VII. Basics of probabilistic modelling
Given by Sven Laur
Brief summary: The concept of probability. Frequentism and Bayesianism. Corresponding design goals. Confidence intervals. Prior beliefs. Informed and uninformed observers. Bernoulli distribution. Binomial distribution. Naive-Bayes classifiers. Bayesian networks.
Slides: PDF
Video: UTTV(2016) UTTV(2015) UTTV(2014)
Literature:
- Bishop: Pattern Recognition and Machine Learning pages 67 - 120
- Weiss, Indurkhya, Zhang & Damerau: Text Mining: Predictive Methods for Analyzing Unstructured Information pages 52 - 70
- Bishop: Pattern Recognition and Machine Learning pages 137 - 161
- Ricci: Fitting distributions with R
Complementary exercises:
- Bishop: Pattern Recognition and Machine Learning exercises from pages 127 - 136 that are related to practical tasks
- Bishop: Pattern Recognition and Machine Learning pages 220 - 224
- Some practical exercises to confirm various hypotheses about how some variables are distributed in real life.
- Build a naive Bayes filter for detecting the spam. Use the Spambase Data Set for training and testing.
Free implementations:
- Various distribution in built-in stats package in R and
qqplot. - MASS package in R:
fitdistr,mvrnorm. - Predbayescor package in R that implements naive Bayes model