Support ROOT serialization

[LovelyBuggies]

We might need to design file media for cross-platform development. I think ROOT file is a good option. Maybe we can design functionality to serialize/deserialize our histograms or axis objects to/from ROOT files.

[henryiii]

This is handled by Aghast, but yes, we absolutely need to have Hist talk to Aghast so we can save and load ROOT Histograms, and help push Aghast forward as well.

[LovelyBuggies]

Duplicate with Serialization via Aghast in Hist plans.

[HDembinski]

boost-histogram should also support this.

[LovelyBuggies]

Since we can convert our histograms to numpy.histogram objects (or nparray objects) and uproot could save and load numpy.histogram objects. Wouldn't that be an easy thing? Are we expected to convert them directly (bh/hist.histogram to ROOT without numpy)?

import hist
import uproot
import numpy as np
import matplotlib.pyplot as plt

file = uproot.recreate("tmp.root")
h = hist.Hist(hist.axis.Regular(15, -3, 3, name='x'))
h.fill(np.random.normal(size=1_000))

w, data = h.to_numpy()

file['title'] = 'A histogram'
file['h'] = np.histogram(data[:-1], bins=data, weights=w)
file2 = uproot.open("tmp.root")
print(file2['title'])
file2['h'].show()

# A histogram
#                  0                                                         159.6
#                  +-------------------------------------------------------------+
# [-inf, -3)   0   |                                                             |
# [-3, -2.6)   4   |**                                                           |
# [-2.6, -2.2) 9   |***                                                          |
# [-2.2, -1.8) 13  |*****                                                        |
# [-1.8, -1.4) 49  |*******************                                          |
# [-1.4, -1)   81  |*******************************                              |
# [-1, -0.6)   119 |*********************************************                |
# [-0.6, -0.2) 152 |**********************************************************   |
# [-0.2, 0.2)  147 |********************************************************     |
# [0.2, 0.6)   136 |****************************************************         |
# [0.6, 1)     107 |*****************************************                    |
# [1, 1.4)     87  |*********************************                            |
# [1.4, 1.8)   63  |************************                                     |
# [1.8, 2.2)   17  |******                                                       |
# [2.2, 2.6)   10  |****                                                         |
# [2.6, 3)     4   |**                                                           |
# [3, inf]     0   |                                                             |
#                  +-------------------------------------------------------------+

[LovelyBuggies]

If we can do it in this way, I think what we need to do is:

• To see whether all bh/hist's histograms could be converted to numpy.histogram (for some special axes)
• Encapsulate a function to write some params to the file like the axis names