I am running Pytorch (.3) successfully on Win 10
Were you able to follow along with the first lesson?
No problem, Richard. I’ll keep you posted, should I make any further progress.
I’m fairly convinced I should be able to configure a workstation for deep learning no matter what libraries I use, since that’s a valuable skill to learn by its own.
So you managed to get pytorch working properly on windows 10, which is substantial! Do you have other libraries installed and/or other versions of cuda?
Did I manage to run lesson1 nbs? Yes, but with cpu-only pytorch, wich leaves me dissatisfied.
I could switch to linux, but for a number of reasons, I’d rather follow with windows 10, primarily since I have other things to do on that windows box, and rebooting continuously is a nonoption.
Thanks for your feedback Aditya.
Do you have other versions of cuda installed and/or other DL libraries in other anaconda envs? Thanks!
Yep Cuda 8 and cudnn 6.0
Have Keras, Tensorflow, Pytorch in my environment
Had installed Pytorch from Anaconda Peterjc channel
Could you attach the output of “conda list” for your env? Furthermore, I suppose you are using Cuda 8.0 for both TF and Pytorch, am I right?
I am already in week 4 of Part 1 (v1) course. Shall I first finish the V1 and then come back to V2 or is it a better idea to restart the V2 course from the beginning?
Will update when I get access to my laptop…
Currently not at home…
Thanks, I received your list by email.
You got your environment all setup Aditya?
Could you possibly help me setup a working environment so that I could follow along with the lessons?
On Win 10
FYI, there is also a Beginner forum here.
I have two tips sheets which may be helpful for beginners:
If you have a question on what directory you can put your project notebook, there is info in there.
@jeremy I cannot find the post on “How to Ask for Help”. Perhaps we can add it to main wiki?
Now, in order to update instructions to get windows users up and running, this is what worked for me (windows 10 box, with gtx 1070):
a) Create a new anaconda env with python 3.6 and some additional packages:
>conda create -n fastai python=3.6 numpy cython statsmodels opencv
b) Install pytorch for cuda. If you already have a keras/TF setup which works only with cuda 8.0, and want to avoid headaches, you would rather install the cuda 8.0 version:
>conda install -c peterjc123 pytorch cuda80
However, keep us posted should you manage a successful installation of the cuda9 version alongside TF and cuda8
>pip install fastai
be patient, that will install a lot of stuff.
d) Now you are done, but you may want to install the ipython kernel in order to use lessons’ notebooks and do your own experiments. In my case:
>python -m ipykernel install --user --name fastai --display-name "fastAI custom"
That would be all.
Now, one may want to log in into your workstation remotely. In that case, there are various options, for example:
- Using remote desktop (mind that it’s slow if you have a slow or high-latency connection)
- Just leaving the notebook server on (unpractical: you would not be able to run administration tasks)
- Using “linux on windows” with tmux (install LoW from windows store)
Hope this helps.
Footnote: created a new topic on the beginner forum.
Beginners, you may find this helpful:
Where can I put my Jupter Notebook?
(I also added it to the Wiki for Lesson 1)
The fast library simplifies things a lot compared to part 1 v1; for a non-programmer like me, v2 is actually easier than v1. Hope to become a programmer by the time v3 is published
- After v2 is launched on the course website, will the v1 notebooks and videos still be available?
- Is it possible to create a video example to explain the core advise that the time is best spent playing with the application side of the code rather than the theory and fundamentals; for example, I do not understand the difference when @jeremy says I want to work in bash and not in the shell or something like that, but I should ignore this for now because it is not on my shortest path to get a feel for the model
- Also, is there a way to show how many parameters the model is trying to fit, when we are trying to determine if it needs more data or improved optimizer etc?
Everything will remain with just some shuffling of directories like putting all notebooks or n a directory named
part1v2or something equivalent…
if we have the model named as
mthen in Pytorch we can do
m.parameters(). This will display Everything…(it’s a bit ugly so I tried it on small networks to make sense out of it…)
Tensorboard for Pytorch…