Introduction

This tutorial assumes that you have already completed the “Getting Started” tutorial in the tutorial.helpers package. If you haven’t, do so now. It is quick!

Installing Python on your machine is sometimes a complex process. Feel free to do so but, for this tutorial, we will work with Python in the cloud using GitHub Codespaces.

Sign up for GitHub.

Learn how to explore a data set using functions like describe(), info(), and sample().

Exercise 1

Before you start doing data science, you must import the libraries you are going to use. Use the import statement to load the polars library with the alias pl. This is the standard convention in Python. Click “Run Code.” The check mark which appears next to “Exercise 1” above indicates that you have submitted your answer.

import ... as pl

In Python, we import libraries to access their functions and data. The import statement makes the library available, and using as pl creates a short alias so we can type pl instead of polars every time.

Exercise 2

In this tutorial, you will sometimes enter code into the exercise blocks, as you did above. But we will also ask you to run code in the Python Console. (You will do this in the other Positron window, since the Console in this window is currently running this tutorial.) Example:

In the Python Console, run import polars as pl.

With Console questions, we will usually ask you to Copy/Paste the Command/Response into an answer block, like the one below. We usually shorten those instructions as CP/CR. Do that now.

Your answer should look like:

>>> import polars as pl
>>>

Your answer never needs to match ours perfectly. Our goal is just to ensure that you are actually following the instructions.

Exercise 3

DataFrames are spreadsheet-like data structures in Polars. Let’s load the famous iris dataset. We’ll read it from a URL.

In the Console, run:

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris

CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris

Whenever we show outputs like this after a question, then we are showing our answer to the previous question, even if we do not label it as such.

Exercise 4

In the Console, run iris.describe(). This provides summary statistics for numerical columns.

CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.describe()

This method provides a quick statistical overview of each numerical variable in the dataset. In some cases, the tutorial displays the same object differently from what you were able to copy/paste. And that is OK! Your answer does not need to match our answer.

Exercise 5

In the Console, run iris.sample(). This selects a random row from the dataset.

CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.sample()

Your answer will differ from this answer because of the inherent randomness in methods like sample().

Exercise 6

In the Console, hit the Up Arrow to retrieve the previous command. Edit it to add the argument n = 4 to iris.sample(). This will return 4 random rows from the iris dataset.

CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.sample(n = 4)

Editing code directly in the Console quickly becomes annoying. See the positron.tutorials package for tutorials about using Positron to write and organize your code.

Exercise 7

In the Console, run print(iris). This returns the same result as typing iris.

CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
print(iris)

You can control how many rows to display using iris.head(n) for the first n rows or iris.tail(n) for the last n rows.

Exercise 8

In the Console, run iris.head(3). This returns the first 3 rows of the iris dataset.

CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.head(3)

head() by default gives the top of the DataFrame, so your answer should match our answer. sample(), on the other hand, picks random rows to return. But, in both cases, the result is a DataFrame.

A central organizing principle of Polars is that most methods take a DataFrame and return a DataFrame. This allows us to “chain” commands together, one after the other, creating a pipeline very similar to R’s pipe operator |>.

Exercise 9

In the Console, run help(iris) or iris.info().

The info() method will show you information about the DataFrame including column names, data types, and memory usage.

Copy/paste the output of iris.info() below.

You can find help about pandas functions with help(function_name) or by visiting the pandas documentation at https://pandas.pydata.org/.

Exercise 10

In the Console, run iris.schema. This shows the column names and data types. CP/CR.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.schema

The schema attribute displays information about the DataFrame’s structure including the data types of each column. For example, sepal_length is listed as Float64, meaning it’s a 64-bit floating-point number. You can also use iris.dtypes to see just the data types, or iris.columns to see just the column names.

Exercise 11

In the Console, run import math then math.sqrt(144).

CP/CR.

import math
math.sqrt(144)

The square root function is one of many built-in functions in Python’s math module. Most return their result, which Python then, by default, prints out.

Exercise 12

In the Console, run x = math.sqrt(144).

CP/CR.

import math
x = math.sqrt(144)

The = symbol is the assignment operator in Python. In this case, we are assigning the value of math.sqrt(144) to the variable x. Nothing is printed out because of that assignment.

Exercise 13

In the Console, run x or print(x).

CP/CR.

import math
x = math.sqrt(144)
x

Now that x has been defined in the Console, it is available for your use. Above, we just print it out. But we could also use it in other calculations, e.g., x + 5.

