from_tensor_slices returns ValueError when passing two numpy arrays as arguments

Question

I’ve got two numpy arrays: images_ar (data) and categorical_y_ar (labels).

Both consist of dtype(‘uint8’).

Shape of categorical_y_ar is (978, 126).

Shape of images_ar is (978, 224, 224, 3)

When trying to build a dataset with:

dataset = tf.data.Dataset.from_tensor_slices(images_ar, categorical_y_ar)

I get the following traceback:

ValueError Traceback (most recent call
last) Input In [50], in <cell line: 1>()
—-> 1 dataset = tf.data.Dataset.from_tensor_slices(images_ar, categorical_y_ar)
2 dataset = dataset.repeat().batch(batch_size)
4 val_size = int( num_classes* 0.2)

File
~.condaenvssklearn-envlibsite-packagestensorflowpythondataopsdataset_ops.py:814,
in DatasetV2.from_tensor_slices(tensors, name)
736 @staticmethod
737 def from_tensor_slices(tensors, name=None):
738 """Creates a Dataset whose elements are slices of the given tensors.
739
740 The given tensors are sliced along their first dimension. This operation (…)
812 Dataset: A Dataset.
813 """
–> 814 return TensorSliceDataset(tensors, name=name)

File
~.condaenvssklearn-envlibsite-packagestensorflowpythondataopsdataset_ops.py:4728, in TensorSliceDataset.init(self, element, is_files, name) 4719
for t in self._tensors[1:]: 4720
batch_dim.assert_is_compatible_with( 4721
tensor_shape.Dimension( 4722
tensor_shape.dimension_value(t.get_shape()[0]))) 4724
variant_tensor = gen_dataset_ops.tensor_slice_dataset( 4725
self._tensors, 4726
output_shapes=structure.get_flat_tensor_shapes(self._structure),
4727 is_files=is_files,
-> 4728 metadata=self._metadata.SerializeToString()) 4729 super(TensorSliceDataset, self).init(variant_tensor)

File
~.condaenvssklearn-envlibsite-packagestensorflowpythondataopsdataset_ops.py:668,
in DatasetV2._metadata(self)
666 """Helper for generating dataset metadata."""
667 metadata = dataset_metadata_pb2.Metadata()
–> 668 if self._name:
669 metadata.name = _validate_and_encode(self._name)
670 return metadata

ValueError: The truth value of an array with more than one element is
ambiguous. Use a.any() or a.all()

Please help me to figure out why this error pops up.

Asked By: Sergey Belousov

||

Source

Answer 1

Try to pass the data as a set or list not as a single entity to the tf.data.Dataset.from_tensor_slice(), as shown below

x_train = np.random.randn(978, 224, 224, 3)
y_train = np.random.randint(0,127, size=(978, 127))

dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))

print(next(iter(dataset.take(1))))

Output:

<tf.Tensor: shape=(224, 224, 3), dtype=float64, numpy= ...

Answered By: Mohammad Ahmed

from_tensor_slices returns ValueError when passing two numpy arrays as arguments

Question:

Answers: