Loading learning path...
Develop practical ML judgment from numerical foundations and reproducible data preparation through baseline models, evaluation and safe data releases. Combine implementation problems, visualizers and analytical repository labs.
The curriculum and complete lessons are free with an account. Optional practice resources and Ono usage follow your plan’s allowances.
A practical route centered on data, model behavior and evaluation rather than a claim to cover every ML architecture or production serving stack. Dedicated ML subject and pattern paths provide additional theory and practice.
How to study: Work in order. Predict the result before opening a visualizer, implement before reading the editorial, and explain the invariant and edge cases after each exercise. Repository and review activities are paired: fix the behavior, then assess a proposed change independently. Choose a supported language or repository environment; you do not need to repeat every environment. Existing completion of shared content counts toward this path.
Check your understanding: • Explain the data split and what information may cross it. • Compare a baseline, error analysis and a metric appropriate to the decision. • Justify robustness, cohort visibility and dataset publication with explicit tests.
What Is Machine Learning?
The Machine Learning Pipeline
Key Concepts and Terminology
Machine Learning Success Factors
The Learning Problem
The Bias-Variance Tradeoff
Dataset Summary Metrics Calculator
Model Fit Diagnosis: Identifying Generalization Issues
Vectors and Vector Spaces: The Language of Machine Learning
Matrices and Linear Transformations
Norms and Distance Metrics
Projections and Least Squares
Vector Inner Product Computation
Linear Transformation of a Vector
Pairwise Matrix Product Computation
Feature Covariance Computation
Feature Engineering Mastery
Feature Transformation Pipelines
Feature Selection Theory
Data Normalization Techniques: Z-Score and Min-Max Scaling
Categorical Binary Encoding
Synchronized Data Permutation
Dataset Batch Generator for Training Pipelines
Ordered Polynomial Feature Generator
Holdout Validation
K-Fold Cross-Validation
Stratified and Group Cross-Validation
Time Series Cross-Validation
Nested Cross-Validation
Data Partitioning with Cross-Validation Splits
Random Dataset Subset Generation for Ensemble Learning
Simple Linear Regression
Multiple Linear Regression
Regression Diagnostics
Gradient Descent
The Overfitting Problem
Ridge Regression (L2 Regularization)
Lasso Regression (L1 Regularization)
Closed-Form Linear Model Fitting
Optimizing Linear Model Coefficients via Gradient Descent
L2-Regularized Linear Regression Loss Computation
Sparse Regression Optimizer with L1 Penalty
Binary Classification Problem
The Logistic Regression Model
Decision Tree Fundamentals
Splitting Criteria
Pruning and Regularization
Logistic Sigmoid Transformation
Single Neuron Binary Classifier with Sigmoid Activation
Computing Gini Impurity for Classification Labels
Recursive Classification Tree Builder
Classification Metrics
ROC and Precision-Recall Curves
Multi-class Classification Metrics
Regression Metrics
Custom and Business Metrics
Debugging ML Models
Classification Accuracy Metric
Precision Evaluation Metric for Binary Classification
Sensitivity Score for Binary Classifier
Weighted Harmonic Mean of Precision and Recall
Binary Classification Outcome Matrix
Prediction Error Magnitude Calculator
Coefficient of Determination Calculator
Clustering Problem Formulation
K-Means Clustering
Clustering Evaluation
Principal Component Analysis Theory
Anomaly Detection Fundamentals
Statistical Methods for Anomaly Detection
Centroid-Based Data Partitioning
Principal Component Extraction
Trailing windows include both time endpoints and every observation tied at the anchor time, separately for each group
hampel proposal 1
Use the lower weighted median: the smallest value whose cumulative weight reaches at least half the total
hampel proposal 2
Treat values at the Hampel threshold as inliers, including zero deviation with zero MAD, and classify the anchor by its absolute deviation
hampel proposal 4
Probability Fundamentals
Estimation Theory
Bootstrap Methods
Include subjects failing or censoring at a time in that time risk set
survival proposal 1
Respect delayed entry in every risk set
survival proposal 2
The Kaplan-Meier median is the first time survival is at most one half, and survival updates use the full risk-set fraction
survival proposal 4
Observe a period on its final report day
Observe a period on its final report day
Repair historical visibility and positive-net retention
Repair historical visibility and positive-net retention
Interpretability Fundamentals
Fairness in Machine Learning
Bias Detection and Mitigation
ML System Design
Valid measurements are rejected at the quality limits
Accept readings on a calibration limit
Captured measurements change after a calibration rollout
Correct the order of temperature calibration
One bad dataset row consumes several quality allowances
Count bad dataset rows once
Published datasets differ from the snapshot that passed review
Publish the rows captured by the quality run
Failed dataset publishing advances a channel without an audit
Fence channel publication with the current generation
Design a Machine Learning Feature Store
Complete this path to earn a personalised certificate. You choose whether to make it public.