Getting error on Kaggle kernel when using cnn_learner function

Part 1 (2019)
1

When I ran the command learn.lr_find the kaggle kernel throughs an error OSError: [Errno 30] Read-only file system: ‘…/input/purge-tmp.pkl.

So I saw the forum, there it was mentioned to update the fastai library and add the argument model_dir in the function cnn_learner. I follow the steps but still I am getting the following error:

TypeError: __init__() got an unexpected keyword argument 'modeldir'

I am using ‘1.0.61’ version of fastai.

2

Do not add it as argument. Set it after making an instance of cnn_learner.
Like this:
learn = cnn_learner(data, models.resnet50)
learn.model_dir = “/kaggle/working”

2 Likes
3

@moein That worked. Thanks.

But I still need to tell the learner to go up a directory. Like I am getting the error Can't write to '../input/humpback-whale-identification/../kaggle/working'. What should I do?

Like for general Python, we can write !cd .. but that is not working for the learner

4

could you clarify what do you want to do?
you shouldn’t give it the kaggle input path, because that’s not writable.
just put the model_dir equal to “/kaggle/working”

5

@moein Thanks you, the issue has been resolved.

The path to learner which was been given by me was ../ouput/kaggle/working but it was picking up the rest of the path from image bunch. Below are my old code:

src = (ImageList.from_csv(’…/input/humpback-whale-identification/’,
‘train.csv’, folder = ‘train’).split_by_rand_pct(0.2).label_from_df(label_delim=’ '))

learn.model_dir="…/output/kaggle/working"

So, I just edited the path, and the new code is:

learn.model_dir="…/kaggle/working"

1 Like