Data Preprocessing & Feature Engineering Cheat Sheet: Imputation, Scaling, Encoding, ML Pipelines, Study Guide PDF

A data preprocessing cheat sheet that shows what each step learns from training data: imputation, scaling, encoding and pipelines, drawn with numbers.
THIS IS A DIGITAL DOWNLOAD. NO PHYSICAL ITEM WILL BE SHIPPED.
You will receive printable PDF files in one ZIP file. No printed or laminated sheets are included. The photos show how they look once printed.

SEE WHAT EACH PREPROCESSING STEP LEARNS
A visual study guide to data preprocessing and feature engineering for machine learning, on two sheets printed on both sides. Each step is shown as a small card of numbers it learns from the training rows and then applies to any rows, so the difference between fit and transform becomes something you can see. One small invented table of library loans (8 training rows, 3 test rows) runs through all four pages, and every value on the pages comes from really running the printed code (scikit-learn, pandas and NumPy).

SHEET 1, FRONT - FIT LEARNS, TRANSFORM APPLIES
- the learned state, drawn: a median imputer learns 3.55 from the training rows and fills the test gap with 3.55, not with the test rows' own median
- fit, transform and fit_transform: what each call returns
- the missing indicator: a 0/1 "was missing" column, and when it is added
- which imputation? median, mean, most frequent or constant: what each learns, what it is good for, what to watch out for

SHEET 1, BACK - NUMBERS: SCALE OR TRANSFORM?
- the three formulas: StandardScaler, MinMaxScaler and RobustScaler, with the population SD and the quartile rule
- one column, three scalers: the same eight values on four aligned number lines; the 86-day loan becomes 2.59, 1 and 8.48
- what changes if the extreme value had been 23 instead of 86 (mean and SD move, median and IQR stay)
- log1p for right-skewed values, two histograms of the same data
- scaling guidance by model (k-NN, linear models, neural nets, trees), and why pandas and the scaler give different SDs

SHEET 2, FRONT - CATEGORIES AND DATES
- one-hot encoding drawn as a matrix: a category never seen in training becomes a row of zeros
- the error you get without handling unknown categories, and grouping rare categories
- ordinal codes for a truly ordered column, and the equal-spacing assumption
- weekday on a circle (sine and cosine), so Sunday sits next to Monday
- target encoding explained briefly: smoothing and cross-fitting

SHEET 2, BACK - PIPELINES
- the whole sheet in one pipeline: each column routed to its own branch (impute and scale, one-hot, ordinal, weekday circle), joined, then one model
- the 8 output feature names, as the pipeline really returns them
- each cross-validation fold learns its own medians: four fold cards with their held-out rows
- predict vs transform, and where a feature-selection step belongs

Every page ends with a short "Your turn" exercise on fresh numbers, with the answer, and 2 or 3 common traps.

TEST YOURSELF VERSION
The same four pages with the four Your-turn answers left blank and lines to write on: what an imputer learns, the three scalers on new values, a one-hot table with an unknown category, and tracing a fitted pipeline step by step. Fill them in from memory, then check yourself against the cheat sheet.

WHAT YOU GET: 8 PDF FILES (in one ZIP)
- Cheat Sheet and Test Yourself
- 2 colour styles: Color, and Gray (uses less coloured ink)
- 2 paper sizes: A4 and US Letter
- each file is two sheets: 4 pages, print double-sided ("flip on long edge")
- a short READ ME with printing tips
The code on the pages is printed for reading and learning. No code files, notebooks or data files are included.

WHO IT IS FOR
Students in a first machine learning or data science course, bootcamp learners, and junior analysts moving into modelling. Assumes pandas basics (a DataFrame, columns, NaN), the idea of train and test sets, and the mean.

HOW IT IS MADE
Typeset in LaTeX; every table, number line and diagram is drawn by code from the real outputs, no AI image generator was used. The content and the code were prepared with help from an AI assistant. Every code snippet on the sheet was run by a script (tested with scikit-learn 1.9.1, pandas 3.1 and NumPy 2.5), its printed output compared, and the key numbers recomputed independently: 125 automatic checks. The content went through several independent reviews.

HOW IT WORKS
1. Purchase the listing.
2. Download the ZIP file instantly from Purchases & Reviews in your Etsy account (or from the link in your confirmation email) and unzip it.
3. Print at home or at a print shop at 100% / "actual size", double-sided, or open the PDF on a tablet.

PLEASE NOTE
- This is a digital product. Nothing will be shipped.
- This is a focused study guide to preprocessing, not a complete library reference. Library behaviour can change in later versions; the pages say which versions were tested.
- Due to the nature of digital downloads, refunds are not accepted. Send us a message if you have any problem with your files and we'll gladly help.
- Colours may vary slightly between screens and printers.
- For personal use only. You may print as many copies as you need for yourself. Teachers: please message us about classroom use. Reselling or sharing the files is not allowed. (c) MIRAICO Academy, all rights reserved.
- Independent educational resource. Not affiliated with or endorsed by the projects mentioned.

Also in the shop: the Training, Validation & Leakage, Supervised Machine Learning, Model Evaluation & Metrics and pandas cheat sheets, in the same visual style.

At MiraicoAcademy, we draw what the code learns, so the numbers make sense.