(@ecdrid , that’s what I would like not to do if its possible, not to loose things already there )
I @miguel_perez unfortunately once a machine is made we cannot overwrite it without deleting all data. I would recommend creating a new machine or you could also upgrade Pytorch to 0.3 on your existing machine which is what caused the issue.
Unfortunately that requires a little bit of legwork.
Another option would be to shoot an email to support@paperspace.com and they can convert you to a “team” which will give you access to shared drives. Then your data can live on a single drive that is shared among many instances.
Sorry none of these are ideal We are really working hard to streamline this process and should have some new announcements soon.
@dillon, thanks for your answer, its ok, I will drop my current machine and create a new one, it is not such a big deal
I had a small suggestion for the learn.fit() functions, these take about 5-10 minutes on p2 instances during which I read some other stuff. Is it possible to make a notification sound once the fitting finishes to avoid keeping them in idle states?
@init_27 I havent tryied it but maybe you can check this out : https://gist.github.com/iandanforth/fd9c98dc485592e354974e394ff6ae28
To verify that you’ve installed pillow-simd (a faster version of pillow) you can enter the following in Python:
import PIL
PIL.__version__
And you should see a version number (at or above 4.3.0) and most importantly something like “post0” at the end:
Out[2]:
'4.3.0.post0'
And not:
Out[2]:
'4.3.0'
(I’m storing this piece of knowledge here for my future self and anyone else who can use it.)
It’s not important any more however, since we use opencv, not Pillow.
All,
This might help someone having problems updating the conda env. Had been a while I hadn’t updated my environment. On trying to update using the recommended command:
conda env update
I got a rather unhelpful message:
NoPackagesFoundError: Dependencies missing in current linux-64 channels:
- pytorch >=0.2.0 -> mkl >=2018
- torchvision >=0.1.9 -> pytorch >=0.3 -> mkl >=2018
Panicked around a bit. Finally, the following worked for me:
- Activate the conda environment using:
source activate fastai
- Then update conda:
conda update conda
- Then update the env:
conda env update
Oh, thanks… even i encountered this and was wondering why… BTW, is there any material that can compare openCV to PIL?