wrathematics

f = function(expr)
{
  for (i in 1:2)
    print(system.time(eval(expr))[[3]])
}

f(Sys.sleep(.4))
## [1] 0.401
## [1] 0

wrathematics

wrathematics

FROM alpine:3.10.2
RUN apk add R R-dev bash build-base

CMD ["/bin/bash"]

wrathematics

pbd = c(
  "hdfio",
  "hpcvis",
  "kazaam",
  "pbdBASE",
  "pbdCS",
  "pbdDEMO",
  "pbdDMAT",
  "pbdIO",
  "pbdML",

wrathematics

library(memuse)
library(hdfio)

f = nycflights13::flights

for (compression in c(0, 4, 9)){
  suppressWarnings(file.remove("flights.h5"))
  write_h5df(f, "flights.h5", compression=compression)
  print(Sys.filesize("flights.h5"))
}

wrathematics

library(inline)
csrc = "
  SEXP ret;
  PROTECT(ret = allocVector(LGLSXP, 2));

  LOGICAL(ret)[0] =  1;
  LOGICAL(ret)[1] = -1;

  UNPROTECT(1);
  return ret;

wrathematics

---
title: "A Bit on the Numerics Behind R's Linear Model Fitters"
output: html_document
---

The classical linear regression setup is that you want to solve a system of equations that looks something like:
$X_{m\times n}\beta_{n\times 1} = y_{m\times 1}$

In statistics/data analysis, it's pretty typical for $m>n$. If $X$ has ["full rank"](https://en.wikipedia.org/wiki/Rank_(linear_algebra)) (more on this later). In this case, [we can derive a closed form solution](https://en.wikipedia.org/wiki/Least_squares#Linear_least_squares) for $\beta$:

wrathematics

#include <stdio.h>
#include <papi.h>
#include <stdlib.h>

#define NUM_EVENTS 2

int main(int argc, char **argv)
{
  int a = atoi(argv[1]);
  int b = atoi(argv[2]);

wrathematics

#include <stdio.h>
#include <papi.h>

#define NUM_EVENTS 2

int main()
{
  int events[NUM_EVENTS] = {PAPI_L1_DCM, PAPI_L2_DCM};
  long long values[NUM_EVENTS];
  

wrathematics

// ----- count.c
// build with R CMD SHLIB count.c (or put it in the package, w/e)

#include <R.h>
#include <Rinternals.h>

SEXP colsum_which_eq0(SEXP nrows_, SEXP ncols_, SEXP J_)
{
  SEXP ret;
  const int nrows = INTEGER(nrows_)[0];