Loading diabetic_retinopathy_detection dataset throws error

[sararob]

Short description When I try to run:

train_data, test_data = tfds.load(
    name="diabetic_retinopathy_detection"
)

I get the following error:

AssertionError: Manual directory /root/tensorflow_datasets/downloads/manual/diabetic_retinopathy_detection does not exist. Create it and download/extract dataset artifacts in there.

Just to confirm this wasn't an environment or versioning issue I tried changing the dataset I was loading to mnist and it worked.

Environment information

• Operating System: Colab
• Python version: 3.6.8
• tensorflow-datasets/tfds-nightly version: tfds 1.2.0
• tensorflow/tensorflow-gpu/tf-nightly/tf-nightly-gpu version: Tried with tf 1.14.0 and 2.0.0-rc2

Reproduction instructions Run the code snippet above with both diabetic_retinopathy_detection and mnist. Loading the DR dataset should throw an error.

[Conchylicultor]

As the error message says, diabetic_retinopathy_detection do not supports automatic download, so you should manually download the file and put in in the requested location instead.

[sararob]

Is it possible to make the error message and / or docs clearer to explicitly say you need to download the data into a directory with that name? In the current error message it isn't clear that this specific dataset doesn't support automatic download.

On this page it might be helpful to add something along the lines of "datasets that aren't in Google Cloud Storage don't support the load method. To use them with tfds, download them from the URL and place them in a directory in the following format (zip, raw files, etc.)..."

[Conchylicultor]

Yes, we should make a better job to document this. There is https://github.com/tensorflow/datasets/issues/936 for better surfacing this info on the doc

[vijayphoenix]

This issue can be closed