Load tfrecord dataset from google drive when using TPU in colab

TensorFlow General Discussion
, ,
1

I am using TPU runtime in Google Colab, and I know tpu runtimes can only access data in GCS as only the GCS file system is registered.
However I found StreamingFilesDataset can read tfrecord file from local.(tensorflow/tensorflow/python/tpu/datasets.py at master · tensorflow/tensorflow · GitHub)

I wrote like this
from tensorflow.python.tpu import datasets as tf_tpu_ds
tf_tpu_ds.StreamingFilesDataset( dataset_path, ‘tfrecord’, file_reader_job=‘localhost’)

Then I got

AttributeError: ‘OwnedIterator’ object has no attribute ‘string_handle’

‘string_handle’ seems tensorflow version 1 function.

So, Is StreamingFilesDataset version 1 only supported, or bugs?

3

Hi Sabe, can you share a code snippet of the loading you are trying please?