R workshop XX: Parallel Computing with R

Contributed by

Yuan Huang, PM Intern of SupStat Inc,

Tong He, Data Scientist of SupStat Inc.

Vivian Zhang, CTO of SupStat Inc, will deliver this workshop.

Content:

Our slides can be found:

http://nycdatascience.com/slides/parallel_R/index.html#1

http://nycdatascience.com/slides/parallel_R/examples_general.html

http://nycdatascience.com/slides/parallel_R/example_crossvalidataion.html

http://nycdatascience.com/slides/parallel_R/example_web.html

We will go over the steps toward parallel computing.

1.Whether the problem is parallel-able ?

2.Tips to improve the parallel computing's efficiency.

3.Implementation in R.

We will discuss how to do load balance, how to reduce parallel over-head, how to make sure each nodes have different random number and the few statistical models to be paralleled.

And do a overview of

1.Rmpi ( R interface to MPI; flexible; powerful, but more complex.)

2.Snow (will be used for backends with foreach package today)

3.multicore (work only on a single node and Linux-like machine)

4.parallel (hybrid package containing snow and multicore)

5.foreach (parallel backends doSNOW / doMPI / doMC)

In the end, we will give examples by using foreach package:

1. Bootstrapping: calculate CI for median.

2.Random Forest

3. Calculate the pairwise distance

4.Cross Validataion

5. Web scrapper