Hi all,
Trying to use several ML algorithms to see how they compare for Predictive Maintenance.
I'm using the (public) dataset from https://www.kaggle.com/behrad3d/nasa-cmaps .
I have most algorithms working, and comparing them is not a problem. For Neural Network, not so much. I'm getting the following error:
Error: Neural Network (185): Neural Network: Error in predLoess(object$y, object$x, newx = if (is.null(newdata)) object$x else if (is.data.frame(newdata)) as.matrix(model.frame(delete.response(terms(object)), :
Screengrab with full errorlog, with macro messages enabled: https://imgur.com/a/E8QLxQn
Running the latest version of Alteryx Designer (on a student licence) as admin: 2021.3.2.54175
I'm confused by the following conflicting information:
- The official help site ( https://help.alteryx.com/20213/designer/neural-network-tool ) says the target variable has to be a string.
- The sample workflow in Alteryx (Help --> Sample workflows --> Learn one tool at a time --> Predictive --> Neural Network) seems to agree, although it predicts the strings "Yes" or "No".
- Several other sources (example: https://community.alteryx.com/t5/Alteryx-Designer-Knowledge-Base/Tool-Mastery-Neural-Network/ta-p/30... ) say a numerical target variable is possible: "Neural networks are a predictive model that can estimate continuous or categorical variables."
In my own workflow, I'd like to predict a numerical value - somewhere between -5 and ~200, integers only at the moment. Is this not possible, or do I have to fiddle with settings somewhere?
Conclusion: I don't know what to believe anymore. 😉
I've attached my workflow. Tips, tricks and solutions are appreciated.
Solved! Go to Solution.
Hi @AxiestheCollector ,
there are a couple of things. First, you were using Target as both your predictor and target variable, which is naughty but wouldn't break anything.
The reason you were getting the error is the number of nodes required by adding the number of fields you have in your data. Effectively you didn't have enough nodes to handle the field density.
I've increased this to 30 and it's now working. You might need to tweak it down until you get the optimal number of nodes.
Also, you were then scoring the train dataset, so I've added the test dataset, created a Source field to split after the processing steps, so you should now have your answer.
Hope this helps,
M.
Awesome, thanks for the help. I remember changing the number of nodes, apparently just not high enough.
Any thoughts about my confusion mentioned in the topic?
(not sure if accepting your post as the solution closes the topic, so I'll do that later today or tomorrow)