Type error with new tensorflow 2.18

[maxsch3]

Issue description

When using tensorflow 2.18 I'm getting the following error:

TypeError: `output_signature` must contain objects that are subclass of `tf.TypeSpec` but found <class 'list'> which is not.

Investigation

From the looks of it, new tensorflow wants multiple inputs in x and y structures to be tuples, not lists as it is now

I would like to

make batch shaper to return x and y as tuples

[maxsch3]

The changes were pulled through other PR #147 because it prevented from tests passing

[maxsch3]

Actually, the fix was applied in the wrong place. Reopening

Details

Tensorflow 2.18 now wants structures with tuples e.g. ((x1,x2), y1). The fix above suggests to enforce tuples if multiple inputs in x are provided as list. This can be achived easier and more straightforward the structure itself must be a tuple, not a list.

so if one provided x_structure=(("x1", x1_encoder), ("x2", x2_encoder)), this would have solved the issue at user's end without any fix. It would also retain complete transparency about the structures, that the module had before the fix - you get what you specified.

I would like to

• revert the changes
• update tests to use tuples in structures
• update documentation