Models and methods students should know

[bensoltoff]

I've been thinking about next term and exactly what models and methods to cover. There's obviously a ton of potential approaches. Right now we have a lot of the basics covered but I wonder if we should be more ambitious and expose our students to more advanced modeling procedures like neural nets and other deep learning algorithms. The third term as currently structured focuses a lot on distributed computing platforms and methods (e.g. parallel processing, Hadoop, MapReduce, Spark, AWS, SQL and relational databases). At least last year, CS 123 covers a lot of that in probably far more detail than we'd want to do. We could still cover some of that in our spring Perspectives course focusing more on application to social science datasets, but I'm not sure how much we want to duplicate content they are learning elsewhere.

Instead, we could bleed Perspectives on Modeling into Perspectives on Advanced Computational Topics and add more algorithms and methods to students' toolkits. Right now, topics I can think of specific to modeling are:

1. Model/theory building
2. Data generating processes
3. Model estimation procedures
   • Maximum likelihood estimation
   • Generalized method of moments
4. Generalized linear models
   • Ordinary least squares
   • Logistic regression (binary, ordinal, and multinominal)
   • Poisson/negative binomial regression (regression for count variables)
5. Model assessment
   • Cross-validation
   • Bootstrapping
   • Ensemble model averaging
6. Flexible linear methods
   • Polynomial terms
   • LOESS
   • Generalized additive models
7. Tree-based methods
   • Decision trees
   • Random forests
8. Support vector machines
9. Kernal density estimation
10. Neural networks
11. Nearest-neighbors
12. Unsupervised learning
    • Clustering
    • Principle components analysis
    • Latent Direchlet allocation
13. Structural/theoretical models

Certainly that all doesn't fit into a 10 week course. But what about treating the final terms as a 20 week course? Spend 14 or 15 weeks on methods like the ones above and the remaining 5-6 weeks on distributed computing methods. Thoughts @rickecon?

[bensoltoff]

Closed per discussion with @rickecon