Epid 814 - Advanced Modeling Methods

Course schedule

Topics are subject to change and rearrangement! (We may also not get to all the topics listed)

Main ideas
- How to go from a question to a diagram to equations to code?
- What kind of model to use under different circumstances?
To Do: Everyone bring a specific question that they might want to use a model to answer, and we’ll workshop it together for the 9/16/19 class.
- Think about what knowledge about the system you have—what are the causal processes? What features/variables/processes might be important to include to answer your question? What do you think we might be able to ignore or simplify out?
- Also, if you think you might want to estimate parameters for the model, think a about what kind of data you might want to use! What can you measure/observe, e.g. through epidemiological studies or biological experiments?
We will go through each question/problem together and discuss how we might build a model that can answer that question, what the assumptions and limitations of different kinds of models are, etc.
Discussion prompts

(moved this up since it might be useful for making data sets we can play with for the ML and networks sections below)

Twitter data collection lab
Using R to pull data using API’s - Twitter, Facebook, and Google Maps
Collecting and working with twitter data
An interesting article about the ethics of using public data and our perception of privacy

Overall topics of interest: k-means clustering, support vector machines, singular value decomposition (possibly a little linear algebra review also?), basic neural networks
Clustering Methods Lecture
- Density-sensitive version of DBSCAN: OPTICS
- Another overview of DBSCAN and some of its variants
Clustering Lab
- Other useful tutorials:
  - Cluster analysis in R
  - Clustering in R Overview
Classification
Lab: Naive Bayes classification with twitter data
- Twitter topic datasets: Aliens, Vaccines, Vaccines 2
- Other useful tutorials
  - Naive Bayes in R for movie reviews
  - Naive Bayes with Titanic survivorship
  - Naive Bayes Overview - includes more details on cross validation, confusion matrices, Laplace smoother, etc.

Networks lecture
- Review of network concepts (centrality, modularity, etc.)
- Network simulation methods (Barabasi-Albert, Erdos-Renyi, etc.)
- Configuration models
Network data sets
- Mark Newman’s list of data sets
- UC Irvine Network Data Repository
- Facebook social network data from kaggle
- Colorado network data set
- Transitland: transit data for major cities and visualization packages for it: transitland github, visualizations 1, more visualizations