Deployment View

RapidMinerUser12

Hello,

I've gone through the videos, and I fail to see if the Deployment View can be used to deploy models that weren't made with Auto Model?
If they can, can you share how you did it?

Thank you in advance.

IngoRM

Hi, at the moment, the Deployments view only supports models generated by AM or by AM processes.  While it is technically possible to also add manually created models to the Deployments view, it is a rather complex process and will be simplified in the future.  The easiest way for now would be to modify an AM process, create the same results, and override an existing deployment.  No guarantees that this works in all cases though...

May I ask why you prefer a manual model over one created by Auto Model?  We are always eager to learn how we can improve Auto Model in the future.

Cheers,
Ingo

RapidMinerUser12

Hi Ingo, thank you for your reply.

Although AM is a really useful tool for swift model development, I prefer the manual model creation because it gives me more control over the parameters of the operators created in the process.
However, I really like the Deployments view. Can you explain more how to override an exiting deployment?

Like let's say I deploy a decision tree made in AM, and then I open that AM model in the design view process and edit it. How do I redeploy that process, the one edited in the Design View?

Thank you in advance.

IngoRM

Hi & thanks again for coming back.  Here is a follow up: can you please share a bit more details what you mean by "control over parameters"?  Is it more parameters you would like to tune?  Is it different ones?  Is it just trying out more or longer?  Or is it about the preprocessing?  I just try to understand what is missing from your point of view in AM, that's all.  Hope this makes sense...

"Like let's say I deploy a decision tree made in AM, and then I open that AM model in the design view process and edit it. How do I redeploy that process, the one edited in the Design View?"

We will definitely work on making this simpler for (semi-)manual models in the future, but for the time being this is the rough idea:

1) go once through a manual AM run on Server,

2) which will save the results in a repository you have defined, and

3) change the generated processes (one for each model) once by changing the data input of each process / parameters / or whatever you want to change as long as the same type of results are produced.

The advantage of using the Server remote execution is that all processes are already setup to store the results again in the repo.  If you would do the same with AM running in Studio and press Save Results at the end, the outcome is almost the same but you would need to change the processes by defining the correct output locations in the context for each process for all 20+ results.  Possible, but the Server approach is just simpler…

Hope this helps,
Ingo

RapidMinerUser12

Hi, thank you for the lengthy explanation.

What I mean by more control over parameters is that, for example, I can tune the pruning and max depth. etc by myself, that kind of gives me a feeling of being more in control. Of course I can then open the process from AM and edit everything manually, but it's a lot more work to find where everything is in the generated process than to generate everything by myself. But that's just because I'm used to developing in the design view.

Thank you.