Udacity - Machine Learning Nanodegree

Udacity's Machine Learning Nanodegree project files and lecture notes.

This repository contains project files and lecture notes for Udacity's Machine Learning Engineer Nanodegree program.

Lecture note reference

Model evaluation and validation

Topics covered in this section:

• Model Evaluation
  Confusion matrix, F1 score, F-beta score, ROC curve
• Model selection
  Types of errors, various types of cross validation, learning curves, grid search

See lecture notes: here

Supervised learning

Topics covered in this section:

• Linear regression
  Absolute trick, advantages / disadvantages, L1 regularisation, L2 regularisation
• Decision trees
  Entropy, information gain, hyperparameters
• Naive bayes
  Prior probability, posterior probability, naive bayes
• Support vector machines
  Idea, different types of errors, basic working principle, etc.

See lecture notes: here

Unsupervised learning

Topics covered in this section:

• Clustering
  K-means clustering
• Hierarchical and density-based clustering
  Hierarchical clustering, single-link clustering, complete-link clustering, average-link clustering, ward's method, DB scan
• Gaussian mixture model and cluster validation
  EM algorithm, cluster validation, external indices, internal indices, adjusted rand indices, silhouette coefficient
• Feature scaling
• PCA
• Random projection and ICA
  Johnson-Lindenstrauss lemma, ICA, applications

See lecture notes: here

Reinforcement learning

Topics covered in this section:

• RL framework
  Reinforcement setting, episodic and continuous tasks, rewards hypothesis, cumulative reward, discounted reward, Markov decision process, Bellman equations, optimality, action-value functions,
• Dynamic programming
  Iterative policy evaluation, estimation of action values, policy improvement, policy iteration, truncated policy iteration, value iteration
• Monte Carlo methods
  Predicting state values, estimating action-values, incremental mean, policy evaluation, policy improvement, exploration-exploitation dilemma, GLIE MC control algorithm, constant-alpha GLIE MC control algorithm
• Temporal difference learning
  TD(0) prediction, action value estimation, solving the control problem, Sarsamax (Q-learning), expected Sarsa
• Deep reinforcement learning
  Discrete and continuous spaces, discretization, coarse coding, tile coding, function approximation, kernel functions, coarse coding
• Deep Q-Learning
  NNs as value functions, Monte Carlo learning, TD learning, Q-learning, Sarsa vs. Q-learning, experience replay, fixed Q-targets, different types of DQNs
• Policy-based methods
  Policy function approximation, stochastic policy search, policy gradients, Monte Carlo policy gradients, constrained policy gradients
• Actor-critic methods

See lecture notes: here

Deep learning

Topics covered in this section:

• Neuronal networks
  Perceptron trick, perceptron algorithm, sigmoid activation, maximum likelihood, cross entropy, logistic regression, perceptron and gradient descent
• Deep neural networks
  Regularization, dropout, vanishing gradients and activation function, momentum, keras optimisers
• Convolutional neural networks
  Model validation, image augmentation

See lecture notes: here