Book Image

Mastering Machine Learning Algorithms. - Second Edition

By : Giuseppe Bonaccorso
Book Image

Mastering Machine Learning Algorithms. - Second Edition

By: Giuseppe Bonaccorso

Overview of this book

Mastering Machine Learning Algorithms, Second Edition helps you harness the real power of machine learning algorithms in order to implement smarter ways of meeting today's overwhelming data needs. This newly updated and revised guide will help you master algorithms used widely in semi-supervised learning, reinforcement learning, supervised learning, and unsupervised learning domains. You will use all the modern libraries from the Python ecosystem – including NumPy and Keras – to extract features from varied complexities of data. Ranging from Bayesian models to the Markov chain Monte Carlo algorithm to Hidden Markov models, this machine learning book teaches you how to extract features from your dataset, perform complex dimensionality reduction, and train supervised and semi-supervised models by making use of Python-based libraries such as scikit-learn. You will also discover practical applications for complex techniques such as maximum likelihood estimation, Hebbian learning, and ensemble learning, and how to use TensorFlow 2.x to train effective deep neural networks. By the end of this book, you will be ready to implement and solve end-to-end machine learning problems and use case scenarios.
Table of Contents (28 chapters)
26
Other Books You May Enjoy
27
Index

Q-learning

This algorithm was proposed by Watkins (in Watkins C.I.C.H., Learning from delayed rewards, Ph.D. Thesis, University of Cambridge, 1989 and furtherly analyzed in Watkins C.I.C.H., Dayan P., Technical Note Q-Learning, Machine Learning 8, 1992) as a more efficient alternative to SARSA. The main feature of Q-learning is that the TD update rule is immediately greedy with respect to the Q(st+1, a) function (assuming that the agent received the reward rt after performing the action at while in the state st):

The key idea is to compare the current Q(st, at) value with the maximum Q value achievable when the agent is in the successor state. Assuming , the previous equation can be transformed into a TDerror structure:

The first term is the current reward, the second is the discounted maximum reward that the agent can theoretically achieve using its current knowledge and the last one is the estimation of the Q function. As the policy must be GLIE, the convergence...