Machine Learning in Game Design

Explore what machine learning is and how supervised learning uses linear regression, multiple linear regression, and polynomial regression to predict a monster's challenge rating in Dungeons and Dragons.

Learn how machine learning uses data and training to build models. Focus on supervised learning, data labeling, and predicting outcomes through classification and regression, with applications in game design.

Explore linear regression, a method for predicting a single dependent value from a single independent value, and work with a simple output function.

Learn to implement linear regression in python, loading data with pandas, training with a learning rate, plotting results with matplotlib, and evaluating on a train-test split.

Learn multiple linear regression with several independent variables to predict miles per gallon, use feature elimination to improve accuracy, and apply dummy variables for categorical data.

Apply multiple linear regression in Python to predict monster challenge rating using hit points and categorical type, encoding with dummy variables and training with stochastic gradient descent.

Explore polynomial regression as a curve-based alternative to linear regression and learn forward and backward degree selection, gradient descent optimization, and scikit-learn implementation.

Develop a polynomial regression model in Python with scikit-learn, using degree three polynomial features, fitting a linear model, and evaluating with mean squared error on prepared data.

Identify that data is not all useful; in Dungeons and Dragons, monster hit points predict challenge rating, and apply linear, multiple linear, and polynomial regression with scikit learn, pandas, matplot.

Learn reinforcement learning fundamentals by training an agent to maximize scores in a Unity game using Q-learning, with actions such as do nothing, left, or right, and state-action-reward concepts.

Learn Q-learning, a reinforcement learning algorithm that updates a Q-value table from rewards and the maximum future Q value with a discount factor, and balance exploration and exploitation.

Implement a Q-learning workflow in Python and Unity, using ML-Agents to train a blue box agent to maximize reward via discrete actions and an epsilon-greedy policy.

State action reward state action, or sarsa, is a variation of q-learning that uses the next state and action, making it safer. A cliff with -100 reward illustrates updates.

Implement the SARSA algorithm in Python and Unity, training a policy by updating q-values with the next state and action, and compare with Q-learning while testing and inferring in Unity.

Explore reinforcement learning, including q-learning and SARSA, and how agents learn from state-action-reward in discrete environments. Evaluate when to use q-learning or SARSA, and balance greedy and epsilon-greedy exploration.

Explore evolutionary algorithms, including population, fitness, selection, crossover, and mutation, and train neural network–based agents in Unity to balance the ball on their heads.

Train an evolutionary algorithm for 12 agents in a Unity ML-Agents environment using Python, DEAP, and a neural-net brain, evolving continuous actions through selection, crossover, and mutation.

Explore evolutionary algorithms inspired by natural evolution, using individuals, selection, crossover, and mutation to evolve neural networks and behaviors in game design, with distributed Python tools guiding retraining across episodes.