Python 3.11+pandas 3.xIntermediate~4h

pandas

Predict the table, then run it

Filtering, grouping, joining and reshaping all change the size of a table, and being good at this is being able to say what size before you look. Seven runs of drills and builds, ending with the question none of the tutorials ask: when you assign into a frame, does the write actually land?

Start this project See the modules

Modules: 7
Drills: 10
Build steps: 7
Time: ~4h

Walk out able to

Predict a merge’s row count from its keys, move a table between long and wide, and say whether a write reaches the frame — before running any of it.

What you work through

Frames, dtypes, selection

What a column really holds, and reaching one exact row.

Grouping and joining

One row per group — and the merge that doubled your rows.

Long, wide and back

Melt, pivot and stack, as arithmetic rather than magic.

Where a write lands

The warning, what replaced it, and the report you build.

The pitch

What you practise, and what you leave with

A report script: two tables in, unmatched rows discarded, totals by town and year, one wide CSV out, checked against a total.

You will practise

Selecting by label, position and mask
Grouping, aggregating, and missing keys
Merge row counts and column names
Long and wide, and where a write lands

Afterwards you can

Call a merge’s row count before running it
Move a table between long and wide
Spot chained assignment on sight
Read pandas 2.x code on pandas 3.x

Modules

7 modules, 25 items

Lessons explain one idea. Drills repeat it until it sticks. Build steps make something that exists afterwards.

1
Get a frame and see what is in it
~22 min
Build a frame, read its dtypes, and say how many values are really there.
- LessonWhy a frame and not a list of dicts4 min
- Build stepBuild your first frame and read it7 min
- DrillSay what dtype a column will get11 min
2
Select the rows you meant
~42 min
Reach any subset by label, by position or by mask — and know the count first.
- LessonTell loc and iloc apart for good11 min
- Build stepFilter sixty rows with a mask8 min
- DrillCount the rows a filter keeps12 min
- DrillTell a label from a position11 min
3
One row per group
~40 min
Split by a key and predict both the value and the shape of what comes back.
- LessonCollapse a frame to one row per town10 min
- Build stepTotal fifty readings by town6 min
- DrillCount the rows a grouping returns12 min
- DrillAggregate a column with gaps in it12 min
4
Line two frames up on a key
~41 min
Predict a merge’s row count from the keys, and catch the multiplication early.
- LessonDecide which rows a merge keeps11 min
- Build stepPredict an outer join, then run it6 min
- DrillCount the rows a merge returns13 min
- DrillName the columns a merge produces11 min
5
Long and wide are one table
~28 min
Move a table between the two encodings and say what shape it lands in.
- LessonTurn a long table wide, and back10 min
- Build stepPivot forty-five readings into a grid6 min
- DrillPredict the shape after a reshape12 min
6
The copy you thought was a view
~33 min
Say whether a write reaches the frame — on your pandas, and on the one before it.
- LessonSpot chained assignment on sight10 min
- LessonRead pandas 2.x code on pandas 3.x10 min
- DrillSay where a write lands, on sight13 min
7
Build the report
~42 min
Recall the calls from memory, then turn two tables into one wide report.
- DrillWrite the call that does what you meant12 min
- Build stepThrow away the visits with no town10 min
- Build stepTotal the minutes by town and year8 min
- LessonTurn the totals into a wide report12 min