A single log does not always tell you the truth and you may want to aggregate multiple logs to see the "average behavior", like how quickly the optimality gap falls on average.
This of course requires that the logs are relatively similar as otherwise there is no use in this. If you try to aggregate the logs of different problems, you cannot expect to get anything useful out of it. However, imagine you are solving a scheduling problem every day and want to see the average behavior of CP-SAT or detect certain outliers.
One can start by overlaying the plots, but could also do boxplots of the other metrics, such as runtime, number of variables, etc. Could require to extend the parser first to allow to access more of the values.
A single log does not always tell you the truth and you may want to aggregate multiple logs to see the "average behavior", like how quickly the optimality gap falls on average.
This of course requires that the logs are relatively similar as otherwise there is no use in this. If you try to aggregate the logs of different problems, you cannot expect to get anything useful out of it. However, imagine you are solving a scheduling problem every day and want to see the average behavior of CP-SAT or detect certain outliers.
One can start by overlaying the plots, but could also do boxplots of the other metrics, such as runtime, number of variables, etc. Could require to extend the parser first to allow to access more of the values.