Machine Learning for Aspiring Data Scientists: Zero to Hero

Course Curriculum

Chapter 1: Machine Learning Models

Lecture 1: Modeling an epidemic

Lecture 2: The machine learning recipe

Lecture 3: The components of a machine learning model

Lecture 4: Why model?

Lecture 5: On assumptions and can we get rid of them?

Lecture 6: The case of AlphaZero

Lecture 7: Overfitting/underfitting/bias/variance

Lecture 8: Why use machine learning

Lecture 9: Notes on machine learning models

Chapter 2: Linear regression

Lecture 1: The InsureMe challenge

Lecture 2: Supervised learning

Lecture 3: A quick note on the word "features"

Lecture 4: Linear assumption

Lecture 5: Linear regression template

Lecture 6: Non-linear vs proportional vs linear

Lecture 7: Linear regression template revisited

Lecture 8: Loss function

Lecture 9: Training algorithm

Lecture 10: Code time

Lecture 11: R squared

Lecture 12: Why use a linear model?

Lecture 13: Kaggle notebook on linear regression

Lecture 14: Notes on supervised learning and linear regression

Lecture 15: Finding closed-form solution to linear regression (optional)

Chapter 3: Scaling and Pipelines

Lecture 1: Introduction to scaling

Lecture 2: Min-max scaling

Lecture 3: Code time (min-max scaling)

Lecture 4: The problem with min-max scaling

Lecture 5: What's your IQ?

Lecture 6: Standard scaling

Lecture 7: Code time (standard scaling)

Lecture 8: Model before and after scaling

Lecture 9: Inference time

Lecture 10: Pipelines

Lecture 11: Code time (pipelines)

Lecture 12: Kaggle notebook on scaling and pipelines

Lecture 13: Notes on scaling and pipelines

Chapter 4: Regularization

Lecture 1: Spurious correlations

Lecture 2: L2 regularization

Lecture 3: Code time (L2 regularization)

Lecture 4: L2 results

Lecture 5: L1 regularization

Lecture 6: Code time (L1 regularization)

Lecture 7: L1 results

Lecture 8: Why does L1 encourage zeros?

Lecture 9: L1 vs L2: Which one is best?

Lecture 10: Kaggle notebook on regularization

Lecture 11: Notes on regularization

Chapter 5: Validation

Lecture 1: Introduction to validation

Lecture 2: Why not evaluate model on training data

Lecture 3: The validation set

Lecture 4: Code time (validation set)

Lecture 5: Error curves

Lecture 6: Model selection

Lecture 7: The problem with model selection

Lecture 8: Tainted validation set

Lecture 9: Monkeys with typewriters

Lecture 10: My own validation epic fail

Lecture 11: The test set

Lecture 12: What if the model doesn't pass the test?

Lecture 13: How not to be fooled by randomness

Lecture 14: Cross-validation

Lecture 15: Code time (cross validation)

Lecture 16: Cross-validation results summary

Lecture 17: AutoML

Lecture 18: Is AutoML a good idea?

Lecture 19: Red flags: Don't do this!

Lecture 20: Red flags summary and what to do instead

Lecture 21: Your job as a data scientist

Lecture 22: Kaggle notebook on validation and cross-validation

Lecture 23: 30-minute code assignment with new dataset!

Lecture 24: Notes on validation and testing

Lecture 25: Extra reading: Model retraining

Lecture 26: Extra reading: The Difference between Statistics and Machine Learning

Chapter 6: Common Mistakes

Lecture 1: Intro and recap

Lecture 2: Mistake #1: Data leakage

Lecture 3: The golden rule

Lecture 4: Helpful trick (feature importance)

Lecture 5: Real example of data leakage (part 1)

Lecture 6: Real example of data leakage (part 2)

Lecture 7: Another (funny) example of data leakage

Lecture 8: Mistake #2: Random split of dependent data

Lecture 9: Another example (insurance data)

Lecture 10: Mistake #3: Look-Ahead Bias

Lecture 11: Example solutions to Look-Ahead Bias

Lecture 12: Consequences of Look-Ahead Bias