JuliaCon2016: Recap of Sessions and Discussions

1. JuliaCon2016
2. Actions
3. Wednesday Sessions
4. Friday Sessions

1. JuliaCon2016

"The third Julia conference will take place June 21st-25th, 2016 at the Massachusetts Institute of Technology in Cambridge, Massachusetts. "

2. Actions

0.4 changelog
xdata, oore foundation
distributed, high performance,
check remaining work on 0.5 https://github.com/JuliaLang/julia/milestones/0.5.0
what were the failures in CS education wrt the web in 1990
what were the goals in college programs that emphasized C support for core algorithms
what is the difference between expected work and education and does that drift create issues
Automatic Differentiation http://alexey.radul.name/ideas/2013/introduction-to-automatic-differentiation/
getLaplacian
Jacobians
Conjugate Gradient Methods for 2D Laplacian
performance profiling for FP with allocation, memory, and time space
http://quant-econ.net/jl/index.html
Quantitative Economics translated to JavaScript
https://www.manning.com/books/julia-in-action

3. Wednesday Sessions

3.1. Fortress Features and Lessons Learned, Guy Steele

https://en.wikipedia.org/wiki/Fortress_(programming_language)

Covered four of the 20 crazy programming ideas implemented in Forturs.

3.1.1. Exmaple: histogramWater

For an array of integers create a histogram, then let it rain on the histogram and find the volume of water.

sum the volume of water for each item
each item looks left and right to find the max

3.1.2. Features

libraries
symmetry
whitespace
preserve whiteboarded mathematics
requires understanding the precedence of operators of all operators
clear point: don't force yourself to have a parser define precendence built into a langage that would not be used consistently: use parenthesis and don't make it Facebook link-bait
general big sigma and comprehension
generators and reductions filter through the type system
overloading with types- trait Boolean extends { Any } excludes { String }

3.1.3. Notes

3.2. Gallium, Keno Fischer

https://github.com/Keno/Gallium.jl

3.2.1. Example: 1+2.0

  using Galllium
  @enter 1.0+2

Reviewed: s, bt, edit, nc, si
@code_llvm @code_native

3.2.2. Example: plotting

3.2.3. Example: gcd

  breakpoint(gcd)
  methods(gcd)

Any of the calls to the types supported by calls
macro expansion
new error message from repl
generator functions

3.3. Juno, a Julia IDE

http://junolab.org/

workspace available with 0.5
based off of Gallium changes
code vs. exploration wih Notebook

3.4. Julia for data science: current progress and future plans, Simon Byrne

dataframes and integration with Python, R, Spark, GraphLab
allocation of memory between sources
switching to maps (without union operations)
compound operations (filter then average) are greedy

3.5. TypedTables: type-safe data containers, Andy Ferris

DataFrames
TypedTables
GeneratedTypes

3.6. Music Information Retrieval in Julia, Jong Wook Kim

3.7. Using Julia as a Quick and Dirty Code Generator, Arch D. Robison

3.8. Automatic differentiation techniques used in JuMP, Miles Lubin

https://github.com/JuliaDiff/ReverseDiffSource.jl

nodes in memory can be a performance constraint
billions of long-lived objects that would need GCC
single vector of immutable objects
register functions

http://arxiv.org/abs/1508.01982

3.9. ForwardDiff.jl: Fast Derivatives Made Easy, Jarrett Revels

https://github.com/JuliaDiff/ForwardDiff.jl

Example: you inherit a function where you can't symbolically

automatic differentation (AD)

df(x)/dx = lim h→0, f(x+h)−f(x)/ h
central difference

lim h -> 0; f'(x) = f(x) + f(x + h) / h

complex step method
dual number method

Profiling:

f(x) = e^x / sqrt(sin^3(x) + cos^3(x))

Jacobians: J(g)(f)
Dual numbers: less, sin, * are implementation of chain rule and product rule

Test:

  function cumprod

  end

test the function: cumprod

cumprod(A[, dim])
Cumulative product along a dimension dim (defaults to 1). See also
cumprod!() to use a preallocated output array, both for performance and
to control the precision of the output (e.g. to avoid overflow).

write the Jacobian
create test of the input size
provide the timing
graph the error rates

3.10. Enabling reverse communication solvers and embedding Julia, Andy Greenwell

3.11. Patterns for building web apps with Escher.jl, Shashi Gowda

3.12. Unums 2.0: Implementing projective intervals and sets in Julia, Jason Merrill

https://www.crcpress.com/The-End-of-Error-Unum-Computing/Gustafson/p/book/9781482239867

second version of an idea that's under active development
good ideas that show issue
ranges
open and closed sets
issues with 0 and -0

3.12.1. Universal numbers

Superset of IEEE types,both 754 and 1788
Integers->floats->unums
No rounding, no overflow to ∞, no underflow to zero
Projective infinity implies closed interval on interval arithmetic

3.12.2. Tests

  var g = function(x) { return 1 / (x + 3 - x - 3); }

3.13. The design and use of extended precision floats, Jeffrey Sarnoff

from 1986: always double the bit width
float64 calculation with the LHC they would use double-doubles
example: precision on non-precise items: lasers cutting tumors don't have the level of precision
time is generally a better example
risk is also not that precise: individual or organizational risk thresholds are likely a single order