KNN regression
Introduction
You and your group will implement KNN regression. I will break down the process for you below. Your group will build on this assignment over the next 2-3 weeks as we learn more content. Therefore, it is important that you communicate with and contribute to your group! Asking clarifying questions, explaining your thought process, throwing ideas out there, etc. all count as contributing.
There will be a series of small deliverables roughly 1-2 times a week. Even though you are working as a group, I want you to submit your deliverables individually. This is so you can continue to develop your skills as a coder and work at a pace that’s more comfortable for you!
Group introductions
Introduce yourself: name, majors, anything at all
Talk about how coding in R (base R or tidyverse) is going for you so well. Some people in the group may be more nervous about or proficient in coding, but I think it’s good to name it early on.
Find a time and place to meet outside of class this week
Step 1: compare pseudocode
Compare your pseudocode for implementing KNN regression. This could look like:
Having each person summarise the “order of events” in their pseudocode, before diving into each component
Having someone talk through their pseudocode from beginning to end, then having the other members discuss what they did similarly or different
Discussing what you were confused about
Step 2: write a group pseudocode
Now that you’ve compared your individual pseudocode, I want you write a new/final version as a group. Once you are happy with your pseudocode, call Becky over to have her check it.
Step 3: implement KNN
Once Becky has confirmed your pseudocode is detailed enough, your group can begin coding the implementation of KNN regression. The real fun begins!
Clone the GitHub project called knn_regression and work in the file knn_regression_implementation.Rmd.
Your final implementation must be a function called .knn() (note the period) that takes in the “minimal amount” of inputs and returns a vector of the predictions for the test data. “Minimal amount” means only the arguments that cannot be obtained/derived within the function itself (i.e. what the user of your function must specify in order to make your function run).
You may assume that the user only wants Euclidean distance for this implementation.
Deliverable
Once you’ve finished your implementation, check it by seeing if you get the same results as I did for the non-standardized LRUG mites data in the KNN regression slides.
Specifically, use your brand new .knn() function and demonstrate that you obtain the same predictions as I did for the same set of test data (plus or minus some rounding error). Then, use your function to predict for a different choice of \(K\).
General advice
Test your code frequently to make sure it’s doing what you want it to do! This is especially true of any functions that you create.
- This means you might want to create your train / test data now
Start “big picture”:
Begin by defining your function and its arguments using
function(). Remember, you can always add to the arguments as you progress through your coding if you realize your function requires additional inputs.If you need to iterate the same procedure many times, begin by writing out the loop and then add comments for each step within the body of the for loop. Then start coding each step.
Write functions that you can call within your implementation.
Be as reproducible as possible. Try to avoid “hard-coding” if you can. This means you should not use specific variable names or indices. You want to be able to use your implementation for any future data set!
If you get stuck, please ask Becky or Doug for help!
Submission
Once you’ve finished, push your changes to GitHub. Upload a PDF of your implementation and your confirmatory “code check” to Canvas.