Is there a maximum limit to the number of features given as input to a PNN? And how to find the optimum spread value?

Hello, I am trying to classify ECG arrhythmia conditions using the PNN . Right now I'm considering 5 classes. I have 52 features in the feature matrix. What is the maximum number of features I can give to the PNN? IS there any limit like that? Also how to select the spread value for the PNN? I havent tried implementing any neuralnetworks till now. This is my first attempt. Please forgive if my query is silly.

Answer 1

[Walter Roberson]

The practical limit on number of features is determined by memory. The theoretical limit... Ummm, if you are training on gpu then it might be 2^14 = 16384 in some cases, as matrix that large on each side would be 2^28 elements and at 4 bytes each that would be 2^32 bytes (2 gigabytes) which might be the address limit on some older gpu.

[SSG_newbiecoder]

So, I have to reduce the feature set and bring it down to a lesser number? 52 features will be too much right?

[Walter Roberson]

No. I said 2^14 which is 16384. 52 is not much. You could potentially run out of memory if you had tens of thousands of images with 52 features, perhaps.

[SSG_newbiecoder]

Ok great. Thank you:)