Problem (bug?) TextDataBunch/ TextList from_csv function

fastai
1

Hello All,

I am trying to recreate the IMDB example from lesson3.
I am using the IMDB_SAMPLE dataset.
When trying to load the csv file using:

data_lm = (TextList.from_csv(path, 'texts.csv', cols='text')
           .split_from_df(col=2)                
           .label_from_df(cols=0)
           .databunch(bs=bs))

…

after executing:
learn.lr_find()

I get the following error:

Expected input batch_size (81792) to match target batch_size (48).

Also all GPU memory is used and the server gives a warning.

When using:
data = TextDataBunch.from_csv(path, 'texts.csv', bs=bs, num_workers=0)

After:
learn.lr_find()

I get the following error:

CUDA out of memory. Tried to allocate 2.68 GiB (GPU 0; 11.17 GiB total capacity; 8.94 GiB already allocated; 1.80 GiB free; 123.06 MiB cached)

I think this error is related to the error above.

When using the full IMDB dataset, which contains folders instead of a csv file and loading it using:

data_lm = (TextList.from_folder(path)
            .filter_by_folder(include=['train', 'test', 'unsup']) 
            .random_split_by_pct(0.1)
            .label_for_lm()           
            .databunch(bs=bs))

and calling learn.lr_find all is ok.

I’ve included a link which demonstrates the problem:
https://colab.research.google.com/gist/jj321/c71bade2e999c1806a085be965096f9a/imdb.ipynb

“Option 1” and “Option 2” give an error, and the section “This is ok” runs ok.
I’ve also tried other datasets (csv’s) to no avail.
I don’t know if it is relevant but i am using Google Colab.

Am i doing something wrong or is this a bug in the from_csv function?

Kind regards,

Jaap

2

I agree that there’s something wrong in the from_csv() databunch creation. I’ve hit both CUDA OOM bugs and this one:

RuntimeError Traceback (most recent call last)

<ipython-input-35-174310cf6578> in <module>() ----> 1 learn.fit_one_cycle(20, 1e-2, callbacks=[SaveModelCallback(learn, every=‘epoch’, monitor=‘accuracy’, name=path + ‘/new-’ + model_name + ‘-data-aug-on-’ + currtime)])

9 frames

/usr/local/lib/python3.6/dist-packages/fastai/metrics.py in accuracy(input, targs) 28 input = input.argmax(dim=-1).view(n,-1) 29 targs = targs.view(n,-1) —> 30 return (input==targs).float().mean() 31 32 def accuracy_thresh(y_pred:Tensor, y_true:Tensor, thresh:float=0.5, sigmoid:bool=True)->Rank0Tensor:

RuntimeError: Expected object of scalar type Long but got scalar type Float for argument #2 ‘other’

image
image.png1450×387 38.9 KB

If I rearrange the dataset (dog breed classification) in subfolders and use from_folder() everything runs smooth.

1 Like