How to use sets of families for the BicopFamily class

[dmey]

I tried to set the controls for a set of family rather than its members as mentioned in the docs (https://vinecopulib.github.io/pyvinecopulib/_generate/pyvinecopulib.BicopFamily.html#pyvinecopulib.BicopFamily) as:

pv.FitControlsVinecop(family_set=[pv.BicopFamily.all])

however these families are not members of the BicopFamily and result in AttributeError: type object 'pyvinecopulib.BicopFamily' has no attribute 'all'. Is this meant to be called in a different way?

[tvatter]

The sets are part of the main namespace of pyvinecopulib, so you can do pv.all.

[tnagler]

Ah.. We should clarify this in the documentation. Also, wouldn't it be better to have this as pv.families.all or similar?

[tvatter]

Agreed for the clarifications. Regarding the namespace, I'm not sure that it's worth it. If anything, I would actually have pv.BicopFamily.all (i.e., an object of class BicopFamily to return a set of objects).

[dmey]

I see, OK and then this is set by passing e.g. family_set=pv.all to FitControlsVinecop. So in the example above this would be:

pv.FitControlsVinecop(family_set=pv.all)

Thanks for the clarification.

[tnagler]

Wouldn't it be confusing (if at all possible) to have an enum that holds a list of enums of the same type?

[tvatter]

@dmey exactly! However, pv.all is the default anyway :)

[tvatter]

@tnagler The enum would'nt "hold" the list, it would just have a method allowing to create such a list.

[tvatter]

@dmey I just noticed that the documentation of three of those sets was incorrect, see #65 (it's just the doc, the sets are actually available)

[tvatter]

Since the doc has been corrected for a while, I'm closing. Feel free to reopen if needed.

[tvatter]

Actually, arguments of the controls should be better documented anyway. I'm reopening.

[kevindarby]

Perhaps related, is there a way to specify the family for each bicop and then fit them to data?

Do I have to fit each bicop separately then? If so, does the bicop.fit() just take the leftmost columns in the dataset?

For instance, I'd like to set up:

tree0 = [
pv.Bicop(pv.BicopFamily.bb7, rotation = 0), #, parameters = None),
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
pv.Bicop(pv.BicopFamily.indep, rotation = 0),
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
]
tree1 = [
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
pv.Bicop(pv.BicopFamily.bb7, rotation = 0),
]
...

then

order = np.arange(1, len(data.columns) + 1)
vine_struct = pv.DVineStructure(order) #eg [1,2,3...
fit = pv.Vinecop(vine_struct, pair_copulas = [tree0, tree1, tree2, tree3, tree4])

but now

fit.select(data)

can change families

and

controls = pv.FitControlsVinecop(family_set=[pv.BicopFamily.bb7, pv.BicopFamily.indep],
fit.select(data, controls = controls)

can make some bb7 and some indep,

[tvatter]

@kevindarby Since last release, we now have a fit method, so it is indeed possible. See Vinecop.fit. I'm closing but feel free to reopen if this does not solve your issue.