Led by Gaussian Regression Simulacrum

About this course
Gauss invented the method of least squares in 1795 to find the orbit of a lost asteroid from sparse, noisy observations. Every regression algorithm taught in this course is a variation on the same problem: given noisy data, find the best-fitting function.
This course covers five regression algorithms in Python with scikit-learn — simple and multiple linear regression, polynomial regression, SVR, decision tree regression, and random forest regression — plus the model evaluation framework that tells you which one to trust.
Based on the published writings of Carl Friedrich Gauss.
If you found this course useful, consider becoming a patron and supporter. Support Universitas Scholarium →
Led by Gaussian Regression Simulacrum
The question
Gauss minimised the sum of squared residuals because it gives a unique, analytically tractable solution. What does R-squared actually measure — and when should you use adjusted R-squared instead?
Outcome
The student can implement simple and multiple linear regression, interpret R-squared, and apply backward elimination.
Sub-units
Led by Gaussian Regression Simulacrum
The question
When the data bends, linear regression fails. Polynomial regression adds power terms; SVR finds the widest tube within which all points lie. Which is more robust to outliers — and why must SVR features always be scaled?
Outcome
The student can implement polynomial regression and SVR, explaining when each is appropriate.
Sub-units
Led by Gaussian Regression Simulacrum
The question
A decision tree partitions space into rectangles and predicts the regional mean. It overfits. A Random Forest trains hundreds of trees on bootstrap samples and averages them. Why does averaging reduce variance — and how many trees is enough?
Outcome
The student can implement both, explain ensemble averaging, and interpret feature importance.
Sub-units
Led by Gaussian Regression Simulacrum
The question
R-squared always increases when you add a feature, even if it is noise. How does adjusted R-squared fix this — and what does a large gap between training R-squared and test R-squared tell you?
Outcome
The student can compare five regression models systematically on test-set R-squared.
Sub-units
Led by Gaussian Regression Simulacrum
The question
No regression model is best in general. The right choice depends on dataset size, linearity, interpretability requirements, and outlier robustness. Given a new dataset, how do you choose?
Outcome
The student can select and justify a regression model for a specific dataset and business context.
Sub-units