Datasets: DatasetBuilder is failing to load a certain datasets

[shikhar2707]

Short description
The tfds DatasetBuilder is failing to load a certain dataset.
I wrote a basic piece of code to check if it can load all the registered datasets or not.
The code -
Dataset_List = tfds.list_builders()
Count = 0
for i in range(len(Dataset_List)):
try:
tfds.builder(Dataset_List[i]).info
except:
Count += 1
which in turns returns the error
ERROR:absl:Failed to construct dataset my_dataset
ERROR:absl:Failed to construct dataset open_images_v4
ERROR:absl:Failed to construct dataset wmt_translate

Environment information

• Operating System: Windows 10 x64 4GB RAM
• Python version: 3.7
• tensorflow-datasets/tfds-nightly version: Tensorflow 2.0
• tensorflow/tensorflow-gpu/tf-nightly/tf-nightly-gpu version:

Expected behavior
The Datasets should load comfortably.
on

[Conchylicultor]

• mydataset: We don't have any datasets like this in our repository, it's probably comes from your codebase where you define class MyDataset
• wmt_translate: This is expected:

      raise ValueError(
          "The raw `wmt_translate` can only be instantiated with the config "
          "kwargs. You may want to use one of the `wmtYY_translate` "
          "implementation instead to get the WMT dataset for a specific year."
      )

• open_images_v4: This may be a bug. Which error is raised ?

[Eshan-Agarwal]

@Conchylicultor open_images_v4 runs fine see colab

[shikhar2707]

my_dataset is one from my repo. But the other two? It shows absl:failed there.

[Conchylicultor]

@Eshan-Agarwal Thanks for confirming. Closing this. Please reopen a bug with the specific open_images_v4 error message if you still encounter this.

[Eshan-Agarwal]

my_dataset is one from my repo. But the other two? It shows absl:failed there.

Please try it with tfds-nightly or after cloning tfds repo, maybe you are using pip package, Also for wmt_translatesee above comment should call it by some configs

[Conchylicultor]

@shikhar2707 You should try with pip tfds-nightly. It's possible we have fixed the open_images bug already in the nightly release.