🐍 Python, the parts you actually need

You will not memorize a library. You will read code that a model wrote and decide whether it does what it claims. That requires knowing what the pieces mean.

This is short on purpose. It is the vocabulary, not a tutorial.

Imports¶

import ibis
from plotnine import ggplot, aes, geom_line

import ibis makes the name ibis available. from ... import ... pulls specific names directly into your namespace.

When a model’s code fails with NameError, an import is usually missing. When two pieces of code disagree about what read_csv means, two different libraries both define it.

Assignment¶

co2 = ibis.read_csv("co2.csv")

The name co2 now refers to that object. Assignment is not a copy and not a computation — it is a label.

A common way analyses go wrong: reassigning the same name to different things as a script goes on, so that by line 60 nobody knows what df contains. Give things names that say what they are.

The dot¶

The . means “look inside this thing.”

co2.filter(...)     # call the filter method belonging to co2
co2.year            # access the column named year

Everything to the left of the dot determines what is legal to the right. ibis tables have .filter(); a plain number does not. When you see AttributeError, the object was not the type the code assumed — which is usually the actual bug, not a typo.

Method chaining¶

result = (
    co2
    .filter(co2.year >= 1990)
    .select("year", "average")
    .order_by("year")
)

Each method returns a new table, so the next method can act on it. Read a chain top to bottom: start with co2, keep rows from 1990 on, keep two columns, sort.

The outer parentheses let you break the chain across lines. Put each step on its own line. A chain you can read line by line is a chain you can check line by line.

Lazy evaluation¶

ibis does not do the work when you write the chain. It builds a description of the work and runs it when you ask for results.

result          # in interactive mode, shows a preview
result.to_pandas()   # actually executes, returns the whole thing in memory

This is why ibis can query a file far larger than your session’s memory: it pushes the filtering and grouping down to DuckDB, which streams the file and returns only the answer. to_pandas() is the moment you pull data into memory, so it is the moment to ask how much data you are about to pull.

The verbs¶

Nearly everything in this course is a combination of six or seven table operations — filter, select, mutate, group_by, agg, order_by. That vocabulary is small, consistent, and worth learning once; the next page, single table verbs, works through it properly.

Easy to remember, easy to get wrong¶

ibis.read_csv(...).filter(...).select(...) is not hard to write. That is the point of a well-designed package — the syntax is not the obstacle.

What is hard is everything the syntax does not tell you:

.filter(_.stressor == "CO2 - combustion - air") runs perfectly and quietly excludes cement, lime, and peat emissions.
.agg(total=_.value.sum()) runs perfectly and quietly adds kilograms to kilotonnes.
Reading one file when the dataset is twelve files produces a clean chart of a twelfth of the record.

None of those raise an error. Correct syntax is the beginning of the work, not the end of it.