Neural Net Topology and Prediction in RM Studio 7.4

Hi cakman, 

I assume you are using the Cross Validation operator.

To see the model (the Neural Net), you need to connect the "mod" output of the Cross Validation to the process output.

To see the predictions, the "tes" output must be connected.

Regards,

Marius

Hi Marius,

Thank you for yr kind reply.

I used Split validation, anyhow tried the steps you mentioned, but now faced another problem in my data set (from repository). 

The system detected a problem and the error message read as "this may not be portable when sharing process"

I really don't understand, I made every steps successfully in Rapid Miner 7.3 without any error messages.

Now in Rapid Miner 7.4 many error messages that I cannot clear. See most of the features have been changed.

I really don't know how to fix the error message

Have a nice day.

Ceren

That error relates to how you set up your process and data repository. It's just giving a warning to say that if you were to share this process and data, it's not in a relative path type of format and may break on some other users PC. Nothing to worry about unless you are collaborating with others. 

Dear Thomas,

Many thanks for yr kind reply.

I fixed the problem however now facing with another difficutly.

I have 62 months data set and tried to prdict 63th month.

I run all the processes from yr videos successfully but cannot see any 63th month data.

I know I'm missing a point, but could not find it. I need to show it on my PhD thesis.

I'd be glad if you'd assist.

Thanks you for yr kind concern.

Have a nice day.

BRgds,

Ceren

Ok, is this a time series problem? If yes, are you using the Sliding Window Validation operator?

Take a look at the sample process I posted here: http://community.rapidminer.com/t5/RapidMiner-Studio-Forum/Recall-Error/m-p/37302#U37302

That will do point forecasting on a time series. I used an SVM but you can swap it out for a ANN.

Dear Thomas,

Yes this is a forecasting problem, and I used sliding window validation and forecasting performance.

I utilized many of your videos, appreciate them; very helpful.

I tried to divide the data set manually as training and test then did the processes for forecasting.

 Here attached you can find the process I run.

I checked the link that you sent, but I haven't tried a loop operator before.

Is there any saved repository for the link you have sent or any video that I can watch to have an idea?

Brgds,

Ceren

That process is one I created from Bala Depreshande's book, Chapter 10 I think. Looping operator's are fantastic and in that process it takes the last date and then generates new dates with the predictions. So if  the horizon is 5, it would take generate 5 new predictions for the next 5 days. It's pretty cool but heavily reliant on order of execution of the operators. 

Dear Thomas,

Thank you so much for the refences.

Have a nice day.

Ceren