Taking mean of all regrets instead of sum(max_regret - regret(i))

q5/q5.py

@@ -26,7 +26,7 @@ count=1
 for instance_id, grp1 in df.groupby(['instance']):
     #if instance_id == 'instance-1':
     if True:
-        plt.figure(count)
+        plt.figure(count, figsize=(8, 6))
         count = count + 1
         for algorithm, grp2 in grp1.groupby(['algorithm', 'epsilon']):
             rlist = []
@@ -36,9 +36,10 @@ for instance_id, grp1 in df.groupby(['instance']):
                 algo_label = algo_label + " " + str(algorithm[1])
             for horizon, grp3 in grp2.groupby(['horizon']):
                 hlist.append(horizon)
-                maxr = grp3['regret'].max()
-                rlist.append(grp3['regret'].apply(lambda x: maxr - x).sum())
-            plt.plot(hlist, rlist, label=algo_label)
+                #maxr = grp3['regret'].max()
+                #rlist.append(grp3['regret'].apply(lambda x: maxr - x).sum())
+                rlist.append(grp3['regret'].mean())
+            plt.plot(hlist, rlist, label=algo_label, marker=".")
         plt.xscale("log")
         plt.yscale("log")
         plt.legend()