First Steps

Goals

• Launch the notebook environment and run a first code cell.
• Prompt an LLM to read a data file and print its contents.
• Prompt an LLM to compute and interpret the mean and median of a column.

Starting the Notebook

How do I start a notebook and create a first cell?

• A notebook is a document that mixes text, code, and output in one file
  • Jupyter notebooks have the extension .ipynb; each piece of code or text is called a "cell"
  • Each cell can hold either code or explanatory text; you run code cells one at a time and the output appears directly below
• Your instructor will give you a URL to open the notebook environment in your browser
  • If you are running Jupyter on your own laptop, open the Terminal app (Mac) or Command Prompt (Windows), type jupyter notebook, and press Enter; a browser tab will open automatically
  • Either way, you will see a file browser; click "New" and choose "Python 3" to create a fresh notebook
• To add a code cell, click the + button in the toolbar or press B when a cell is selected
• To run a cell, press Shift+Enter
  • The output appears directly below the cell
  • If the cell causes an error, the error message appears there too

What should I type in the first cell to make sure everything is working?

• Paste print("hello") into the first cell and press Shift+Enter
  • If you see hello appear below the cell, the environment is working
  • If you see an error like ModuleNotFoundError, a required package is not installed

Reading the Data

Read a CSV file of monthly climate observations and print its first five rows.

• The dataset for this session is the Environment and Climate Change Canada Historical Climate Data [eccc-climate2025]

  • Go to ECCC Climate Data and search for a station near you (e.g., Toronto Pearson International Airport or Vancouver International Airport)
  • Select "Monthly" data, choose all available years, and download as CSV
  • The file is named something like en_climate_monthly_ON_6158731_1840-2025_P1M.csv
  • It has one row per month with columns for mean temperature, precipitation, and other observations

• Polars is a Python library for working with tables of data

  • It is already installed in the course environment; you do not need to do anything to set it up
  • Including "Using Polars" in your prompt tells the LLM which tool to use; without it, the LLM might choose a different library and produce code that does not run in your environment

Use Polars to read a CSV file called climate.csv, skip the first row which is a title line, treat empty cells as missing, and print the first five rows.

• The LLM will produce something like:

import polars as pl

df = pl.read_csv("climate.csv", skip_rows=1, null_values=[""])
print(df.head())

• Paste the code into a new notebook cell and run it
  • The output shows five rows with columns including Year, Month, and Mean Temp (°C)
  • If the code crashes with FileNotFoundError, the CSV is not in the same folder as the notebook

Mean and Median

• The mean is the sum of all values divided by the count (also called the average)
• The median is the middle value when all observations are sorted from smallest to largest

Compute the mean and median of the monthly mean temperature column.

• If that doesn't work, try a more specific prompt:

Using Polars, read climate.csv the same way, then compute and print the mean and median of the 'Mean Temp (°C)' column, excluding missing values.

• The LLM will produce something like:

import polars as pl

df = pl.read_csv("climate.csv", skip_rows=1, null_values=[""])
temps = df["Mean Temp (°C)"].drop_nulls()
print(f"Mean:   {temps.mean():.2f} °C")
print(f"Median: {temps.median():.2f} °C")

• Run the cell
  • For most Canadian cities the annual mean will be between -10° and +15° C
  • If the number is in the hundreds, the LLM likely read the wrong column

Why would the mean and median of a dataset ever be different numbers?

• The mean is the sum of all values divided by the count
  • It is pulled toward extreme values: a few very cold winters raise the magnitude of negative values and pull the mean down
• The median is the middle value when the data is sorted
  • Exactly half the values are above it and half are below
  • A few extreme months do not move it much
• For Canadian climate data, the mean and median are often close because temperatures follow a roughly symmetric seasonal cycle
  • They can differ dramatically in a dataset with outliers
• When outliers matter for your question, report the mean; when you want a typical value unaffected by extremes, report the median
  • Temperature records: usually mean, because the total heat budget drives most climate effects
  • Housing prices: usually median, because a handful of luxury properties distort the mean

Saving the Notebook

What does it mean to save a notebook, and why does it matter?

