compare ratio of weighted histograms (#1845)

* first attempt just from current knowledge

runs but has two main issues observed

1. lots of runtime warnings about invalid division. This comes from 0/0
   since 1/0 is converted to np.nan or np.inf
2. ratio uncertainty is crazy large, probably because the weights for
   the signal samples are also extreme

* silence NumPy divide-by-zero warnings since the defaults are what we want

* use poisson-ratio instead of raw poisson counts

Author: tomeichlersmith

ComparePlots/_differ.py

@@ -9,6 +9,7 @@
 import uproot
 import numpy as np
 import hist.intervals
+import mplhep
 
 # us
 from ._file import File
@@ -144,11 +145,19 @@ def plot1d(
         den_h, _den_art = raw_histograms[0]
         bins = den_h.axes[0].edges
         for num_h, num_art in raw_histograms[1:]:
-            (den_h/num_h).plot1d(
+            # NumPy nicely warns us when we divide by zero,
+            # but its defaults are sensible to use
+            # x/0 = inf, 0/0 = nan
+            with np.errstate(divide="ignore", invalid="ignore"):
+                ratio = num_h.values() / den_h.values()
+            mplhep.histplot(
+                ratio,
+                bins = bins,
                 ax = ratio_ax,
                 yerr = hist.intervals.ratio_uncertainty(
                     num = num_h.values(),
-                    denom = den_h.values()
+                    denom = den_h.values(),
+                    uncertainty_type = 'poisson-ratio'
                 ),
                 histtype='errorbar',
                 color = num_art[0].stairs.get_edgecolor()