.md .pdf

repository open issue suggest edit

Contents

Week 6 — Two Variables (Sep. 27–Oct. 1)

Attention

The first midterm exam is on Tuesday.

This week’s learning outcomes are:

1. Display two potentially-related numeric variables for exploratory analysis.

2. Compute correlation coefficients between variables

3. Run a linear regression

Since we have the exam this week, the lecture load is significantly reduced.

🧐 Content Overview

Element	Length
🎥 Two Variables Intro	4m36s
🎥 Displaying Variables	3m45s
🎥 Correlation	11m12s
🎥 Regression	6m3s
🎥 Features	4m20s

This week has 0h30m of video and 0 words of assigned readings. This week’s videos are available in a Panopto folder and as a podcast.

📅 Deadlines

• Midterm A Tuesday 9:00–10:15 AM (in class)

• Quiz 6 Thursday by 8AM

🚩 Midterm A

The first midterm is on Tuesday. It is written to take about an hour, and covers material up through and including Week 5.

Format

The exam is a written in-person exam. It will contain a variety of questions to assess your ability to understand and apply concepts from the class. Question formats include:

• Multiple-choice

• True/false

• Matching

• Fill-in-the-blank

• Short answer

I may ask you to do a range of things on the exam, including (but not limited to):

• Define a concept

• Compute a metric from a small quantity of data

• Interpret a chart

Exam Rules

• You may have 1 note sheet, letter- or A4-sized, single-sided. (For the final, you will be allowed a two-sided note sheet.)

• You should not need a calculator, but may bring one if you wish.

• You may answer in either pen or pencil.

Study Tips

• Review the previous quizzes and assignments.

• Review lecture slides to see where you are unclear on concepts and need to review.

• Skim assigned readings, particularly the section headings to remind yourself what was in them.

• Review the course glossary, keeping in mind that it does contain terms we haven’t gotten to yet.

🎥 Introduction

This video introduces the week’s topic.

Video (4m36s)

Slides

CS 533INTRO TO DATA SCIENCE Michael Ekstrand TWO VARIABLES Learning Outcomes (Week) Explore relationships between two variables Understand correlation and covariance Photo by Sonny Ravesteijn on Unsplash We’ve Seen Computing within a group Paired t-tests Two-sample t-tests Both relate a numeric outcome to a categorical predictor. Paired Observations Observe more than one variable for the same object Top Critics rating All Critics rating # of MovieLens ratings Relating Observations Observations happen at different levels A student graduated, or did not The students at a school graduated at some rate The students of a particular race at a school graduated at some rate Predictions Often, we group two (or more) variables into two: Outcome variable describes a result Dependent Endogenous Predictor variable(s) are used to explain or predict outcomes Explanatory Independent Exogenous The Week Two Variables Midterm Exam (Wed. PM through Saturday, 72 hours) Covers through Week 5 Notebook + Conceptual Questions Concept Qs more specific than quizzes – apply! Wrapping Up A lot of interesting problems will have two (or more) variables per observation. We’re going to look at how to measure these variables’ relationships. Often an outcome and 1 or more predictors.

🎥 Displaying Variables

This video discusses how to display related numeric variables.

Video (3m45s)

Slides

CS 533INTRO TO DATA SCIENCE Michael Ekstrand DISPLAYING VARIABLES Learning Outcomes Display relationships between two variables for exploratory analysis. Extend this to more than two. Photo by Greg Jeanneau on Unsplash Scatterplots Remember them? Scatter plots show relationships between two variables: Each point is an observation One variable on X axis Other on Y axis sns.scatterplot plt.scatter Including Distributions Can include marginal distributions with scatter Histograms on margins (jointplot) “Rug plots” along axes Useful for exploratory analysis. Trend Lines - regplot Pairwise Correlations - pairplot Wrapping Up We want to explore the relationships between variables. Scatter plots and their variants let us do that, with possible augmentations.

🎥 Correlation

This video discusses how to compute the correlation coefficient between two variables.

Video (11m12s)

Slides

CS 533INTRO TO DATA SCIENCE Michael Ekstrand CORRELATION Learning Outcomes Compute correlation coefficient Understand the relationship between covariance and correlation Photo by Scott Webb on Unsplash Correlation We can see the relationship. But how correlated are the two variables? Variance Covariance Covariance Correlation Correlations Figure from Wikipedia. CC BY-SA 3.0. Correlation and Independence Estimating and Testing Wrapping Up Correlation measures the degree of linear relationship between two variables. It is a normalized version of covariance. Pandas: cov and cor.

Warning

In this video, I list the Pandas correlation function as cor. The correct name is corr.

🎥 Regression

This video discusses how to fit a line between two variables.

Video (6m3s)

Slides

CS 533INTRO TO DATA SCIENCE Michael Ekstrand REGRESSION Learning Outcomes Compute a linear regression with one predictor Photo by Andre Benz on Unsplash Trend Line Where did this come from? Variables and Model Regression Model coef – strength of relationship R2 – frac. of variance explained Ignore p-values and standard errors for now Wrapping Up Regression predicts one value with another. Lets us go beyond same-scale (paired t-test) to determine strength of relationship. Photo by Lucy-Claire on Unsplash

📓 Correlation Notebook

The correlation notebook shows how to compute the metrics in this week’s videos, and has the code I used to produce the charts in the slides.

🎥 Features

This video introduces the idea of feature engineering

Video (4m20s)

Slides

CS 533INTRO TO DATA SCIENCE Michael Ekstrand FEATURES Learning Outcomes Compute new features for observations Think about feature relatedness Preview: split data Photo by Logan Clark on Unsplash Features Penguins: Bill length Bill depth Flipper length Body mass Can compute: Bill aspect ratio Augments data with a new feature Featured Selection Things we’ll want to do: Extract features Select features Rescale, normalize, and transform features Combine features into new ones These can improve performance and interpretability. Desired Feature Properties Well-distributed Normal often easiest to work with Interpretable, if we’re doing inference Uncorrelated Two correlated features may not contribute more value! Cause problems with some models! Beware Don’t look at correlation between outcome and predictors in exploration! Biases your feature selection Hold out testing / validation data, then explore your “training data”. Wrapping Up We want to take care in curating and transforming features. We’re going to need to split data into pieces soon. Photo by James Fitzgerald on Unsplash

🚩 Week 6 Quiz

The Week 6 quiz is due before class on Thursday as usual.

📩 Assignment 3

Assignment 3 is due October 10.

previous

Week 5 — Filling In (9/20–24)

next

Week 7 — Getting Data (Oct. 4–8)