Quantile export

[scrouthtv]

In order to accurately plot box and whiskers plots using third-party libraries, we have to export the lower and upper (first and third) quantile.

Describe the solution you'd like

diff --git a/evo/core/metrics.py b/evo/core/metrics.py
index 1b6396f..dc10bef 100644
--- a/evo/core/metrics.py
+++ b/evo/core/metrics.py
@@ -59,6 +59,8 @@ class StatisticsType(Enum):
     min = "min"
     max = "max"
     sse = "sse"
+    q1 = "q1"
+    q3 = "q3"
 
 
 @unique
@@ -152,6 +154,10 @@ class PE(Metric):
             return np.min(self.error)
         elif statistics_type == StatisticsType.std:
             return float(np.std(self.error))
+        elif statistics_type == StatisticsType.q1:
+            return np.quantile(self.error, 0.25)
+        elif statistics_type == StatisticsType.q3:
+            return np.quantile(self.error, 0.75)
         else:
             raise MetricsException("unsupported statistics_type")

Describe alternatives you've considered
Using the plots that are saved by evo. Downsides: cannot integrate well with some typesetting systems.

Additional context
New output of evo_results:

APE w.r.t. translation part (m)
(not aligned)


                   rmse      mean    median       std       min      max  \
fasterlio.csv  0.873564  0.727618  0.775811  0.483411  0.004898  1.57153   

                       sse        q1       q3  
fasterlio.csv  1379.711516  0.246562  1.10841  

[MichaelGrupp]

Hi,

looks like a useful addition. The diff you posted should be enough to add that, so feel free to make a PR 🙂