Fastai library issues (Part 1 - 2018)

Part 1 (2018)
1

I’m having issues loading fastai library. I’m using a linux P2xLarge EC2 instance. In particular, running

from fastai.transforms import *

gives me the error below:

ile "fastai/transforms.py", line 12
    r,c,*_ = im.shape
        ^
SyntaxError: invalid syntax

I suspect the issue might be related to Pillow. I was able to pip3 install Image with no problem. However, (sudo) pip3 install Pillow returns the error below:

Command "/usr/local/bin/python3.6 -u -c "import setuptools, tokenize;__file__='/tmp/pip-build-ybtlqb2o/Pillow/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" install --record /tmp/pip-bqhd9q9o-record/install-record.txt --single-version-externally-managed --compile --install-scripts=/usr/bin" failed with error code 1 in /tmp/pip-build-ybtlqb2o/Pillow/

Any solution to this?

2

Looks like you’re using an older version of Python. We require minimum 3.6. Use the recommended conda installation method to ensure you have everything you need.

3

Thanks @jeremy for your help and for this great MOOC. I’m actually using Python 3.6. Will use conda instead then.

4

Ideally, use the fastai AMI. Here’s a great walkthru from @reshama https://github.com/reshamas/fastai_deeplearn_part1/blob/master/tools/aws_ami_gpu_setup.md

5

Thanks @jeremy for this great tutorial, I just started working on it.

I have installed the requisite libraries in Mac OS 10.13.3 and when I try to run the following line on lesson1.ipynb:

arch=resnet34
data = ImageClassifierData.from_paths(PATH, tfms=tfms_from_model(arch, sz))
learn = ConvLearner.pretrained(arch, data, precompute=True)
learn.fit(0.01, 6)

I got the following warning and it crashed my system.

fastai/courses/dl1/fastai/core.py:23: UserWarning: volatile was removed and now has no effect. Use with torch.no_grad(): instead.
x = Variable(T(x), volatile=volatile, requires_grad=requires_grad)

I raised this issue with PyTorch forum and here is the reply I received:
Are you sure you’re not running out of memory? volatile greatly decreases the usage, but was deprecated in favor of torch.no_grad(). Try wrapping the code that uses x in this context manager.

I also tried reducing the variable sz=100, then it worked. But the accuracy I am getting is only 95%. Is there any fix for this?

6

@shankarrajus sounds like you’re not using pytorch 0.3 ?

7

I had the same problem and, as @jeremy said, due to the python version. PIL downgrades python to 2.7. As @sepehr mentioned, pillow should be used instead of PIL. These are the commands that made my conda environment work properly:

conda install nb_conda matplotlib pillow bcolz scipy opencv seaborn
pip install graphviz sklearn_pandas isoweek pandas_summary
conda install tqdm
conda install pytorch torchvision cuda90 -c pytorch
pip install torchtext

The instruction “conda install pytorch torchvision cuda90 -c pytorch” it’s generated by http://pytorch.org/ and may differ with your configuration.

Hope it helps.

1 Like
8

Facing the same problem how do you resolve this? really annoyed

1 Like
9

i have python 3.6 when i do which python it shows 3.6

10

I am having the same issue.
I have tried both python 3.5 and 3.6. None of them solved this.

I can not move on with this great lecture series.

11

Thank you,
It is now fixed with the following ticket over github : https://github.com/fastai/fastai/pull/352

12

Please, I can’t import these lines of code
from fastai.transforms import *
from fastai.conv_learner import *
from fastai.model import *
from fastai.dataset import *
from fastai.sgdr import *
from fastai.plots import *
I am using anaconda on my windows 10 laptop