I would like to have to save runs: if you press the play button, an entry will be made with start date, start time,end time.
Clicking on such an entry should show the output of the run (preferably in the workflow indicated by a kind of “output sink” all tables that link to this sink are part of the saved output). Double-clicking such an entry should reset each node to the used parameters for this run.
Reason for this is: while experimenting I change a parameter, I have to scribble down the output, manually change another parameter, comparing with the previous output(s) etc and finally switch back to the parameters settings with the best output. This takes a long time and many manual actions, the proposed wish should make this a lot easier.
One option would be to use the Parameter Optimization Loop to feed different parameters with each loop run and collect the results together with the parameters yielding the result. Afterwards you will get a set of parameters yielding the best performance (please check knime://EXAMPLES/04_Analytics/11_Optimization category on our EXAMPLES server for examples). However, it won’t allow you to dynamically adjust the parameters based on the result of the previous run.
Another option would be to incapsulate your nodes to a wrapped meta node and to use the quickform inputs to be able to input all the parameters at once (please check knime://EXAMPLES/09_Enterprise/02_WebPortal category on our EXAMPLES server for examples). However you will still have to input those parameters manually.
I know the parameter optimization loop, but in my workstyle I often run some experiments (trial&error) to see what are the most important factors. Your solution would fit in after I detemined by prototyping what to do.
My wish goes a bit further: if implemented well you also get an auditable sequence of actions and enhance reproduciability of analysis. Another leader in the magic quadrant (rapidminer) has this functionality (except for the double click part).