Skip to content

Latest commit

 

History

History
89 lines (80 loc) · 4.57 KB

discussion.md

File metadata and controls

89 lines (80 loc) · 4.57 KB
Raw

What would you like your students to learn in a 2nd course on statistical computing?

Make it run, make it right, make it fast.

Background

  • Offering STA 633: Statistical computing and computation in Spring 2015
  • Schedule is 2 lectures (75 mins) and 1 lab (75 mins) per week
  • This is a 2nd course in statistical computing - pre-req is Colin Rundel's class
    • STA 523
    • Quite fast-paced - recommended books are
      • Advanced R - Wickham
      • R Packages - Wickham
    • Will cover Unix shell, make, git, markdown and programming in R
    • We will have a pretest to determine eligibility if students have not taken STA 523

Proposed learning objectives

  • Basically teach all the computing that we would personally like to see in a PhD student or postdoc working with us
    • Comfortable using both high (Python/Julia?/R) and low level languages (C/C++)
    • Understand data management and use of relational database
      • Working with "bad" data
        • Examples?
      • Hands-on exercise building a normalized database from a spreadsheet and querying it via SQL
    • Can build reproducible data analysis pipelines (testing + make + literate programming)
    • Can convert a statistical model (e.g. from manuscript or textbook) into a numerical algorithm
      • Understanding of basic algorithms for optimization, simulation and smoothing * Building blocks for large classes of statistical algorithms * What algorithms should students know?
      • Pragmatic usage of libraries for established numerical routines * Recommendations for C/C++ libraries
    • Can write code that is correct
      • How much and what kind of testing is appropriate?
      • How to test code with stochastic elements
    • Can write code that runs fast
      • Trade-off between computation and programmer time (premature optimization)
      • Some understanding of complexity trade-offs for algorithms and data structures
      • Benchmarking and profiling
      • JIT compilation
      • Writing native code
      • Exploiting multiple cores (threading, multiprocessing, OpenMP)
      • Exploiting multiple machines (MPI)
      • Exploiting GPUs (CUDA, maybe OpenCL)
      • Working with really big data (MapReduce)

Units

Unit 1: Reproducible analysis and introducing Python as a glue language (10%) Unit 2: Working with data - data munging and relational databases (10%) Unit 3: Exploratory data analysis and visualization (10%) Unit 4: Core statistical algorithms and libraries (40%) Unit 5: C bootcamp, code profiling and writing native code (15%) Unit 6: Parallel computing and working with big data (15%)

Discussion

  • Overall course objectives?
  • Overall course content?
    • Are there useful classes of topics we have left out?
    • Within each topic, what content should students learn?
      • Unit 1: Reproducible analysis and introducing Python as a glue language (10%)
      • Unit 2: Working with data - data munging and relational databases (10%)
      • Unit 3: Exploratory data analysis and visualization (10%)
      • Unit 4: Core statistical algorithms and libraries (40%)
      • Unit 5: C bootcamp, code profiling and writing native code (15%)
      • Unit 6: Parallel computing and working with big data (15%)
  • How can programming be taught effectively?
    • Every good programmer I know is self-taught ...
    • MCQs for rapid sanity check on level of understanding each week
    • Less talking, more doing - mini-project after each unit
    • Individual or group work?
  • What are statistical algorithms students should know?
    • Know the theory and how to use a good implementation
      • Teach understanding with toy example
      • Use library to solve more realistic problem
    • Examples
      • Linear algebra e.g. projection, normal equations
      • Optimization - e.g. Newton, IRLS, multivariate gradient descent, EM
      • Simulation - resampling methods, Monte Carlo, MCMC
      • Others? Smoothing, interpolation etc
  • What are good data sets and problems to use for teaching?
    • Bad data
    • Big data
    • Slow and fast versions