Allow several 1-sigma bars in the best fit graph

Problem happened when a 1D likelihood shows two intervals below 1-sigma. likelihood_scan.pdf.

This MR allows to draw two lines (or more) in the ExclusionAndBestFit, only one of them with the best fit. exclusion.pdf

An ideal case would be to modify evaluate_likelihood_scan_1d but it may impact more functions so I restricted myself to modifying the plot_exclusion_and_bestfit_1d only, for people who are interested in a quick fix.

dhi/plots/exclusion.py

@@ -179,9 +179,10 @@ def plot_exclusion_and_bestfit_1d(
 
     # perform scans to extract best fit values
     scans = []
+    n_intervals = 1
     for d in data:
         if not d or d.get("nll_values") is None:
-            scans.append(None)
+            scans.append([None])
             continue
 
         # preprocess values
@@ -190,18 +191,40 @@ def plot_exclusion_and_bestfit_1d(
             (poi, d["nll_values"][scan_parameter]), shift_negative_values=True, remove_nans=True,
             min_is_external=poi_min is not None, origin="entry '{}'".format(d["name"]))
 
+        # check for disjoint intervals 
+        dnll2_below_1sigma = np.where(dnll2_values < 1)[0]
+        dnll2_below_1sigma_intervals = np.split(dnll2_below_1sigma, np.where(np.diff(dnll2_below_1sigma) != 1)[0]+1)
+
         # evaluate the scan
-        scans.append(evaluate_likelihood_scan_1d(poi_values, dnll2_values, poi_min=poi_min,
-            origin="entry '{}'".format(d["name"])))
+        if len(dnll2_below_1sigma_intervals) <= 1:
+            scans.append([evaluate_likelihood_scan_1d(poi_values, dnll2_values, poi_min=poi_min,
+                origin="entry '{}'".format(d["name"]))])
+        else:
+            print '{} disjoint intervals of dnll2 values below 1 sigma'.format(len(dnll2_below_1sigma_intervals))
+            # perform scans 
+            scan_intervals = []
+            for i in range(len(dnll2_below_1sigma_intervals)):
+                dnll2_values_tmp = dnll2_values.copy()
+
+                # exclude points of other intervals 
+                for j in range(len(dnll2_below_1sigma_intervals)):
+                    if i != j:
+                        dnll2_values_tmp[dnll2_below_1sigma_intervals[j]] = 1.1
+                scan_intervals.append(evaluate_likelihood_scan_1d(poi_values, dnll2_values_tmp, poi_min=None,origin="entry '{}' (interval {})".format(d["name"],i)))
+
+            # Order them by minimum of dnll2
+            order_dnll2 = np.argsort([dnll2_values[idx].min() for idx in dnll2_below_1sigma_intervals])
+            scans.append([scan_intervals[i] for i in order_dnll2])
+            n_intervals = max(n_intervals,len(scan_intervals))
 
     # draw best fit values
     if any(scans):
         f = int(show_best_fit_error)
         g_bestfit = create_tgraph(n,
-            [(scan.num_min() if scan else -1e5) for scan in scans],
+            [(scan[0].num_min() if scan[0] else -1e5) for scan in scans],
             [n - i - 0.5 for i in range(n)],
-            [(f * scan.num_min.u(direction="down", default=0.) if scan else 0) for scan in scans],
-            [(f * scan.num_min.u(direction="up", default=0.,) if scan else 0) for scan in scans],
+            [(f * scan[0].num_min.u(direction="down", default=0.) if scan[0] else 0) for scan in scans],
+            [(f * scan[0].num_min.u(direction="up", default=0.,) if scan[0] else 0) for scan in scans],
             0,
             0,
         )
@@ -210,6 +233,22 @@ def plot_exclusion_and_bestfit_1d(
         draw_objs.append((g_bestfit, "PZ" + opt("E")))
         legend_entries.append((g_bestfit, "Best fit value", "P" + opt("L")))
 
+        # add potential additional intervals 
+        if n_intervals > 1:
+            for idx in range(1,n_intervals):
+                g_interval = create_tgraph(n,
+                    [(scan[idx].num_min() if len(scan)>idx else -1e5) for scan in scans],
+                    [n - i - 0.5 for i in range(n)],
+                    [(f * scan[idx].num_min.u(direction="down", default=0.) if len(scan)>idx else 0) for scan in scans],
+                    [(f * scan[idx].num_min.u(direction="up", default=0.,) if len(scan)>idx else 0) for scan in scans],
+                    0,
+                    0,
+                )
+                r.setup_graph(g_interval, props={"MarkerSize": 0., "LineWidth": 1})
+                draw_objs.append((g_interval, "PZ" + opt("E")))
+                
+
+
     # theory prediction
     if x_min < 1:
         sm_value = poi_data[scan_parameter].sm_value
@@ -233,14 +272,14 @@ def plot_exclusion_and_bestfit_1d(
     for i, (d, scan) in enumerate(zip(data, scans)):
         # name labels
         label = to_root_latex(br_hh_names.get(d["name"], d["name"]))
-        if scan:
+        if scan[0]:
             if summary :
                 label_summary = to_root_latex(hh_references.get(d["name"], d["name"]))
-                label = label_tmpl_summary % (label, scan_label, scan.num_min.str("%.2f", style="root",
-                    force_asymmetric=True, styles={"space": ""}), label_summary)
+                label = label_tmpl_summary % (label, scan_label, ' / '.join([s.num_min.str("%.2f", style="root",
+                    force_asymmetric=True, styles={"space": ""}) for s in scan]), label_summary)
             else :
-                label = label_tmpl_scan % (label, scan_label, scan.num_min.str("%.2f", style="root",
-                    force_asymmetric=True, styles={"space": ""}))
+                label = label_tmpl_scan % (label, scan_label, ' / '.join([s.num_min.str("%.2f", style="root",
+                    force_asymmetric=True, styles={"space": ""}) for s in scan]))
         else:
             label = label_tmpl % (label,)