Description
• Load the dataset “fisheriris” into the workspace.
o Study the dataset in terms of (a) Number of classes, (b) Number of features, and
(c) What the data represents, i.e., gain some intuition about the problem domain.
Based on your study, would you expect the features to perform well in this
problem?
• Compute the following quantities for each feature. Do you observe anything of interest
from these statistics?
Sepal Length Sepal Width Pedal Length Pedal Width
Minimum
Maximum
Mean
Variance
Within-Class
Variance
𝑠𝑠𝑠𝑠(𝑖𝑖) = ∑ 𝑃𝑃𝑗𝑗𝜎𝜎𝑗𝑗𝑗𝑗
𝑀𝑀
𝑗𝑗=1 , where 𝜎𝜎𝑗𝑗𝑗𝑗 is variance of i-th feature in class j,
and 𝑃𝑃𝑗𝑗 is a-prior probability of class j
Between-Class
Variance
𝑠𝑠𝑠𝑠(𝑖𝑖) = ∑ 𝑃𝑃𝑗𝑗(𝜇𝜇𝑗𝑗𝑗𝑗 − 𝜇𝜇𝑖𝑖) 𝑀𝑀 2
𝑗𝑗=1 , where 𝜇𝜇𝑗𝑗𝑗𝑗 is mean of i-th feature in class
j, and 𝜇𝜇𝑖𝑖 is the mean of the i-th feature
• Compute and display the correlation coefficients exactly as shown below (left figure).
Do you observe anything interesting from this display?
• Display each of the four features versus the class label, exactly as shown below (right
figure). What can you state about how well the features may perform in classification?
• Perform the following classification tasks.
Setosa Vs. Versi+Virigi All Features Batch_Perceptron and LS
Setosa Vs. Versi+Virigi Features 3 and 4 Only Batch_Perceptron and LS
Virgi Vs. Versi+Setosa All Features Batch_Perceptron and LS
Virgi Vs. Versi+Setosa Features 3 and 4 Only Batch_Perceptron and LS
Setosa Vs. Versi Vs. Virigi Features 3 and 4 Only Multiclass LS
• For each case, (a) report whether the method converged, (b) No. of epochs, (c)
Computed weight vector, (d) No. of training misclassifications, and whenever
appropriate, (e) plot of feature vectors, as well as the computed decision boundary.
Upload your m-file and report to Blackboard before midnight Friday, Feb 15.
