Utility for constructing Dimensions from Parameters?

[jbednar]

An hv.Dimension is quite like a param.Parameter, in that both establish a named dimension with bounds, and they are often used for the same purpose, e.g. to determine the type of widget and the numerical range covered by that widget (such as a slider). In the code below, the declared kdims list is a copy of data already stored and declared on the underlying param.Parameterized object ig.DifferenceOfGaussians, and it would reduce redundancy and increase generality if those values could simply be looked up on the Parameterized object.

So, maybe we could have a utility that will accept a Parameterized object and then return a list of Dimensions for use as kdims, possibly with specified filtering (whitelist and blacklist) on the parameter names? Coupled with #1189, this approach could get rid of all the duplication between the DynamicMap and the DifferenceOfGaussians object, making it possible to use the same code with any such object to give widgets for whatever Parameters it found there.

import numpy as np
import holoviews as hv
import imagen as ig
import imagen.transferfn as tf
from imagen.image import FileImage
from imagen.transferfn.sheet_tf import Convolve
hv.notebook_extension()
%opts Image (cmap="gray") {+axiswise}

FileImage.filename = "/Users/jbednar/imagen/doc/images/ellen_arthur.pgm"

kdims = [hv.Dimension('size',                   range=(0.1, 2)),
         hv.Dimension('positive_size',          range=(0.1, 2)),
         hv.Dimension('negative_size',          range=(0.3, 4)),
         hv.Dimension('positive_aspect_ratio',  range=(1.0,10)),
         hv.Dimension('negative_aspect_ratio',  range=(1.0,10))]

def filtered_images(size,positive_size,negative_size,positive_aspect_ratio,negative_aspect_ratio):
    convolver = Convolve(kernel_pattern=ig.DifferenceOfGaussians(**locals()))
    return FileImage(output_fns=[convolver])[:]

FileImage()[:] + hv.DynamicMap(filtered_images, kdims=kdims)

[jlstevens]

Seems like a very plausible utility, either to just declare kdims from parameters or to declare the entire dynamic map...

[jbednar]

Filtering by precedence would also be useful, so that by default hidden parameters aren't included, but the precedence value can be increased or decreased to get more or fewer widgets. That way there can be filtering that works across a set of Parameterized objects without knowing anything about their internals.

[jlstevens]

After some discussion we weren't convinced of the value of such a utility as it seems rather limited. Unless there are any new suggestions, I'm popping this issue off the 1.8 milestone. I suspect we might decide against the idea in future in which case the PR and this issue could then be closed.

[philippjfr]

Do we still want this? I really can't see an implementation that's actually going to be useful. In any case I don't think it will happen for 1.9.

[jlstevens]

@jbednar I'm happy to close this issue now but also happy to consider any updated proposals you might have in a new issue. Is that ok?

[github-actions]

This issue has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.