tf.data.array() cannot consume a Tensor, despite the documentation claiming so

[AdmiralSnyder]

System information

• windows 11, edge
• TensorFlow.js -> script link to latest (as of 2024-06-10)
• Browser version: 125.0.2535.92 (Official build) (64-bit)
• Tensorflow.js Converter Version: N/A

Describe the current behavior i am trying to create a dataset from a tensor, but that failed, see code:

Describe the expected behavior the dataset should be create-able

Standalone code to reproduce the issue

let myTensor = tf.randomUniform([100], -2, 2); // returns a tensor as docs claim
// myTensor.print(); // prints the random data in this tensor
let myDataset = tf.data.array(myTensor); // returns something null-y - definitely not a dataset made from this tensor, 
console.log(myDataset); // outputs this: r {size: null}

[gaikwadrahul8]

Hi, @AdmiralSnyder

I apologize for the delayed response and I was trying to replicate the same behavior from my end with your code snippet and I'm getting same result for reference I have added screenshot below:

When I'm trying something like below it seems like working fine, if I've missed something here please let me know. Thank you for your cooperation and patience.

[github-actions[bot]]

This issue has been marked stale because it has no recent activity since 7 days. It will be closed if no further activity occurs. Thank you.

[AdmiralSnyder]

Hey, the documentation clearly states that data.array should be able to consume a Tensor, rather than a Tensor[]:

Quote: Parameters: items (tf.void|number|string|TypedArray|tf.Tensor|tf.Tensor[]|{[key: string]:tf.Tensor|number|string}[]) An array of elements that will be parsed as items in a dataset.

which I read as: data.array consumes these types of things:

• tf.void
• number
• string
• TypedArray
• tf.Tensor --> a single tensor
• tf.Tensor[] --> an array of tensors
• an array of objects with key - value specifications where the key is a string and the value is either a tensor or a number or a string

-> and since the simple case where you enter the single tensor into the data.array function clearly doesn't work, this is either a bug in the implementation or in the documentation -> the workaround of putting the tensor between square brackets is not acceptable.