🎉 🎉 RAPIDMINER 9.10 IS OUT!!! 🎉🎉
Download the latest version helping analytics teams accelerate time-to-value for streaming and IIOT use cases.
Data structure in classification task with Neural Net
I used RapidMiner for a student project at my university. First, I think it would be helpful to briefly describe my problem. So here we go:
I have machine / process data (Force in kN) extracted from a manufacturing machine. There is one set of data for each manufacturing cycle, stored in a Excel Table. Each row contains the Force-Values for one cycle. Each cycle takes 100s.
Right now, I am handling (fictional) data with 20 different cycles (= 20 rows in xls) and 100 individual Force-values for each cycle.
(the amount of data is going to increase dramatically when in use - up to 5000 individual values per cycle)
The goal is to classify each machine cycle with the help of the recorded Force-data. The algorithm I am using is a Neural Net with Backpropagation.
First question: Is it really wise to structure the data this way and feed it into RapidMiner? My first attempts resulted in a Neural Net with 100 input Neurons, which is logical but I am wondering, if there's a better way of pre-processing the data?
Second question: I am following the "Right Way to Validate Models"-Guide by Ingo Mierswa, but didn't really get far.
edit: I found my mistake - the "Test" Data I was trying to get results for is not really test data, but really unknown data with no classification - that's why no result was found.
When trying to calculate the test error, my classification accuracy is "unknown" and there's just no results. So I am assuming a very stupid, basic mistake. Can anyone help me?
Here's how i build my process: (multiply is enabled for training but disabled for retrieving the "unknown" - 5V_1D_dataset.
(I also found out I have to remove the label for the training data - so I added the Select Attributes-Operator after Multiply)
Thank you so much in advance.