Why doesn't `The calling iterator did not fully read the dataset being cached.` appear on Google Colab? about tensorflow HOT 2 OPEN

Hi @miticollo ,

Colab usually filters Info and Warning logs. That's why you are not able to see it. The warning descriptions states if you are caching and then applying take() or batch() method. If your dataset is large caching the whole dataset might be a problem and can leads to truncation of dataset. This is just a warning to inform the user about possible data truncation.

Since tf.data.Dataset is a pipeline object you can first batching the dataset and then caching.

dataset = tf.data.Dataset.from_tensor_slices((X_valid, y_valid))
dataset = dataset.map(lambda x, y: (x, y), num_parallel_calls=tf.data.AUTOTUNE)
dataset = dataset.batch(32)
dataset = dataset.cache()
dataset = dataset.prefetch(buffer_size=tf.data.AUTOTUNE)
dataset = dataset.apply(tf.data.experimental.prefetch_to_device(device="/gpu:0"))

from tensorflow.

I tried to reduce dataset to 64 bytes and 70 bytes (I hope I did it correctly).
Then I run the notebook different times and I noticed that the warning sometimes appears and sometimes not.
I don't know why, could be an issue?

Anyway these are outputs from 6 sequential runs (every time I turned off and then on the IPython runtime):

• 1: the warning appears twice.
• 2: the warning appears only in the second case.
• 3: see case 2.
• 4: see case 2.
• 5: see case 2.
• 6: see case 1.

from tensorflow.