This is the design document for Unified Histogram Indexing (UHI). Much of the original plan is now implemented in boost-histogram. Other histogramming libraries can implement support for this as well, and the “tag” functors, like sum and loc can be used between libraries.


The following examples assume you have imported loc, sum, rebin, underflow, and overflow from boost-histogram or any other library that implements UHI.


v = h[b]          # Returns bin contents, indexed by bin number
v = h[loc(b)]     # Returns the bin containing the value
v = h[loc(b) + 1] # Returns the bin above the one containing the value
v = h[underflow]  # Underflow and overflow can be accessed with special tags


h == h[:]             # Slice over everything
h2 = h[a:b]           # Slice of histogram (includes flow bins)
h2 = h[:b]            # Leaving out endpoints is okay
h2 = h[loc(v):]       # Slices can be in data coordinates, too
h2 = h[::rebin(2)]    # Modification operations (rebin)
h2 = h[a:b:rebin(2)]  # Modifications can combine with slices
h2 = h[::sum]         # Projection operations # (name may change)
h2 = h[a:b:sum]       # Adding endpoints to projection operations
h2 = h[0:len:sum]     #   removes under or overflow from the calculation
h2 = h[v, a:b]        #   A single value v is like v:v+1:sum
h2 = h[a:b, ...]      # Ellipsis work just like normal numpy


# Single values
h[b] = v         # Returns bin contents, indexed by bin number
h[loc(b)] = v    # Returns the bin containing the value
h[underflow] = v # Underflow and overflow can be accessed with special tags

h[...] = array(...) # Setting with an array or histogram sets the contents if the sizes match
                    # Overflow can optionally be included if endpoints are left out
                    # The number of dimensions for non-scalars should match (broadcasting works normally otherwise)

All of this generalizes to multiple dimensions. loc(v) could return categorical bins, but slicing on categories would (currently) not be allowed. These all return histograms, so flow bins are always preserved - the one exception is projection; since this removes an axis, the only use for the slice edges is to be explicit on what part you are interested for the projection. So an explicit (non-empty) slice here will case the relevant flow bin to be excluded.

loc, project, and rebin all live inside the histogramming package (like boost-histogram), but are completely general and can be created by a user using an explicit API (below). underflow and overflow also follow a general API.

One drawback of the syntax listed above is that it is hard to select an action to run on an axis or a few axes out of many. For this use case, you can pass a dictionary to the index, and that has the syntax {axis:action}. The actions are slices, and follow the rules listed above. This looks like:

h[{0: slice(None, None, bh.rebin(2))}] # rebin axis 0 by two
h[{1: slice(0, bh.loc(3.5))}]          # slice axis 1 from 0 to the data coordinate 3.5
h[{7: slice(0, 2, bh.rebin(4))}]       # slice and rebin axis 7

If you don’t like manually building slices, you can use the Slicer() utility to recover the original slicing syntax inside the dict:

s = bh.tag.Slicer()

h[{0: s[::bh.rebin(2)]}]   # rebin axis 0 by two
h[{1: s[0:bh.loc(3.5)]}]   # slice axis 1 from 0 to the data coordinate 3.5
h[{7: s[0:2:bh.rebin(4)]}] # slice and rebin axis 7

Invalid syntax:

h[1.0] # Floats are not allowed, just like numpy
h[::2] # Skipping is not (currently) supported
h[..., None] # None == np.newaxis is not supported

Rejected proposals or proposals for future consideration, maybe hist-only:

h2 = h[1.0j:2.5j + 1] # Adding a j suffix to a number could be used in place of `loc(x)`
h2 = h[1.0] # Floats in place of `loc(x)`: too easy to make a mistake


For a histogram, the slice should be thought of like this:


The start and stop can be either a bin number (following Python rules), or a callable; the callable will get the axis being acted on and should return an extended bin number (-1 and len(ax) are flow bins). A provided callable is bh.loc, which converts from axis data coordinates into bin number.

The final argument, action, is special. A general API is being worked on, but for now, bh.sum will “project out” or “integrate over” an axes, and bh.rebin(n) will rebin by an integral factor. Both work correctly with limits; bh.sum will remove flow bins if given a range. h[0:len:bh.sum] will sum without the flow bins.