Although Polars includes hundreds of methods for data manipulation, the most important are filter(), select(), sort(), with_columns(), and group_by() with agg(). These work very similarly to their R tidyverse equivalents!

Exercise 1

Let’s warm up by examining the tips dataset. Type the following and hit “Run Code”:

import polars as pl
tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")
tips

import polars as pl
tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")
...

The tips dataset contains information about restaurant tips, including total bill, tip amount, and other variables.

Exercise 2

Run tips.describe() to see summary statistics.

import polars as pl
tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")
tips.describe()

Note that this gives us statistics for the numerical columns in the dataset.

Exercise 3

Use .drop_nulls() to remove rows with missing values. In Polars, we pipe operations by calling methods one after another, just like in R! Try:

tips.drop_nulls()

import polars as pl
tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")
tips.drop_nulls()

Note the number of rows in the DataFrame after drop_nulls(). This dataset actually has no missing values, so all rows remain.

We can chain methods by writing tips.drop_nulls().head() to first drop NA values and then show the first few rows.

Exercise 4

Chain .filter() to filter rows. Use pl.col("time") == "Dinner" as the argument. This is very similar to R’s filter()!

import polars as pl
tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")
tips.filter(pl.col("time") == "Dinner")

This workflow — in which we chain DataFrame methods together — is very common in Polars and very similar to R’s pipe workflow.

The resulting DataFrame has the same number of columns as tips because filter() only affects the rows. But there are fewer rows now.

Exercise 5

Continue the chain with .select() to choose specific columns. Use ["total_bill", "tip", "sex", "day"] as the argument. You can use the “Copy Code” button to avoid retyping.

import polars as pl
tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")
tips.filter(pl.col("time") == "Dinner").select(["total_bill", "tip", "sex", "day"])

Because select() doesn’t affect rows, we have the same number as after filter(). But we only have 4 columns now. This is just like R’s select() function!

Exercise 6

Copy previous code. Continue the chain with .describe().

tips.filter(pl.col("time") == "Dinner").select(["total_bill", "tip", "sex", "day"]).describe()

This gives us summary statistics for our filtered and selected data.

Exercise 7

Copy previous code. Replace .describe() with .drop_nulls().

tips.filter(pl.col("time") == "Dinner").select(["total_bill", "tip", "sex", "day"]).drop_nulls()

The number of rows stays the same because there are no missing values in this subset.

Exercise 8

Continue the chain with .sort("tip") to sort by the tip column.

tips.filter(pl.col("time") == "Dinner").select(["total_bill", "tip", "sex", "day"]).drop_nulls().sort("tip")

The sort() method sorts the rows of a DataFrame. By default, it sorts in ascending order. This is just like R’s arrange() function!

Exercise 9

Copy the previous code. Add descending=True as an argument to sort() to sort in descending order. This is like using desc() in R!

.sort("tip", descending=True)

Got to respect someone who tips $10!

Exercise 10

Let’s make a plot! For plotting, we’ll use plotnine, which works exactly like ggplot2 in R. Copy the previous code, but remove the .sort() line. Instead, save the result to a variable called tips_dinner, then convert it to pandas with .to_pandas() (plotnine works with pandas DataFrames), and create a plot using ggplot().

Here’s the pattern:

import polars as pl
from plotnine import ggplot, aes, geom_point

tips = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv")

tips_dinner = (tips
  .filter(pl.col("time") == "Dinner")
  .select(["total_bill", "tip", "sex", "day"])
  .drop_nulls()
  .to_pandas())

ggplot(tips_dinner, aes(x='total_bill', y='tip')) + geom_point()

from plotnine import ggplot, aes, geom_point

tips_dinner = (tips
  .filter(pl.col("time") == "Dinner")
  .select([...])
  .drop_nulls()
  .to_pandas())

ggplot(tips_dinner, aes(x='...', y='...')) + geom_point()

This creates a scatter plot showing the relationship between total bill and tip amount. Notice how similar this is to ggplot2 in R! We use ggplot(), aes(), and geom_point() just like in R.

Exercise 11

Copy previous code. Now let’s add jitter to see overlapping points better, just like in R! Change geom_point() to geom_jitter().

from plotnine import ggplot, aes, geom_jitter

ggplot(tips_dinner, aes(x='total_bill', y='tip')) + geom_jitter()

This is exactly like using geom_jitter() in R’s ggplot2! The jitter adds a small amount of random noise to help us see overlapping points.

