LearnMachine Learning Fundamentals

Machine Learning
Fundamentals.

The supervised-learning loop the way working data scientists run it — regression, classification, regularization, cross-validation, pipelines. Every model trains live in your browser on the Lendly P2P-lending dataset, built on scikit-learn 1.3.

Course Overview

Modules

Duration

~8 hours

Live in-browser sklearnLendly running datasetInteractive animationsPipelines that don't leak

WHAT YOU SHOULD KNOW FIRST

Prereqs (links go straight to free courses)

This course assumes you can write Python, wrangle a DataFrame, and reason about probability. If any of these feels rusty, fill it in first — every minute spent here pays back in Modules 3–8.

Python Fundamentals

Syntax, functions, comprehensions, OOP. ~6h.

Pandas Fundamentals

DataFrames, groupby, merge — the data wrangling layer. ~5h.

Probability for Data Scientists

MLE, distributions, Bayes — the math under classification. ~7h.

Statistics Foundations

Mean, spread, distributions. The metric vocabulary. ~4h.

Learning Modules

Each module combines animated explanations, hands-on sklearn practice, and a 10–15 question knowledge check. Every ML concept is anchored to the same fictional lending platform — Lendly — so you build one mental model, not eight.

Module 1: The ML Frame & the Lendly Dataset

Beginner

What ML actually is, what it isn't, and the vocabulary that makes the next 7 modules click. Tour of the Lendly P2P-lending dataset you'll use across the whole course.

65 min

5 sections

Sections

1What Machine Learning Actually Is

2Features, Targets, and the Hypothesis Class

3Regression vs Classification — Same Frame, Different Output

4Train, Validate, Test — Why You Need Three Sets

5Meet Lendly — Your Running Dataset for All 8 Modules

5 sections

Interactive

Start

Module 2: Linear Regression: From Line-Fitting to Loss

Beginner

The model under every modern ML system. Mean squared error, the normal equation, gradient descent, and how to read coefficients on Lendly interest rates.

75 min

6 sections

Sections

1The Linear Model — A Weighted Sum of Features

2Mean Squared Error — Why Squared, Why Mean

3The Normal Equation — Closed-Form Solution

4Gradient Descent — When Closed Form Won't Scale

5Reading the Output — Coefficients, R², Residuals

6Lendly Case: Predicting Interest Rate from Borrower Features

6 sections

Interactive

Start

Module 3: Regularization & the Bias–Variance Tradeoff

Intermediate

Why one extra feature can crash your test score, and the two penalties (Ridge = L2, Lasso = L1) that fix it. Bias-variance, the regularization path, and why standardization is non-negotiable.

65 min

6 sections

Sections

1Overfitting in Lendly — The Symptom

2Bias and Variance — The Two Failure Modes

3Ridge Regression — Squeeze the Coefficients

4Lasso — Squeeze Some Coefficients to Zero

5Picking Alpha and Why Standardization Is Non-Negotiable

6Lendly Case: Regularized Rate Prediction

6 sections

Interactive

Start

Module 4: Classification: Logistic, kNN, Decision Tree

Intermediate

Three classifiers with completely different inductive biases — logistic regression, kNN, decision tree. Same dataset, three views, three sets of strengths.

70 min

6 sections

Sections

1The Classification Setup — Predicting Default on Lendly

2Logistic Regression — Sigmoid, Log-Odds, and Why

3k-Nearest Neighbors — Geometric, Lazy, Powerful

4Decision Trees — Recursive Yes/No Splits

5Three Models, One Dataset — How They Disagree and Why

6Lendly Case: Default Prediction Bake-Off

6 sections

Interactive

Start

Module 5: Classification Metrics & Threshold Choice

Intermediate

Confusion matrix, precision, recall, F1, ROC AUC, PR curves — and the most underrated lever in classification: the decision threshold. Pick it from cost, not from 0.5.

65 min

6 sections

Sections

1The Confusion Matrix — Where Every Other Metric Comes From

2Precision, Recall, F1 — Tradeoffs, Not Synonyms

3ROC and PR Curves — Whole-Model Reports

4Threshold Choice — The Most Underrated Lever in ML

5Class Imbalance — When Accuracy Lies

6Lendly Case: Picking a Default-Risk Threshold

6 sections

Interactive

Start

Module 6: Cross-Validation, Leakage & Stratification

Intermediate

K-fold, stratified k-fold, time-series CV. The six common forms of data leakage and the Pipeline pattern that prevents all of them.

65 min

6 sections

Sections

1The Problem with One Train/Test Split

2K-Fold Cross-Validation — One Idea, Many Variants

3Stratified K-Fold — Keep Class Balance Across Folds

4Data Leakage — The Silent Model Killer

5Time-Series CV — Why Random Folds Are Wrong for Temporal Data

6Lendly Case: 5-Fold CV the Right Way

6 sections

Interactive

Start

Module 7: Pipelines, Preprocessing & GridSearchCV

Intermediate

The production-style pattern: ColumnTransformer for mixed types, SimpleImputer for missing, OneHotEncoder for categoricals — all in a Pipeline. Tune it with GridSearchCV.

65 min

6 sections

Sections

1The Anti-Leakage Pattern — Why Pipelines Exist

2ColumnTransformer — One Transformer per Column Type

3Imputation, Scaling, Encoding — The Trio

4GridSearchCV — Searching the Hyperparameter Space

5RandomizedSearchCV — Smarter When Grids Explode

6Lendly Case: A Production-Style Pipeline

6 sections

Interactive

Start

Module 8: Capstone: End-to-End on Lendly

Intermediate

The full loop top to bottom — EDA, splitting, pipeline, CV, threshold from cost, final test report, model card. The artifact every senior data scientist ships.

65 min

6 sections