improve validation for param.Array

[JoyMonteiro]

Hello,

are there some examples on how I could use param for numpy arrays?
I have some code where I'm validating the input manually:
https://github.com/JoyMonteiro/spyRaTE/blob/b943aee2cd2afdb13b6703c8f2ea8ff1b8b8b203/spyrate/rte/rte_interface.py#L712-L734

esp. the assert_shapes code. Is this something I could do more elegantly with Param?
The traittypes project seems to allow something like this, but not sure how Param and traittypes compare, or which I should prefer.

TIA,
Joy

[jbednar]

Param offers an Array parameter type that accepts a numpy.ndarray object, but it doesn't currently offer any other validation. I'd love to see param.Array improved to allow optional validation of at least the shape and dtype of the array, and a PR to achieve that would be gratefully accepted. Validating the shape is a bit tricky; in your example you are validating the complete shape, but I can imagine that some people might want to accept any 2-column array, any 3-column array, etc., not just a specific shape, so it should presumably allow None for any part of the shape tuple, and None for the entire shape to skip shape validation entirely (to match the current default). There may be other things that one might want to validate about arrays, and if those are easily expressed, could be added too. Hopefully the validation in other Parameter examples makes it clear how to add validation in this case, but if not, you can ask us on gitter.im/pyviz/pyviz.

[JoyMonteiro]

I would be happy to create a PR! I want to understand the program flow when using Param so that I can come up with a sensible API. So please let me know if the following is what I would do to (for instance):

class MyArrays(Parameterized):
    arr1 = param.Array(shape=(None, None, None), dtype=np.float64)
    arr2 = param.Array(shape=(None, 2), dtype=np.float32)

or if I don't want class attributes:

class MyArrays(object):
    def __init__(self, **kwargs):
          self.arr1 = param.Array(kwargs['arr1'], shape=(None, None, None), dtype=np.float64)
          ...

The second version makes sense to me if you are using the class just as a data container, with
assignments of attributes only happening during object instantiation. Or is the
first version the preferred style?

Also, how would one implement functionality where arr2's shape is dependent on arr1's shape (or dtype)?