Here are a few examples that highlight the functionality of UHI:

Example 1:

You want to slice axis 0 from 0 to 20, axis 1 from .5 to 1.5 in data coordinates, axis 2 needs to have double size bins (rebin by 2), and axis 3 should be summed over. You have a 4D histogram.


ans = h[:20, bh.loc(-.5):bh.loc(1.5), ::bh.rebin(2), ::bh.sum]

Example 2:

You want to set all bins above 4.0 in data coordinates to 0 on a 1D histogram.


h[bh.loc(4.0):] = 0

You can set with an array, as well. The array can either be the same length as the range you give, or the same length as the range + under/overflows if the range is open ended (no limit given). For example:

h = bh.Histogram(bh.axis.Regular(10, 0, 1))
h[:] = np.ones(10) # underflow/overflow still 0
h[:] = np.ones(12) # underflow/overflow now set too

Note that for clarity, while basic Numpy broadcasting is supported, axis-adding broadcasting is not supported; you must set a 2D histogram with a 2D array or a scalar, not a 1D array.

Example 3:

You want to sum from -infinity to 2.4 in data coordinates in axis 1, leaving all other axes alone. You have an ND histogram, with N >= 2.


ans = h[:, :bh.loc(2.4):bh.sum, ...]

Notice that last example could be hard to write if the axis number, 1 in this case, was large or programmatically defined. In these cases, you can pass a dictionary of {axis:slice} into the indexing operation. A shortcut to quickly generate slices is provided, as well:

ans = h[{1: slice(None,bh.loc(2.4),bh.sum)}]

# Identical:
s = bh.tag.Slicer()
ans = h[{1: s[:bh.loc(2.4):bh.sum]}]

Example 4:

You want the underflow bin of a 1D histogram.


val = h1[bh.underflow]


Axis indexing

TODO: Possibly extend to axes. Would follow the 1D cases above.

Implementation notes

loc, rebin, and sum are not unique tags, or special types, but rather APIs for classes. New versions of these could be added, and implementations could be shared among Histogram libraries. For clarity, the following code is written in Python 3.6+. Prototype here. Extra doc here.

Note that the API comes in two forms; the __call__/__new__ operator form is more powerful, slower, optional, and is currently not supported by boost-histogram. A fully conforming UHI implementation must allow the tag form without the operators.

Basic implementation (WIP):

class loc:
    "When used in the start or stop of a Histogram's slice, x is taken to be the position in data coordinates."
    def __init__(self, value, offset):
        self.value = value
        self.offset = offest

    # supporting __add__ and __sub__ also recommended

    def __call__(self, axis):
        return axis.index(self.value) + self.offset

# Other flags, such as callable functions, could be added and detected later.

# UHI will perform a maximum performance sum when python's sum is encountered

def underflow(axis):
    return -1
def overflow(axis):
    return len(axis)

class rebin:
    When used in the step of a Histogram's slice, rebin(n) combines bins,
    scaling their widths by a factor of n. If the number of bins is not
    divisible by n, the remainder is added to the overflow bin.
    def __init__(self, factor):
        # Items with .factor are specially treated in boost-histogram,
        # performing a high performance rebinning
        self.factor = factor

    # Optional and not used by boost-histogram
    def __call__(self, binning, axis, counts):
        factor = self.factor
        if isinstance(binning, Regular):
            indexes = (numpy.arange(0, binning.num, factor),)

            num, remainder = divmod(binning.num, factor)
            high, hasover = binning.high, binning.hasover

            if binning.hasunder:
                indexes[0][:] += 1
                indexes = ([0],) + indexes

            if remainder == 0:
                if binning.hasover:
                    indexes = indexes + ([binning.num + int(binning.hasunder)],)
                high = binning.left(indexes[-1][-1])
                hasover = True

            binning = Regular(num, binning.low, high, hasunder=binning.hasunder, hasover=hasover)
            counts = numpy.add.reduceat(counts, numpy.concatenate(indexes), axis=axis)
            return binning, counts

            raise NotImplementedError(type(binning))