Machine learning manager exchange resource and how manufacturing can be helped by machine learning

Ignition Exchange
21

With just entering in to using ML, I used clustering with k-means++ and the output I got is what I expected. I need to definately read up more before I venture out trying other algorithms.

There is a function still not working but from what I gather an update is coming out soon to this exchange project. I too am also waiting for the companion guide to make it's debut.

1 Like
22

I'm putting this here, because I think I have a related problem.

I'm currently testing the Machine Learning Manager. I have trained a regression model, and am trying to connect a tag in the designer to my models prediction output - based on https://ml-guide.ia.io/using-model/ and Using Ignition with Machine Learning Libraries | Inductive Automation.

I have set up the gateway project as the project I am working on (as well as drilling down to the ml module - i.e. projectname/exchange/ml) and I now have an issue, where my expression tag gives an error:

Error_ExpressionEval

My tag expression is:

runScript(
'exchange.ml.hook.models.processModelByTagPath',
1000,
'MyModel',
'[default]ML/rand0')

Simultaneously, if I run the processModelByTagPath function in the script editor, I get a 'None' value result.

After looking deeper into the function itself, it seems that the issue is that the GlobalHandler.get() method is returning an empty object.

model = GlobalHandler(name).get()

If I independently access the database, then I can see that my model has been stored in the machinelearning table:

image
image896×136 35.3 KB

I am not sure how best to proceed, and any help/advice would be ideal.

23

Did you ever get this resolved? I’m facing the same issue. If I use views provided in the manager, I can create and train and test models, but if I use the script functions they fail. When I try exchange.ml.hook.models.existing() it returns the models but all say they are untrained even though they show a trained state in the GUI. Any idea?

24

Hi Steve,

No, unfortunately not, and the course I was working on this for has since been reduced in size such that we don’t cover ML modelling anymore.

At the moment, I’m of the opinion that it’s best to keep the modelling workflow separate and upstream of the MES/SCADA, and then manually transfer model coefficients down to where they’re needed.