Training on several data sets

[manerotoni]

Hello, with the latest version of CATS, training on multiple datasets does not seem to work.

I did following steps.

• Open Image1
• put some labels
• Train Classifier
• Save ARFF file
• Close CATS
• Open Image 2
• Load labels from Image1
• Add some more labels
• Train classifier with both datasets checked
• If I want to do Apply classifier I get for each feature following error

Greetings

Antonio

[manerotoni]

CATS_error.zip This is a minimal data set that shows the error

[manerotoni]

There seems to be a work around. Not sure why it works like this. Somehow in the first round the labels for Image1 are corrupted?

• Load labels Image1
• Recompute labels
• apply classifier on ROI
• Load labels 2 (check both boxes)
• Apply classifier (now it works)

[tischi]

Thanks for reporting the error! I will try to have a look soon!

[tischi]

This is really funky though, because I did not touch the code for ages.

Maybe there is something special about this image data set?

Maybe the image is rather small?

Or maybe the image properties (c,t,z) are mixed up?

Or did you maybe play with the advanced settings? There is the option to "only compute difference features" and I think also another option to compute the "log2" of all features. However, normally CATS will examine the ARFF file and then adjust the advanced options according to what it finds in the ARFF file.

[manerotoni]

It is funky because I do not always get this error. For instance now I managed to add many training sets for the large data set. I think what could have happened is that the first ARFF file was corrupted? Can you reproduce the error?

I did create a small data set so that you can try it out, but the error occurred also with a large data set.

[tischi]

Hi @manerotoni, I have not used CATS for a long time. Thus a zoom would probably be the best to tackle this!