New wishlist item? -- Better looping

KNIME Analytics Platform
1

At the moment, as far as I know, the looper only repeats the nodes within it with identical parameters, so the only variation you can get is with randomisation.
I'd like some way to pass a list of parameters to one of the nodes with in the looper, so
that I can, for example, build models with different parameters and test the outcome. At the moment it is necessary to cut and paste the workflow for every parameter variation.

Thanks,

Dave

2

Hi Dave,

version 2.0 will have support for (loop) variables, which will help you to accomplish that sort of things. (As Michael already pointed out in another thread the framework will support these variables, but that doesn't necessarily mean that all of our nodes already know what to do with them.)

As for 1.3.x I would suggest to use the BatchExecutor instead of the looper node to run your experiments.

Regards
Bernd

3

Thanks Bernd,

Do you have any further descriptions or even screenshots of how these variables will work? It's hard to summarize my requests here (probably because I don't understand the current workflow manager in any depth) , but I gues I'd liketo be able to manipulate input parameters more flexibly in the same way as you can do in the batch executor, but within a Knime workflow. It doesn't sound like a separate set of global variables will acheive this, but I don't really understand the details.

Thanks

Dave

4

Hi Dave,

we plan to pass variables along the pipeline (independent branches of the flow don't share the same variables). In particular we will allow individual nodes to add (or read) variables to (from) a stack that we pass around. Think of a file reader node, which adds a (string-)variable containing the URL it reads from; nodes further down the pipeline will then be able to read this variable (for instance the CSV writer will be able to write to a file whose name is composed of that URL). We also plan to have dedicated nodes, which enable the user to set or change variables ... and -- to address your concerns -- we will have looper nodes, in which the user provides sets of variables (and in each iteration it uses another variable configuration).

But once again: 2.0 will have basic support for these variables, that does not mean that we have changed all of the available nodes to make use of these variables.

Regards
Bernd

5

wiswedel wrote:
Hi Dave,

we plan to pass variables along the pipeline (independent branches of the flow don't share the same variables).

This is VERY exciting news!!

-Kirk