Sweep timings - averaging over all instances vs only successful instances

[Sakshi21299]

The main branch was averaging the solve time on all instances (successful and failed convergence) of the parameter sweep. However, the results reported in the paper, only average solve times on the successful instances. PR #65 fixed this in the main branch.
The results before and after this change are reported below:

=========================================
Averaged over all instances

=========================================
Averaged over successful instances

[Sakshi21299]

If we average solve time over all instances, it decreases uniformly. But if the solvetime is averaged over only successful instances, it first decreases and then goes up - as reported in the paper

[Sakshi21299]

@Robbybp Tagging you here to check this out

[Robbybp]

Thanks. Just confirmed that I see this trend when I run on my most recent results. Here is the diff to produce the bad results:

diff --git a/var_elim/scripts/summarize_sweep_results.py b/var_elim/scripts/summarize_sweep_results.py
index f878913..9f8540f 100644
--- a/var_elim/scripts/summarize_sweep_results.py
+++ b/var_elim/scripts/summarize_sweep_results.py
@@ -51,7 +51,8 @@ def summarize_single_model(args, elim_names):
         
         # Avg solve time is calculated only over the successful instances
         # This calculation matches the one in the paper. 
-        ave_solvetime = sum(df[df["success"] == True]["solve-time"])/n_success
+        #ave_solvetime = sum(df[df["success"] == True]["solve-time"])/n_success
+        ave_solvetime = sum(df["solve-time"])/len(df["solve-time"])
         ave_elimtime = sum(df["elim-time"])/len(df["elim-time"])
 
         sweep_data["model"].append(args.model)