Add NDArray trait type

[maedoc]

This adds an NDArray trait type that validates on shape and dtype constraints.

[maedoc]

FWIW the tests pass on py 27 but importing numpy fails with an ImportError for multiarray on py 33. If anyone knows how to make that work on Travis, I suspect the tests would pass on py 33 as well.

[Carreau]

Will look into that with more detail, see also comment on other PR #2 that maybe we want to have ipython/ipython#4659 totally integrated into IPython before pushing this repo forward.

One problem I might see with this as it is, is that it makes traitlets requiring numpy, and I'm not sure it is a good idea. We can probably conditional loading that make NDArray availlable IIF numpy is availlable, but I would like to find a better solution.

[maedoc]

The actual module could test that the class name matches instead of the
instance check, yes? (Isn't that already supported?) The tests still would
require numpy. Would that be satisfactory?
On Jan 3, 2014 9:31 AM, "Matthias Bussonnier" notifications@github.com
wrote:

Will look into that with more detail, see also comment on other PR #2https://github.com/ipython/traitlets/pull/2that maybe we want to have
ipython/ipython#4659 https://github.com/ipython/ipython/pull/4659totally integrated into IPython before pushing this repo forward.

One problem I might see with this as it is, is that it makes traitlets
requiring numpy, and I'm not sure it is a good idea. We can probably
conditional loading that make NDArray availlable IIF numpy is availlable,
but I would like to find a better solution.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1#issuecomment-31510653
.

[maedoc]

Is ipython/ipython#4659 waiting on something in particular?

[takluyver]

I agree that traitlets shouldn't require numpy, but if you have to import it as from traitlets.ndarray import NDArray, I'm fine with that importing numpy - an ImportError seems like a clearer failure mode than trying to do anything clever.

The Python 3 testing is a problem with Travis - you're installing python3-numpy from apt, but on the Ubuntu version Travis uses, that's a package for Python 3.2. There are instructions to use Miniconda on Travis here: https://gist.github.com/dan-blanchard/7045057

As an aside: if anyone has the resources to set up a building/testing service based on conda envs instead of virtualenvs, so we don't need that kind of hack, I think the SciPy ecosystem would be very grateful.

[Carreau]

Is ipython/ipython#4659 waiting on something in particular?

Mainly that I got time to polish it, that is to say move the docs and do cross refs and that we agree on what should be exposed in traitlets where.

Basically I don't really like the fact that you can import Config object directly from traitlets and things like that.

from traitlets.ndarray import NDArray, I'm fine with that importing numpy - an ImportError seems like a clearer failure mode than trying to do anything clever.

I haven't read the code yet, so if it is like that I'm find with it too.

[maedoc]

I didn't touch the traitlets.__init__ module on purpose, so any user code only requires numpy if the traitlets.ndarray module is explicitly imported. I also prefer the obvious ImportError this will generate as code using the NDArray class presumably will depend, independently of traitlets, on numpy.

[edited for clarity]

[rossant]

@maedoc just wondering what's the status of this?
Is it conceivable to have a ndarray trait notify content changes like:

class A(HasTraits):
    arr = NDArray(...)

a = A()
a.arr = np.array(...)  # ==> notify
a.arr[0] = 1  # ==> notify
a.arr += 1  # ==> notify
...

[Carreau]

We will restart Traitlets from scratch after 3.0, (spliting it again from IPython)

Once this is done, the more the merrier.

[rossant]

OK. Any idea of the timeframe? What's wrong with the current implementation?

[Carreau]

OK. Any idea of the timeframe?

A few month.

What's wrong with the current implementation?

Nothing, this just copy and past of old IPython version, which has evolved too much with widgets.
I basically spent a week creating this package with a wrongly set mutex. So the backend reply to me that transaction could not be applied, and I need to reperfom it.

I just need to find a week to say :

"""
Don't F%^& dare touching `Traitlets` code for a week
"""

and re-do the work, then release mutex. It's doable in 2 days if I don't move the documentation now, and I don't try to move the Config part at the same time. Probably a day with cafein, but think one more day to review.

[maedoc]

@rossant I seem to have lost this repo, but a.arr = r_[] could notify while a.arr += 1 would require an ndarray subclass, as any reference to a.arr results in the ndarray and not the trait's data descriptor arr (whic is the thing wrapping access to the ndarray).

[maedoc]

@minrk why no action?

[Carreau]

I think that because most of that is in numtraits package in @astrofrog/numtraits.

[ellisonbg]

I thought we were going to put these types of things here:

https://github.com/jupyter-incubator/traittypes