windows is not supported yet because pytorch 1.0 for windows has not been released.
Better use linux for now
activate your environment and run
pip install dataclasses
that should resolve that issue.
Can we refer to installing "fast"ai on windows as superslowai
Hello,
i followed steps mentioned above but we get following error when we try to run ml1/lesson1-rf.ipynb. uploading screenshot for your reference.
please let me know if i have missed something.
I was unable to get the symlink to work because the “old” directory is not present when you get the latest pull. I point to fastai and it works fine…so far.
from the docs, the installation method is this
conda install -c pytorch -c fastai fastai pytorch-nightly cuda92
but when I tried to do this, conda returned an error and says that it can’t find the channel, any help?
Hi,
I am trying to install the same following above instructions given in AWS windows(m4.xlarge) environment. I am getting below error.
Please let me know how to resolve the same.
Can we refer to installing "fast"ai on windows as superslowai. i try to import fastai.imports i get the error “ModuleNotFoundError: No module named ‘dataclasses’”.
Houses for Rent in Rawalpindi
First time I executed all the steps pf installation, it ran smooth. I was able to run all commands on lesson1.ipynb
The next day I wanted to check lesson1 again, so I started anaconda prompt, changed directory ‘cd fastai’, ran ‘conda activate fastai’, ran ‘jupyter notebook’
As soon as I started executing the program error popped up No module named ‘fastai.imports’
So I again ran all the steps above, didn’t solve the issue.
What do i do?
did you make a symbolic link archive in the directory of the course notebook to fastai ?
Yes I did
Hi, if anyone can help me to install fast,ai in my laptop highly appreciated. I keep having a error while installing and i ain’t got clue what package am i missing.
support for windows will come when pytorch 1.0 has official support for it
@Kaspar when executing mklink /d fastai …\old\fastai
I am getting warning “Cannot create a file when that file already exists.”
Can you please suggest what should I do to overcome it, I have followed all previous steps as per prof. Jeremy mentioned.
that must be because there is already a folder/symlink called fastai. to replace it you have to delete it first
It seems you make another env in this installation.(is this necessary? I installed fastai without making a new env with the tutorial of a CN fastai site(seems official))
I tried to make symlinks like step 8, but still can not import fastai.transform,
is this module stil exsit in version 1.0 ?
I faced the exact same problem as you, but I found a fix! This is what I run on Anaconda Prompt so I don’t face any problems:
cd fastai
activate fastai
git reset HEAD --hard
cd courses\dl1
del fastai
mklink /d fastai …..\old\fastai
cd ….. (no idea why the formatting is changing this to a ‘…’, just type what is given in the first post)
Whenever I am running command “conda env update”
I get below error,
ChunkedEncodingError(ProtocolError(‘Connection broken: OSError("(10054, ‘WSAECONNRESET’)")’, OSError("(10054, ‘WSAECONNRESET’)")))
Did you face this error when updating env.
So now that PyTorch 1.0 is out with Windows and Cuda10 support, does anyone know the time frame for FastAI to support Windows?
1 Like
@jeremy ,
Great lectures. I’m trying to work on lecture 10 for a NLP classification problem on a windows server that has 32 GB RAM and no NVIDIA GPU. Can you please let know if fast.ai can work on a server with no GPU. I know you have mentioned that NVidia GPU is a must but still asking. Thanks