I just had to change:
pat = r'/([^/]+)_\d+.jpg
to
pat = r'([^/]+)_\d+.jpg
Does it make sense to add a function fastai.check_lib_versions()
to check for the recommended library versions?
jeremy
(Jeremy Howard)
6
Yes the paper is written based on how things will be after fastai2 is released. At that time it’ll just be fastai
.
1 Like
jeremy
(Jeremy Howard)
7
BTW best place to discuss fastai2 and the paper is: #fastai-users:fastai-v2
1 Like