CST 383 - Intro to Data Science | Week 5

Learning log 4:

This week we learned more about how important preprocessing is before building a machine learning model. One thing that stood out to me is that missing data is not always obvious. Sometimes it shows up as actual NA values, but other times it can be hidden as values like 0 or “information requested,” depending on the dataset. That made me realize that cleaning data is not just a technical step, but also requires thinking carefully about what the values actually mean.

I also learned why scaling matters, especially for models like KNN. Since KNN uses distance to compare points, features with larger numbers can have too much influence if the data is not scaled. This helped me understand why preprocessing and modeling are connected instead of being separate tasks.

The test/train split and cross validation topics were useful too. I understand that the test set should be saved for checking how well the model works on new data, while cross validation helps compare models or hyperparameters more reliably. One concept I am still trying to fully get comfortable with is how to choose the best value of k in KNN and how much I should trust cross validation results compared to the final test accuracy.

This week helped me see the full workflow better:

  • clean the data
  • scale it when needed
  • train a model
  • then evaluate it carefully

I feel like I am kind of starting to understand why each step matters instead of just memorizing code.

Comments

Post a Comment

Popular posts from this blog

CST 334 - Week 3

Week 3 Summary: This week in CST 334 was a rollercoaster. We started working on PA3, which is all about building a memory allocator. Basically our own version of malloc() and free(). At first, I thought, “okay cool, pointers and structs again,” but this assignment humbled me quick. Managing memory manually, keeping track of chunks, splitting them, and then coalescing them back together sounds simple on paper but actually implementing it without causing segmentation faults or crashing everything was tough. One thing that stood out was just how fragile everything becomes when you're directly manipulating memory. A wrong pointer or forgetting to update a linked list will break the whole thing. A lot of my time was spent debugging and going through each line of my code just to figure out where things were going sideways. The split_node function specifically gave me a hard time. It took me a while to realize how important it is to keep the fwd and bwd pointers clean or everything falls ...
Read more

CST 334 - Week 2

Week 2 Summary:  This week in CST 334, we dug into how operating systems handle processes, and honestly, things are starting to make more sense now. I’ve heard the term “context switching” before, but I never really got what it meant until we broke it down. Learning the difference between voluntary and involuntary switches like when a process is just waiting for input vs. when the OS forces it to stop, helped a lot. It’s kind of wild to think about how much the OS is doing behind the scenes just to keep everything running without us even noticing.  We also got into how processes are created in C with fork(), exec(), and wait(). The fork() function really threw me off at first. I was like is the same program just running twice? But it made sense after printing out a bunch of values and looking at the return codes. It’s pretty cool how the parent and child processes end up doing different things, even though they start from the same place. Later in the week, we went over schedul...
Read more

CST 370-30 - Algorithm Design & Analysis Week 1

 CST 370-30 - Algorithm Design & Analysis Week 1 This week in CST 370, we went over the basics of algorithm design and analysis and what to expect from the course. A big focus was on understanding how to think about problems in terms of best case and worst case scenarios, not just getting an answer. We worked on logic puzzles like the fake coin and ball matching problems, which helped me see how algorithms can be about strategy and guarantees rather than code. These problems showed how careful reasoning can minimize steps while still covering all cases, which I think will be important later in the class.
Read more