• Saving a Jupyter notebook writes all cells and all outputs to a .ipynb file on disk
  • In Jupyter, press Ctrl+S (or Cmd+S on Mac) or click the save icon
  • The saved file is a record of the analysis that anyone with the same data file can re-run
• Reproducibility means that running the same code on the same data produces the same result
  • A notebook achieves this only if cells are run in order from top to bottom
  • Running cells out of order can produce outputs that no longer match the code
  • Use "Kernel → Restart and Run All" before sharing a notebook to confirm everything runs cleanly
• The combination of the data file and the saved notebook is the smallest unit of reproducible analysis

Saving Prompts

I am tired of typing "use Polars" and "treat empty cells as missing" in every prompt. Is there a way to set these once?

• Most LLM tools let you set custom instructions that are added to every conversation automatically
  • In Claude, open Settings and look for "Custom instructions" or create a Project and add instructions there
  • In ChatGPT, open Settings → Personalization → Custom instructions
  • Whatever you write there is sent to the LLM along with every prompt you type, without you having to repeat it
• Useful things to put in custom instructions for this course:
  • "Always use Polars for data manipulation, not pandas."
  • "Always use Altair for charts."
  • "Treat blank cells and the values U, Q, and N as missing unless I say otherwise."
  • "Print the first five rows after reading any CSV file."
• Custom instructions do not replace careful prompting
  • You still need to give the LLM the column names and goal for each specific task
  • They only save you from repeating the same background preferences in every session

Check Understanding

Exercises

Different Seasons

Compute the mean and median for January temperatures and July temperatures separately. Are they closer together within a season than across the full year? Why might that be?

Plot Over Time

Ask the LLM to make a line chart of mean annual temperature over time (one point per year). Save it as a PNG. Describe in one sentence what the chart shows about temperature change at your station.

Count Missing Values

Ask the LLM to count how many missing values appear in each column. Which columns tend to have more missing data, and why might that be for a weather station dataset?

Compare Two Stations

Download data for a second station in a different region (e.g., one coastal and one inland). Ask the LLM to compute the mean annual temperature for each station and display them side by side. Which station is warmer on average?

Skip the Right Rows

The following code reads the climate CSV and tries to compute mean and median temperature, but the first line of output looks like a column header rather than a real measurement. Work with an LLM to find and fix the problem.

import polars as pl

df = pl.read_csv("climate.csv", skip_rows=0, null_values=[""])
print(df.head())
temps = df["Mean Temp (°C)"].drop_nulls()
print(f"Mean:   {temps.mean():.2f} °C")
print(f"Median: {temps.median():.2f} °C")

Right Number, Wrong Column

The following code reads the climate CSV and prints a mean and median, but the values are implausibly large for temperatures in Canada. Work with an LLM to identify which column is actually being used and correct it.

import polars as pl

df = pl.read_csv("climate.csv", skip_rows=1, null_values=[""])
temps = df["Total Precip (mm)"].drop_nulls()
print(f"Mean:   {temps.mean():.2f} °C")
print(f"Median: {temps.median():.2f} °C")

Missing Values Not Declared

The following code reads the climate CSV and computes mean and median, but both values print as None. Work with an LLM to find out why and fix the code.

import polars as pl

df = pl.read_csv("climate.csv", skip_rows=1)
print(f"Temperature column type: {df['Mean Temp (°C)'].dtype}")
temps = df["Mean Temp (°C)"].drop_nulls()
print(f"Non-null rows: {len(temps)}")
print(f"Mean:   {temps.mean()}")
print(f"Median: {temps.median()}")

More Than Mean and Median

The following code computes the mean and median monthly temperature. Work with an LLM to extend it so it also prints the minimum value, the maximum value, and the count of non-null observations.

import polars as pl

df = pl.read_csv("climate.csv", skip_rows=1, null_values=[""])
col = "Mean Temp (°C)"
temps = df[col].drop_nulls()
print(f"Mean:   {temps.mean():.2f} °C")
print(f"Median: {temps.median():.2f} °C")
# TODO: also print the minimum value, maximum value, and count of non-null observations