Supervised Machine Learning

Regression and Classification

• Regression vs classification
• Cost function
• Gradient Descent

Multiple Variable Regression

• Vectorization
• Gradient Descent with multiple variable
• Feature scaling
• Learning curve
• Feature engineering
• Polynomial Regression

Logistic Regression

• Regularization

Machine Learning Yearning

Development set and test set

• test and development
• single number evaluation

Error analysis

• eyeball and blackbox test
• mislabelled data

Bias and variance

• Optimal error rate
• tradeoff
• technique to optimize

Learning curves

Shows if we have high bias or high variance

Human level performance

Helps to determine the benchmark

Train data at distribution

The challenges, and the things to address (put some test data in training data to avoid overfit)

Data mismatch

Use training dev set to debug
Artificial data beware

Optimization verification

Use the score to apply both human result and test result to debug what to improve: the score algorithm or the learning algorithm

End-to-end deep learning

Tradeoffs between end-to-end deep learning and pipeline model
How to debug on pipeline