LESSON 44 · INTERMEDIATE4 min read

Project: A Reproducible Data Analysis Notebook

Answer one revenue question with NumPy, make every cleaning decision visible, and ship an executed notebook that another person can rerun.

Python 3.12+Free to readModule 9 of 19

YOUR NEXT SMALL WINS

By the end, you’ll be able to…

Turn a business question into an explicit dataset schema.
Clean missing and invalid values with documented NumPy masks.
Group totals with unique labels and bincount weights.
Present results and limitations in an executed, restart-safe notebook.

01Start from a question and a schema

A notebook is not a pile of cells; it is an argument. Begin with one question a reader can evaluate, such as which region produced the most net revenue after returns. Write down the unit of a row, the currency, the time range, and the meaning of every column before loading any data.

For this project each row is one order line with a region, a unit count, a unit price, and a returned flag. Revenue is units multiplied by price, but a returned line contributes zero net revenue while still costing the gross amount. Naming that rule now prevents a chart later that quietly disagrees with the finance definition.

schema_check.pyPython

import numpy as npregions = np.array(["North", "South", "North", "East", "South", "North"])units = np.array([3, 5, 2, 4, 1, 6])prices = np.array([19.95, 12.50, 19.95, 8.00, 12.50, 19.95])returned = np.array([False, False, True, False, True, False])gross = units * pricesnet = np.where(returned, 0.0, gross)print(np.round(net, 2))

Outputstdout

[59.85 62.5  0.   32.   0.  119.7 ]

02Clean with visible rules

Real extracts contain missing values, negative quantities, and prices recorded as zero. Clean them with boolean masks before computing a total, and report how many rows each rule removed. A silently dropped row is a result nobody can audit.

np.isfinite rejects both NaN and infinity. Combine it with business conditions using & and wrap each comparison in parentheses. The mask is a copy of the decisions, not a hidden mutation of the source arrays. Keep the cleaned arrays and the count of rejected rows for the notebook's methods section.

clean_rows.pyPython

import numpy as npunits = np.array([3.0, 5.0, np.nan, 4.0, -1.0, 6.0])prices = np.array([19.95, 12.50, 19.95, 8.00, 12.50, np.nan])valid = np.isfinite(units) & np.isfinite(prices) & (units >= 0) & (prices > 0)print(valid)print(int((~valid).sum()))

Outputstdout

[ True  True False  True False False]
3

03Group with explicit labels and weights

np.unique with return_inverse gives each region a stable integer code, and np.bincount sums weights by that code. The pair is a compact group-by for small categorical data. Names and totals stay aligned because both come from the same unique call.

State the axis and the unit in the output. A table of totals is easier to trust when the currency and the row count are visible. Sort by the label or by the total deliberately; do not rely on the order in which regions happened to appear in the extract.

group_totals.pyPython

import numpy as npregions = np.array(["North", "South", "North", "East", "South", "North"])net = np.array([59.85, 62.50, 0.0, 32.0, 0.0, 119.70])names, codes = np.unique(regions, return_inverse=True)totals = np.bincount(codes, weights=net)for name, total in zip(names, totals):    print(f"{name:<6} {total:8.2f}")

Outputstdout

East      32.00
North    179.55
South     62.50

04Make the notebook reproducible

A reproducible notebook runs from top to bottom in a fresh kernel. Put imports and configuration in the first cell, load data with a relative path or a documented URL, and avoid hidden state from cells executed out of order. Restart and run all before you trust the output.

Use markdown cells to state the question, the schema, the cleaning rule, and the limitation. A chart needs labeled axes, a stated unit, and an honest baseline. Export an executed HTML copy with nbconvert so a reader can see the outputs without installing the environment, and keep the notebook and any helper module under version control.

notebook_workflow.shShell

py -m pip install numpy jupyterpy -m jupyter labpy -m jupyter nbconvert --to html --execute analysis.ipynb

LET IT CLICK

The ideas worth keeping.

A notebook should answer one question with a documented schema and unit.
Cleaning rules belong in visible masks with counts, not silent drops.
Grouped totals need stable labels, an explicit axis, and a restart-safe execution.

TURN UNDERSTANDING INTO SOMETHING YOU CAN DO

Make it yours.

Type and test your code right here in the interactive workspace. Run your solution, check the output, and compare with the worked guide when ready.

Compute average revenue per region

Using the grouped example, divide each region total by its row count. Print the region, count, and average with two decimals.

your_turn_1.py🐍Python

Ctrl + ↵

Output

Click Run or press Ctrl+Enter

Output will appear here after you run the code.

Explore a worked solution

one_possible_solution.pyPython

import numpy as npregions = np.array(["North", "South", "North", "East", "South", "North"])net = np.array([59.85, 62.50, 0.0, 32.0, 0.0, 119.70])names, codes = np.unique(regions, return_inverse=True)totals = np.bincount(codes, weights=net)counts = np.bincount(codes)for name, total, count in zip(names, totals, counts):    print(f"{name:<6} {count} {total / count:8.2f}")

bincount without weights counts rows per code. Dividing the weighted total by that count gives the average per row, not per region-day; label the unit accordingly.

Report the cost of returns

Given the original arrays, compute gross revenue for returned rows and print the count and the lost amount rounded to two decimals.

your_turn_2.py🐍Python

Ctrl + ↵

Output

Click Run or press Ctrl+Enter

Output will appear here after you run the code.

Explore a worked solution

one_possible_solution.pyPython

import numpy as npunits = np.array([3, 5, 2, 4, 1, 6])prices = np.array([19.95, 12.50, 19.95, 8.00, 12.50, 19.95])returned = np.array([False, False, True, False, True, False])lost = units[returned] * prices[returned]print(int(returned.sum()), round(float(lost.sum()), 2))

Boolean indexing selects the returned rows before multiplication. The count and lost amount answer a different question from net revenue, so keep both metrics visible.

A SMALL QUESTION. A USEFUL NEXT STEP.

What stayed with you?

Check the ideas while they’re fresh. Your answers stay in this page and feedback appears as you go.

A LITTLE ACTIVE RECALL0 / 3 answered

01What does shape (6,) describe?

ASix rows and one columnBOne dimension with six elementsCSix separate arrays

02Which operation returns a copy?

AA basic sliceBBoolean maskingCReading one element

03What makes a notebook reproducible?

ARunning only the changed cellsBRestarting the kernel and running all cells in orderCSaving the notebook without outputs

⚡CODING CHALLENGE

🔒Required to Mark Lesson Complete

Compute average revenue per region

Using the grouped example, divide each region total by its row count. Print the region, count, and average with two decimals.

EXPECTED OUTPUT

name:<6 count total / count:8.2f

main.pyPython 3

Ctrl + ↵

Execution Console

Click "Run & Verify" (or press Ctrl + Enter) to test your code against the challenge.

Keep exploring.

The official references behind this lesson, when you want a second angle or the full details.

MODULE 09 · PRACTICAL SELF-CHECK

Pause. Put the pieces together.

You’ve reached the end of NumPy and Jupyter Notebooks. Before moving on, use this task to check what you can do independently.

Create a Jupyter notebook that imports NumPy, performs statistical analysis on a dataset, and documents the findings with markdown cells.

Try it in your own environment and write down the result. This is a self-check; reading completion does not evaluate your code.

Review this module

One lesson closer.

Reading is a start. Trying it yourself makes the connection.

Solve the hands-on coding challenge in the knowledge section to verify your code and unlock lesson completion.