Watchdog Agent Prognostics Toolkit

[andy88]

Hi,

You can find all the examples of this toolkit in the NI example list. Just click Help->Find Examples...->Directory Structure->Watchdog Agent Toolkit

Thanks

[k.eto]

Hi all,

I have a question regarding its renewal pricing policy.

Does this product need to be renewed every year to get the latest version?

If so, what would be the renewal pricing policy? does this follow other NI software products' renewal pricing policy? such as 20% of software's list price for ontime renewal.

Regards,

Kei

[nzamora]

I want to use the BP Neural Network for some classifying in a project. I've used the Machine learning toolkit a bit but it seems to fall short in regards to novelty detection. The BP and SVM functions of this toolkit seem to operate on the same principles as the MLT:  train with a dataset with known classes then evaluate a new piece of data using the solution genereated from training to return a class that was learned. My issue right now is I don't understand the classes that feed into the Neural Network Training.VI. The example uses what appears to be a 2D array where each row is a binary number. The notes in the example aren't super clear, so I'm a bit confused by them. Hopefully someone with a bit more knowledge of this tools can chime in and help me out.

[Chuan]

Both Neural Network and Support Vector Machine are supervised learning methods. They are used when we know the training labels(classes). In your case if you do not know the class of your data, what you can do is to try some unsupervised clustering methods, for exmaple K-means in MLT or Self Organizing Map in Watchdog Agent Prognostics Toolkit. Check out their help files to see applicable situations, pros and cons. 

When applying unsupervised clustering methods, abnormal data/feature can be identified by doing 'peer comparision'. For instance in machine condition monitoring, healthy features are aggregated into one cluster, while faulty features can be separated from that cluster.

Good Luck!

[nzamora]

Thanks for the reply. I already understand the general workings of BP and SVM. What confuses me is the method of applying a class to a row of data. I know that when training I have to provide a 2D array of doubles where each column is a feature, and each row is a distinct instance of those features (object 1, 2, 3...). Paired with that is a 2D array of doubles for the classes where each row corrosponds to a row of features. But I do not understand the formatting of this array. The example uses a binary breakdown of values:

[[0,0],

 [0,1],

 [1,0],

 [1,1]]

This seems really counter intuitive to me rather than just using integer values in a 1D array:

[[0],

[1],

[2],

[3]]

I need some explenation of why this is the way it is, or how this array is suposed to be formatted. Also, I'd like to know what the output of the testing vi is supposed to mean.

[Chuan]

In the help of NN it says:

Training Target: Target Neuron Network output for training data, m-by-j array. m equals to training sample number and j equals to the number of output layer neurons.

This way of formatting is usually used in NN becuase there might be multiple neurons in the output layer. Each neuron in the output layer outputs a value. In your particular case there are 2 neurons in the output layer. So [0,0], [0,1], [1,0], [1,1] correspond to class 1,2,3,4, respectively.

[nzamora]

Does this mean that the more classes I have, the more output neurons are [i]required[/i]? If I had 17([1,0,0,0,1]) classes, I'd need 5 output neurons? What would be wrong with leaving the padding 0's and using something like

[[0],[0],[0],[0],[0],[0],[0],[0],

 [0],[0],[0],[0],[0],[0],[0],[1],

 [0],[0],[0],[0],[0],[0],[1],[0],

 [0],[0],[0],[0],[0],[0],[1],[1]]

?

[Chuan]

Basically yes.

That would be very difficult for NN to classify. It takes much longer to run and over fitting is likely to occur.

[胖胖龙]

Hi,

I'm learning SOM (self-organization Feature Maps), so I want to ask that is there SOM VI in the Watchdog Agent Prognostics Toolkit? Thanks.

[andy88]

Yes. It has