Fastai.tabular module not found on kaggle kernal

how to fix it

Kaggle Kernels don’t have fastai v1 yet

how can i install fastai v1

I don’t think anyone has figured out how to get fastai v1 working on Kernels yet. Even if you install the correct version using !pip install inside the notebook, it will still have the wrong version of pytorch. I think we have to wait until pytorch 1.0 has a stable release and then get Kaggle to update the Docker image for Kernels.

1 Like

There are instructions on how to make it work. It worked but so sloooow that you couldn’t work with it really…

(Install latest version of fastai, at the time it was 1.0.6 :slight_smile:)

  1. Enable the Internet & GPU.
  2. Execute

!pip install --upgrade pip
!pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cu92/torch_nightly.html
!pip install fastai==1.0.6

  1. Ctrl+shift+p and select “confirm restart kernel” This will restart the Jupyter Kernel instance and reload the installed libraries.
  2. Wait until the Jupyter Kernel finishes the restart
  3. import fastai shows the correct version.

Plus this fix RuntimeError: DataLoader worker (pid 137) is killed by signal: Bus error

setting num_workers=0 here:

data = ImageDataBunch.from_folder(
path,
ds_tfms=get_transforms(),
tfms=imagenet_norm,
size=256,
num_workers=0
)

2 Likes

is this still working?

i did not try it recently. if you try it let us know if it works :slight_smile: thanks!

Fastai v1 should be available on kaggle kernels soon https://github.com/Kaggle/docker-python/issues/397

1 Like

as per quote…

2 Likes