Neural Networks on Weka

DS 352 Syllabus

last updated 08-Nov-2020

Objectives

1. Explore the construction of an artificial neural network

2. Use Weka once more to reinforce data preparation and parameter selection.

3. Build a neural network with no hidden layers and then 1  or 2 hidden layers.

4. Experiment with model building and the parameters of the multi-level perceptron algorithm.

 

Tasks

Project 1:

Locate iris data (in the default data directory). Use the Multi-layered Perceptron classification algorithm in Weka to use for classification.

Experiment with different network architectures. Set the hidden layers to 0. How good is that model compared to Weka's default?

Explore other hidden layers than the default. Are simpler ones just as effective or can a slightly more complex one be better?

Check the confusion matrix for verification of your model. 

Train with the "Use training set"  and "cross-validation".  The latter requires folds to be less than or equal to the number of instances.

 

Project 2:

Load the example majority of 3 bits from the class and configure a neural net that can classify correctly.  It can be done!

You must convert the classification variable from numeric to nominal for correctness. 

Weka's perceptron algorithm does not use sign or sigmoid "rounding" but when the classification variable is nominal, it will be effective.  The nominal classification will also generate a confusion matrix. Check the confusion matrix for verification of your model. 

You can copy/paste this into a csv file for import into Weka.

x1,x2,x3,y
1, 0, 0, -1
1, 0, 1, 1
1, 1, 0, 1
1, 1, 1, 1
0, 0, 1, -1
0, 1, 0, -1
0, 1, 1, 1
0, 0, 0,-1

Project 3:

For further experimentation, try to configure a Weka network to predict the XOR (which is non-linearly separable). Show me what you tried.

Data for copying and pasting:

x1,x2,y
0,0,-1
0,1,1
1,0,1
1,1,-1

Deliverables

Create a Word or PDF of the Weka results of your favorite experiments and your observations.

Submit to Moodle