The problem scenario of this assignment is from our optional textbook, Fundamentals of Machine Learning for Predictive Data Analytics, with substantial modification.

The European Space Agency wants to build a model to predict the amount of oxygen that an astronaut consumes when performing five minutes of intense physical work. The descriptive features for the model will be the age of the astronaut and their average heart rate throughout the work.

The regression model is:

OXYCON = w[0] + w[1] * HEALTH + w[2] * AGE + w[3] * HEARTRATE

Here is a copy of the same data in csv format: A4-data.csv.

Your tasks:

First, it's obvious that the attribute representing health status (HEALTH) is a categorical one and can't be used directly as a variable in the regression model. You need to design a scheme to transform this categorical attribute to a proper type that's suitable to be used in the model.
You can manually edit the input file, use the transformed value to replace the HEALTH value, and feed the modified input file to your program.

Then, write an error based learning program with your choice of programming language to tune the weights in the above given multivariate linear regression model.

Specifically, your program can set the following (adjustable) constants:

• the learning rate, at least the initial one, is 0.000002;
• the initial weights of the model are set as:

  w[0] = -59.5, w[1] = 5.5, w[2] = -0.15, and w[3] = 0.60;
  

• the iteration number is 50; and
• the acceptable threshold for the model error (the sum of squared errors) is 3.0 (calculated as 0.25 times the number of training instances).

The steps of your program should perform in one iteration are:

1. make a prediction for each training instance using the given model with the current weights;
2. calculate the sum of squared errors for the set of the predictions generated in the previous step as the model error;
3. adjust the weights based on the calculated model error from the previous step and the given learning rate using the gradient descent algorithm;

Repeat the above steps until either the designated iteration number is reached, or the calculated model error is below the given acceptable threshold.

for each iteration, display the model error and the adjusted weights of the model, in an easy to understand format. At the end of the iterations (end of your program), display the original data with an added column that shows your model's prediction.

Lastly, write a document that explains at least the following things:

• your design of transforming HEALTH to a type that's suitable to be used in a linear regression model;
• how to execute your program on csci server;
• if you are not using the suggested constants, then show the learning rate and initial weights used in your program, and explain why you'd like to make the adjustment;
• the final weights and the model error after tuning is done;
• any thing you'd like to bring to the attention regarding your program and/or your model.

What to Submit:

• The document;
• The modified input file;
• The source code file of the error based learning program;
• A sample run output of your program;
• Makefile if one should be used to automate the process of compile and execute your program.

How to submit:

Choose one of the following two ways to submit your work:

• Login to your VIU Learn account, find the CSCI 479 course page, click on the "Assessment" drop-down menu, click on the "Assignments" item, then click on the folder named "A4". Then you can click on the "Add a File" button to browse and upload your document and other files.
• On csci server, in the directory that holds all of your assignment solution files, enter the command
  ~liuh/bin/submit 479 A4 .
  This submit script currently accepts files with the names of *.pdf, *.txt, *.csv, *.h, *.cpp, *.py, makefile.
  If you need to submit any file with different extension names, please contact your instructor before submitting.