Setting up windows environment, torch?

I am trying to set up a windows environment to run these notebooks in. I have used windows before with anaconda package and jupyter notebook and it works fine, but it looks like there are a bunch of other libraries that don’t come with the anaconda install that are used.

I was able to install a lot of the libraries to work my way through the import errors, but the one I am stuck on is torch/torchvision. Looking into this it looks like there is not a windows version of this.
Does that sound correct to anyone?

I will probably have to setup this environment with the windows ubuntu interface, like instructed during the workshop the weekend before last.

Hi Kyler, Jeremy was quite clear about the convenience of NOT using our versions of windows installed Anaconda. Sure it is hard, many things to get used to for us “Windows-ers”, but I begin to almost have fun with it. (Ok, not yet, but I see the day will come…)

Anyway… if you are convinced to try the windows way and Pytorch is the only thing that its in your way, maybe I have good news for you, previously to the course I was able to make an environment work with GPU support just with this line of code: conda install -c peterjc123 pytorch (I also installed torchvision, dont remember if with conda or pip but it worked also fine)

So, good luck with it, even if I plan to do the course 100% bash ubuntu it will be interesting to know if its possible to make the environment work in our gpu-windows versions.

Hi Miguel,
Thanks for the response. I guess I don’t remember Jeremy mentioning that windows is going to be painful. I was just trying it because I have already setup some of the tools and they are well integrated with the built in command prompt which I am used to.
I have Ubuntu setup already too from the workshop, it will be a steeper learning curve for me but that’s fine. I have nvidia cards in my personal computers already so I am curious how they handle the DL tasks. I’ll probably just end up doing the course work on AWS or Crestle anyway though.

1 Like

In my experience doing deep learning or using any open source package on window is too much work.
If you want to save time and spend that time on learning deep learning then it’s better to use Linux.
Using Linux will help you in long term. Very few people use window for deep learning in industry. And if you want to deploy the model on server then Linux has best support. Window is too much struggle.

1 Like

Guys, the thread Windows 10 Installation Notes completely moots this thread.

2 Likes