Week 13 Day 1 GGPLOT2 LAB DAY

Sun, 3 Apr 2016 22:44:39 -0700

See you in AQ 3148.1
We will be making use of this cheatsheet courtesy of the awesome RStudio people

Project 3

Sun, 3 Apr 2016 13:42:30 -0700

On the first day it was noted that “The next person to use your code is future you.” This project embodies that point by expanding on one of your previous projects. Final projects are in the form of an 8 minute presentation + 2 minutes of questions to be given in class Monday April 11. You must submit your slides from the presentation. Project 3 will be graded and combined with your grade from the previous version of the project. The resulting combined score will replace your grade from either project #1 or project #2 (whichever one project 3 extends.)

For grades Gi and Gj from project i and project j, where either i=3 or j=3 and 0≤Gi≤Gj≤100% without loss of generality, the final grade, GF, that will replace your original project score will be taken as:

This average puts much more weight on the higher scoring project. For example:

if Gi = 40% and Gj = 90%, GF = 85%
if Gi = 60% and Gj = 70%, GF = 67%
Or see the plots on the right.

To clarify GF will replace your score from the original project. Only projects 1 & 2 count towards
your final grade. This third project simply updates your score from project 1 or 2.

Beer Project:
Expand on the beer project and fix concerns raised during project #1. This time your task is to visit all of the breweries in the lower mainland. On this tour, patrons will start at brewery #1 (of your choice) and then hop aboard a bus to visit all of the lower mainland breweries (everything from Chilliwack west). Patrons may not survive a trip to all breweries so breweries should be location and quality prioritized. The route may not be the shortest but it should be the best. Make sure that your distance and quality metric is well behaved. The distance between breweries should be fairly compensated or penalized for its quality based on the beer dataset.

You will need to provide a guide with analytics so that patrons will know what they should try at each stop, possibly focusing on beer type or some other simpler summary rather than exact name so that the information can be easily shared with the class). Patrons should also be able to decide when they want to abandon the tour by looking at distance and quality measures. Emphasis is on the analytical value you add to the brewery tour company because that is what sells tickets.

Grading is based on:
Sensibleness of distance vs quality metric 33% (Describe the behaviour - it’s ok to just show the equation and talk about the behaviour)
Showcasing the resulting tour 33% (Map plots will probably require multiple zoom scales for clarity. You will need to outline how you know your answer is a good one.)
Beer Tour analytics and insights: 33% (information should be clear and easy to absorb. Big tables do not meet this goal, however plots usually do.)
Just showing up 1%

SSC Abstract Project:
Consider all of the SSC abstracts from 2011-2015 and model the topics that are contained therein. You could consider time or spatial changes in topics if you like. You must compare the topics studied from at least the 10 most dominant presenter groups (also include SFU if we are not one of them). The end result should be a grad school recommendation engine. Emphasis is on the value extracted from the topics and universities; topics by university, universities within topics… Comparisons should be easy to extract and insightful

Grading is based on:
Sensibleness of data cleaning choices 33% (Describe the process, you may wish to showcase one strange abstract that required some care. Data acquisition and cleaning is one of the big challenges here, convince us you did a proper job of it.)
Sensible choice of k and topic model used. 33% (Why did it make sense for you to choose VEM, VEM fixed, Gibbs, or CTM?, What is the k that you used and how did you choose it?)
Grad school analytics and insights: 33% (You might have too many topics to show them all but choose a few to interpret in your presentation. When you are comparing topics by university and comparing universities by topic the information should be clear and easy to absorb. Big tables do not meet this goal, however plots usually do. Do not use names like ‘topic 1’, ‘topic 2’, ..., make sure your names are informative throughout your presentation.)
Just showing up 1%

Week 12 Day 1

Wed, 30 Mar 2016 10:09:03 -0700

Neural Network Models

Week 11 Day 2: Fancy Model Selection

Wed, 23 Mar 2016 10:15:17 -0700

Thermodynamic Integration
Reversible Jump Markov Chain Monte Carlo
Bayes Factor software:

Week 11 Day 1: ABC and Likelihood-Free

Sun, 20 Mar 2016 22:15:38 -0700

Specifically today will focus on these two papers:

Wood, Simon. Statistical Inference for Noisy Nonlinear Ecological Dynamic Systems. Nature 466 (2010): doi:10.1038/nature09319.

Ryder, R, C P Robert, P Pudlo, and J M Marin. Approximate Bayesian Computational Methods. Statistics and Computing (2011).

Applications of Statistical Computing - Stat 853

Week 13 Day 1 GGPLOT2 LAB DAY

Project 3

Week 12 Day 1

Week 11 Day 2: Fancy Model Selection

Week 11 Day 1: ABC and Likelihood-Free