sample
Joram Mutenge
2025-07-10
Analyzing a very large dataset can be strenuous on your computer, especially if you don’t have a fast machine. Instead of overworking your computer and waiting a long time to see the results, you can perform your analysis on a smaller subset of the data. Below is a dataframe with 1000 rows.
| category | quantity |
|---|---|
| str | i64 |
| "Hat" | 1 |
| "Sweater" | 9 |
| "Sweater" | 12 |
| "Sweater" | 5 |
| "Sweater" | 19 |
| … | … |
| "Sweater" | 2 |
| "Socks" | 19 |
| "Socks" | 18 |
| "Sweater" | 15 |
| "Sweater" | 11 |
Say we want to reduce the size of the dataset to 100 rows for faster analysis. A poor way to get those 100 rows would be to select the first or last 100 rows. Why? Because they don’t provide a true representation of the entire dataset.
For example, if the dataset had null values in the last 100 rows of the quantity column, you might wrongly assume that the entire column contains missing values. Conversely, if you only selected the first 100 rows, you could completely miss those null values.
To get a true representation of the data, you should use the sample function. This ensures that rows are randomly selected until you reach the desired number. Here’s how to use it:
| category | quantity |
|---|---|
| str | i64 |
| "Sweater" | 11 |
| "Hat" | 5 |
| "Socks" | 15 |
| "Sweater" | 5 |
| "Socks" | 1 |
| … | … |
| "Sweater" | 11 |
| "Sweater" | 2 |
| "Socks" | 11 |
| "Sweater" | 9 |
| "Socks" | 13 |
If you want your results to be reproducible, you can use the seed parameter.
| category | quantity |
|---|---|
| str | i64 |
| "Sweater" | 2 |
| "Sweater" | 19 |
| "Socks" | 10 |
| "Sweater" | 19 |
| "Sweater" | 3 |
| … | … |
| "Socks" | 19 |
| "Sweater" | 14 |
| "Hat" | 11 |
| "Sweater" | 1 |
| "Hat" | 19 |
Now, whenever you run the code, you’ll always get the same 100 rows.
Learn more in this Polars course.