fix: writing TProfiles from hist

[ariostas]

This PR adds some flexibility to the hist profiles that can be written to file. The old code was assuming ROOT histograms that were converted to hist with to_hist, which adds some metadata, but if the metadata was not present it just didn't work. Someone should double-check that I'm inputting the right data into to_TProfile. Closes #1531.

[ariostas]

@pfackeldey could you take a look at this since you're good with histograms?

[wiso]

thanks, it seems to work, even if I have to simplify my original code. For example with storage=hist.storage.WeightedMean() it doesn't work #1533 and ND profiles are not supported #1534.

By the way, I get a warning

FutureWarning: .metadata was not set, returning None instead of Attribute error, boost-histogram 1.7+ will error.
  if obj.metadata is not None and "fSumw2" in obj.metadata.keys():

[ariostas]

Thank you, @wiso! I fixed the warning and the WeightedMean storage. I'll follow up on the ND profiles on a separate PR.

[ianna]

@ariostas - looks great! Thanks. Please merge it if you are done with it. Thanks!

[wiso]

I think this should be reopen, the values I get when reading the written object are not correct.

Test case

import uproot
import hist
import numpy as np

axis_x = hist.axis.IntCategory(range(10))
h = hist.Hist(axis_x, storage=hist.storage.Mean())

sample0 = np.array([10, 20, 30, 10])
sample1 = np.array([10, 20, 10, 10, 0])
h.fill([0, 0, 0, 0], sample=sample0)
h.fill([1, 1, 1, 1, 1], sample=sample1)

expected_count0 = len(sample0)
expected_count1 = len(sample1)
expected_mean0 = np.mean(sample0)
expected_mean1 = np.mean(sample1)
expected_variance0 = np.var(sample0) / (len(sample0) - 1)
expected_variance1 = np.var(sample1) / (len(sample1) - 1)

hist_count0 = h[0].count
hist_count1 = h[1].count
hist_mean0 = h[0].value
hist_mean1 = h[1].value
hist_variance0 = h.variances()[0]
hist_variance1 = h.variances()[1]
hist_bintype = type(h[0])



with uproot.recreate("test.root") as f:
    f['h'] = h

with uproot.open("test.root") as f:
    h = f['h'].to_hist()

    # FAIL AttributeError: 'boost_histogram._core.accumulators.WeightedMean' object has no attribute 'count'
#    uproot_count0 = h[0j].count
#    uproot_count1 = h[1j].count

    uproot_mean0 = h[0].value
    uproot_mean1 = h[1].value
    uproot_variance0 = h.variances()[0]
    uproot_variance1 = h.variances()[1]
    uproot_bintype = type(h[0])

print(f"count0: {expected_count0:.1f}  {hist_count0:.1f}")
print(f"count1: {expected_count1:.1f}  {hist_count1:.1f}")
print(f"mean0: {expected_mean0:.1f}  {hist_mean0:.1f} {uproot_mean0:.1f}")
print(f"mean1: {expected_mean1:.1f}  {hist_mean1:.1f} {uproot_mean1:.1f}")
print(f"var0: {expected_variance0:.1f}  {hist_variance0:.1f} {uproot_variance0:.1f}")
print(f"var1: {expected_variance1:.1f}  {hist_variance1:.1f} {uproot_variance1:.1f}")
print(f"bin-type {hist_bintype}  {uproot_bintype}")

with output

count0: 4.0  4.0
count1: 5.0  5.0
mean0: 17.5  17.5 17.5
mean1: 10.0  10.0 10.0
var0: 22.9  22.9 1.2
var1: 10.0  10.0 0.2
bin-type <class 'boost_histogram.accumulators.Mean'>  <class 'boost_histogram.accumulators.WeightedMean'>

I guess the main problem is that when retrieving the histogram from uproot the bin type is WeightedMean and the variances are wrong.