Exercise 12

Finally, add a title and labels using plt.title(), plt.xlabel(), and plt.ylabel(). Consider this example:

import matplotlib.pyplot as plt
import seaborn as sns
tips = sns.load_dataset('tips')

tips_dinner = (tips
               .query('time == "Dinner"')
               [['total_bill', 'tip', 'sex', 'day']]
               .dropna())

sns.scatterplot(data=tips_dinner, x='total_bill', y='tip', alpha=0.6)
plt.title('Dinner Tips vs Total Bill')
plt.xlabel('Total Bill ($)')
plt.ylabel('Tip ($)')
plt.show()

You can make yours look like this, or create your own title and labels.

plt.title('...')
plt.xlabel('...')
plt.ylabel('...')
plt.show()

Note that the code in the exercise block is not saved. If you want to save the code, you can copy/paste it into a Python script file (.py).

Generative AI — tools like ChatGPT, Grok, Claude, DeepSeek and so on — are the future, of data science and everything else. The more you use these tools, the better off you will be. Unfortunately, the tools are changing so much that it is hard for a tutorial like this to stay up-to-date. This section provides some general advice and practice exercises.

Exercise 1

Using any AI you like, ask it to write a one-sentence summary about the Python programming language. Copy the answer below.

Example answer:

Python is a high-level, interpreted programming language known for its simplicity, readability, and versatility, widely used for web development, data analysis, artificial intelligence, and scientific computing.

If you do not want to pay for an AI service, then you will probably need to have free accounts with several different services. That way, if one service cuts you off for the day, you can switch to another.

Exercise 2

Type this in the Python Console and hit Enter:

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.head()

Copy/paste the command and the first few lines of output.

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.head()

When working with AI, you often need to tell it about the dataset. The easiest way to do that is often to just copy/paste the first few rows. That shows the AI what the column names and types are, which is key information for creating plots and data pipelines.

Exercise 3

Copy/paste the top of the iris DataFrame into your AI interface and ask it to create a chain of methods using Polars that calculates the average sepal_length for each species. Run the provided code in the Console. If it fails, show the AI the error and ask for better code.

CP/CR.

Claude gave us this answer:

import polars as pl
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv")
iris.group_by("species").agg(pl.col("sepal_length").mean())

This is a great answer! It uses group_by() just like R’s tidyverse, and then agg() (short for aggregate) with pl.col() to specify which column to calculate the mean for. Notice how similar this is to R:

R version: iris |> group_by(species) |> summarize(mean_sepal_length = mean(sepal_length))

Polars version: iris.group_by("species").agg(pl.col("sepal_length").mean())

Using AI is good! But intelligent use — use in which you understand what the AI has done and try to improve/clarify its answer — is even better.

Exercise 4

Ask AI to create a beautiful plot using the iris dataset and the plotnine/seaborn libraries. Run the provided code in the Console. If it fails, show the AI the error and ask for better code.

Example code from DeepSeek:

from plotnine import (
    ggplot, aes, geom_point, geom_smooth,
    facet_wrap, labs, theme_minimal
)
import polars as pl

# Load iris with Polars, then convert to pandas for plotnine
iris = pl.read_csv("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/iris.csv").to_pandas()

(
    ggplot(iris, aes(x="sepal_length", y="sepal_width", color="species")) +
    geom_point(size=3, alpha=0.7) +
    geom_smooth(method="loess", se=False, span=0.9) +
    facet_wrap("~ species") +
    labs(
        title="Sepal Shape Variation Across Iris Species",
        x="Sepal Length (cm)",
        y="Sepal Width (cm)",
        color="Species"
    ) +
    theme_minimal()
)

Note:

• AI tools are great at generating visualizations, but you should understand the code they provide
• Different AI tools may suggest different approaches - compare and learn from them
• Always test the code and make sure you understand what each line does
• You can ask the AI to explain any part of the code you don’t understand

The key is practice. Use AI every day!

This tutorial introduced you to the Python language for data science. You learned how to work with datasets using Polars, a fast DataFrame library with syntax very similar to R’s tidyverse. You learned how to chain operations using method chaining (just like R’s pipe operator |>), and how to make plots using matplotlib and seaborn.

The key advantage of Polars is that it feels familiar to R users while providing the speed and ecosystem of Python. Functions like filter(), select(), sort(), and group_by() work almost identically to their R counterparts!