code and report

CSC 535/635 – HW 4

Perceptron Implementation Due Date: Friday, November 6, 2020 by 11:59 PM

Instructions: Students are required to work alone on this assignment.

Part 1 [60 points]: For this assignment, you will implement the Perceptron algorithm as given on Module 10’s PowerPoint slides on Blackboard. Assume that there are two classes labeled 1 and -1. Use the training data provided in the file training_data.txt. Use the testing data provided in the file testing_data.txt. The third attribute in the datasets is the class label. Use a threshold activation function, which will work fine for linearly separable datasets.

For each test object, output the object followed by the actual class label followed by the predicted class label. This will be followed by the output of the overall accuracy rate, followed by the learned parameters (weights and bias.) Sample program output is as follows:

[ 5.32 0.37] Actual label: 1.0 Predicted label: 1 [-4.06 -6.25] Actual label: 1.0 Predicted label: 1 [-1.17 -6.83] Actual label: 1.0 Predicted label: 1 [ 7.6 -4.52] Actual label: 1.0 Predicted label: 1 [-1.72 -4.08] Actual label: 1.0 Predicted label: 1 [ 2.58 1.6 ] Actual label: 1.0 Predicted label: 1 [ 2. -4.68] Actual label: 1.0 Predicted label: 1 [-4.31 -4.93] Actual label: 1.0 Predicted label: 1 [-3.45 -7.12] Actual label: 1.0 Predicted label: 1 [ 9.2 -6.23] Actual label: 1.0 Predicted label: 1 [-5.56 7.41] Actual label: -1.0 Predicted label: -1 [-5.87 8.37] Actual label: -1.0 Predicted label: -1 [-0.23 2.23] Actual label: -1.0 Predicted label: -1 [-8.39 4.77] Actual label: -1.0 Predicted label: -1 [-6.69 9.28] Actual label: -1.0 Predicted label: -1 [-9.51 -5.91] Actual label: -1.0 Predicted label: -1 [ 4. 5.59] Actual label: -1.0 Predicted label: -1 [-9.54 1.55] Actual label: -1.0 Predicted label: -1 [-9.97 0.31] Actual label: -1.0 Predicted label: -1 [ 2.8 9.71] Actual label: -1.0 Predicted label: -1

Accuracy rate: 100.00%

Learned weights are: [ 5.23962701 -5.18962127] Learned bias: 0.405526752143

Part 2 [20 points]: Use matplotlib to draw a scatter plot of the given training dataset. Show the objects in each class using a different color and/or marker. This dataset is linearly separable by a line that is not hard to figure out by looking at the plot and the model’s learned parameters. Provide a separate plot of the testing dataset and the learned hyperplane. An example of how to draw the hyperplane is shown in the SVM lecture slides.

Part 3 [20 points]: Write a report similar to previous homework assignments. Follow the format specified in the report template. In the report, you can talk about the stopping criteria you used for the algorithm.

Extra Credit [up to 20 points]: Find a real dataset and test your implementation on this dataset. You can search the web for linearly separable datasets or “almost” linearly separable datasets to use. Make sure to provide a URL and description of the dataset you used. Mention any modifications you may needed to do to make your code work for this dataset. If your dataset is not linearly separable, check if using a different activation function such as sigmoid helps.

What to turn in?

Upload to Blackboard copies of your source code and report describing the stopping criterion you used for the algorithm, the accuracy of the algorithm, and any comments about this homework assignment you may want to share. If you did the extra credit part, then make sure to talk about that in your report.

Please make sure that your code is well organized and properly documented and commented. Please use the Homework Report Template available on Blackboard to write your report.

Name your implementation file hw4.py or hw4.ipynb as appropriate. If you did part 2 in a separate file, name that hw4_plots.ipynb. Name the report hw4_Report.docx. Zip all the files together and upload your zipped file to Blackboard.