Skip to main content

Recitation 4 — Data Cleaning & Transformation

Skills: 2​

Reading: 4.2​

Examining Messy Data​

Here is an intentionally messy table:

SURVEY-RAW = table: name, age, favorite-food, rating, state
    row: "Alice", "19", "pizza", "8", "MA"
    row: "Bob", "20", "PIZZA", "9", "ma"
    row: "Carol", "", "tacos", "7", "California"
    row: "David", "18", "Pizza", "ten", "Mass"
    row: "Eve", "21", "sushi", "", "MA"
    row: "Frank", "19", "ice cream", "6", "massachusetts"
    row: "Grace", "twenty", "Pizza", "8", "CA"
    row: "Henry", "19", "ICE CREAM", "5", "calif"
end

Identifying Data Quality Issues​

First, let's look at the data and identify problems (Inconsistent capitalization, Missing values, Inconsistent formats, Inconsistent state abbreviations)

Discussion: "What questions would be hard to answer?"

Cleaning Text Data​

Standardizing Case​

Can normalize text using string functions -- can build-column a new column using string-to-lower of existing favorite-food column.

Check Understanding: "Why might we want all food names in lowercase?"

Standardizing State Names​

Create a more complex transformation for states -- design a function that first uses string-to-lower but then checks using if if the lowercased version is one of the ma variants or one of the ca variants and replaces each with a standard version.

Handling Missing and Invalid Data​

Dealing with Missing Ages​

What different strategies can we use for missing ages?

  • Can remove the data. In this case, create a new table that only includes data where the age is present.
  • Can put a default age. In this case, add a new column with a cleaned age column, where you fill in a default age (say, 20) in case it was missing.

Discussion: Ask, "When might you choose between filtering vs. filling missing data?"

Converting Text Numbers to Actual Numbers​

If people wrote words instead of numbers, we can convert them. If they were arbitrary numbers, this could be complex, but if we expect numbers to only be 1-10 (e.g., in the rating column), then can write a large if ... else if... else... with cases for "", "ten", "nine", ..., and then the last case we can just convert from a number like "8" to a number 8 with string-to-number.

Design such a helper function, and use it to add a rating-clean column.

Check Understanding: Ask, "Why should we use -1 for missing ratings instead of 0?"

Computing New Columns​

Sometimes we might also want to create computed columns based on existing data. For example, maybe we want a rating-category column that distinguishes between "no rating" (-1, based on previous cleaning), "high" (above 7), medium (above 5), and low.

Add such a column.

Wrap-Up​

  • Incremental Cleaning: Building columns step by step rather than trying to fix everything at once
  • Trade-offs: Different strategies for missing data have different implications

Reflection Questions:​

  • "What would happen if we tried to analyze the original messy data?"