Missing per output of "Loop Parameters" operator
Hi,
Sorry, I'm rigorous but I discovered that the Loop Parameters operator has a 'per' output
in its inner subprocess, but have not 'per' output outside this operator (only 'out' outports).
It can be a little confusing, according to the output we connect, we obtain in fine :
a performance table if connected to the first 'out' output of the operator
or
a collection of performance vectors if connected at an other 'out' output port (2nd, 3th etc.) of the operator.
Thanks you,
Regards,
Lionel
Best Answer
-
jczogalla Employee-RapidMiner, Member Posts: 144 RM Engineering
Hi Lionel,
the inner performance port was added in 7.4 when it was rewritten to be parallelized, together with Parameter Optimization (Grid). If you look into the documentation of Loop Parameters, you will find the parameters log_performance and log_all_criteria. They describe that if the inner performance port is connected and the first parameter is set to true, the performance table that you mentioned will be shown in the Results view. You can extract this data with the Log to Data Operator, using the name of your Loop Parameters Operator as the log_name parameter.
As for connecting the performance vector to another port, this will act like any other Loop Operator and collect the performance vectors of all iterations into a collection.
I can see where this could be confusing, but that's also the reason why the 'per' is marked as optional (half full) and in Performance Vector color.
Regards
Jan
1
Answers
hi @lionelderkrikor - so that's an interesting observation. As you can see from the Help section, it's intended to be very similar to Optimize Parameters. That said, I could see some value in having a second output port that is a collection of performances in addition to the first output port which is a collection of exa. Feel free to put something in the ideas section if you like.
Scott
Hi Jan,
Thanks you for these informations. The operation of 'per' output of Loop Parameters operator is perfectly clear for me now.
Best regards,
Lionel