[FEAT] Add operators (mult/div/sub) for (weighted) histograms/accumulators

[jonas-eschle]

Describe the bug

Dividing/multiplying/subtracting histograms with weights results in an AttributeError, only the addition works. (as also mention here: #601 (comment))

Is this not implemented yet or is this ill-defined how to propagate the errors?

If the latter, is it worth to have a different error message than an AttributeError?

Steps to reproduce

Tested on newest release and on dev branch.

import boost_histogram as bh

h = bh.Histogram(bh.axis.Regular(10, -5, 5), storage=bh.storage.Weight())
h.fill([1.5, 2.5], weight=[0.5, 1.5])
h + h  # works
h / h  # this errors
h * h  # this errors
h - h  # this errors

Error message:

AttributeError: 'boost_histogram._core.hist.any_weight' object has no attribute '__imul__'

Full traceback

Details

---

AttributeError Traceback (most recent call last)
in
----> 1 h * h

~/anaconda3/envs/zfit39/lib/python3.9/site-packages/boost_histogram/_internal/hist.py in mul(self, other)
360 ) -> H:
361 result = self.copy(deep=False)
--> 362 return result._compute_inplace_op("imul", other)
363
364 def rmul(

~/anaconda3/envs/zfit39/lib/python3.9/site-packages/boost_histogram/_internal/hist.py in _compute_inplace_op(self, name, other)
399 # Also takes CppHistogram, but that confuses mypy because it's hard to pick out
400 if isinstance(other, Histogram):
--> 401 getattr(self._hist, name)(other._hist)
402 elif isinstance(other, tuple(_histograms)):
403 getattr(self._hist, name)(other)

AttributeError: 'boost_histogram._core.hist.any_weight' object has no attribute 'imul'

[henryiii]

boostorg/histogram#351 is adding this for division to Boost.Histogram. We try to match Boost.Histogram's support here, so division will be added once that is merged.

[henryiii]

And the tracking issue for the rest of them is here: boostorg/histogram#352

We'll follow those as they go in, and include them here.

[masonproffitt]

Division is in boostorg/histogram now, so it can get included here, right?

[henryiii]

Yep, if it's in Boost.Histogram, it can be added to boost-histogram.

[JMulder99]

What is the status of this issue? In the Hist project, someone opened an issue regarding adding weighted histograms, and another issue is regarding using the .plot_ratio for weighted histograms.

It is possible to divide an weighted histogram by an non-weighted histogram. I think all these issues are caused by the lack of the following or similar attributes:
AttributeError: 'boost_histogram._core.hist.any_weight' object has no attribute '__itruediv__'
or
AttributeError: 'boost_histogram._core.hist.any_weight' object has no attribute '__isub__'

Tested on version 1.3.2

[Superharz]

Dear boost-histogram team,

I would aslo be interesed in this feature. Especially diving two weighted histograms by each other.