Allow adding constants

[HDembinski]

Consider the following code (from my analysis), which sums up histograms generated in different threads:

h_tot = None
for h in generate_histograms():
  if h_tot is None:
    h_tot = h
  else:
    h_tot += h

It would be nicer to be able to write it like this (or something similar)

h_tot = 0
for h in generate_histograms():
   h_tot = h_tot + h

We currently don't allow adding numbers to histograms. So to get this feature, we should allow this. Adding a number to a histogram would only be allowed if that is supported by the underlying accumulator.

[henryiii]

This is useful, though you should be able to do (assuming this is all you are doing):

import functools, operator

h_tot = functools.reduce(operator.add, generate_histograms())

Allowing scalar addition might even enable (there's a caveat in the docs about expecting numeric types here, but I've used it for things like this before, and histograms are numeric of sorts):

h_tot = sum(generate_histograms())

[HDembinski]

You are right, I could write it in more clever ways and then it would already work, but it is also nice to support the naive way (but I think we agree on that). This is again about muscle memory, I know that this trick works with numpy arrays, and I kinda assume it should work with bh.

[HDembinski]

Note that adding may trigger a storage conversion:

bh.Histogram(bh.axis.Integer(0, 3), storage=bh.storage.Int64()) + 0.5

This should convert the Histogram storage to Double() for symmetry with numpy arrays.

When you add two histograms with different storage in Boost.Histogram, the C++ code selects the common type for the result. So int + double yields double, for instance.

[HDembinski]

The same thing happens when you scale.

[HDembinski]

Turns out the Python code currently does not support adding histograms with different storages.

h = bhc.histogram(bhc.axis.regular(10, 0, 1), storage=bhc.storage.int64())
h2 = bhc.histogram(bhc.axis.regular(10, 0, 1), storage=bhc.storage.double())

In [19]: h + h2
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-19-360da8eb62ae> in <module>
----> 1 h + h2

~/Code/boost-histogram/boost_histogram/_internal/hist.py in __add__(self, other)
    127
    128     def __add__(self, other):
--> 129         return self.__class__(self._hist + other._hist)
    130
    131     def __eq__(self, other):

TypeError: unsupported operand type(s) for +: 'boost_histogram._core.hist.any_int64' and 'boost_histogram._core.hist.any_double'

[LovelyBuggies]

I think this issue (adding const & fix add func) could be solved together with #289 , as they are all related to storage type. Besides, I think this snippet should work, but currently NOT.

>>> h = bh.Histogram(bh.axis.Regular(10, 0, 1), storage=bh.storage.Int64())
>>> h2 = bh.Histogram(bh.axis.Regular(10, 0, 1), storage=bh.storage.Double())
>>> assert h == h2

It looks like maybe __eq__ also needs to be renewed.

--- Update ---
Oh, I see they are all marked as 0.8.0 milestone, that's good!

[henryiii]

@HDembinski, is h += 1.0; not supported in Boost.Histogram? This doesn't seem to work in C++:

  std::vector<double> v = {1.0, 2.0, 3.0, 4.0, 5.0};
  auto h = bh::make_histogram(bh::axis::variable<>(v));
  h += 1.0;

Should it be? Or should the be Python only?

[henryiii]

And what does h += 1 do to the flow bins? Should this really just be a special case for 0 false-like, to enable the trick above? (0 or None)

[HDembinski]

It is not supported currently, because the operation seemed to make little sense for a real histogram. But it can easily be added.

[henryiii]

I'm thinking it is probably best in Python to allow False-like (0, 0.0, None, False) to do nothing when added to a histogram, and disallow adding non-False values, since they are ambiguous. For a real histogram, I don't think adding a constant to all bins makes physical sense, especially if bins are not equal size or you have flow bins; if you want to add values using [] = syntax, then you can control exactly what gets added, and that already works.