dantonnoriega

# remove any old tmp files
old_tmp_files <- list.files(pattern = '^tmp', include.dirs = TRUE, full.names = TRUE)
invisible(unlink(old_tmp_files, recursive = TRUE))

# create a dark widget frame via work around
dark_widgetframe <- 
function(widget, background = '#666666FF', width = '100%', height = 420) {
  file = tempfile(pattern = "tmp", tmpdir = '.', fileext = ".html")
  selfcontained = FALSE
  libdir = NULL

dantonnoriega

# recreate the code from https://arxiv.org/pdf/1808.06399.pdf using greta (https://greta-dev.github.io/greta/)
library("DirichletReg")
Bld <- BloodSamples
Bld <- na.omit(Bld)
Bld$Smp <- DR_data(Bld[, 1:4])

# using greta
# !! requires the development version to run!
# devtools::install_github("greta-dev/greta@dev")
# convert data to matrix then greta data

dantonnoriega

# set up ---------------------------
library(furrr)

# use multiple clusters
ssh_username <- 'drn'
remote_ssh_configs <- c('a', 'b', 'e') # names for remote server (found in ~/.ssh/config e.g. Host a)
local_comp <- Sys.info()[["nodename"]] # get local computer name

# build cluster ----------------------------------------------------------------
system(command = "ps -axc | grep ssh | awk '{print $1}' | sort -u | xargs kill")

dantonnoriega

# recreate the code from https://arxiv.org/pdf/1808.06399.pdf using greta (https://greta-dev.github.io/greta/)
library("DirichletReg")
Bld <- BloodSamples
Bld <- na.omit(Bld)
Bld$Smp <- DR_data(Bld[, 1:4])

# simplex function. applies simplex to each row of a matrix.
# HUGE speed gains using matrix mat vs for loop!
simplex_mat <- function(x){
  exp_x <- exp(x)

dantonnoriega

# stan code from https://arxiv.org/pdf/1808.06399.pdf
library("DirichletReg")
Bld <- BloodSamples
Bld <- na.omit(Bld)
Bld$Smp <- DR_data(Bld[, 1:4])

stan_code <- '
data {
  int<lower=1> N;  // total number of observations
  int<lower=2> ncolY; // number of categories

dantonnoriega

library(tidyverse)
library(rstan)

## HOW THE SIMULATION LOOP BELOW WORKS -- ignore the shocks for now --------------
# quick example
I = 3 # individuals
M = 5 # "months"
S = 7 # sims
mat <- matrix(NA, I*M, S) # M rows (months) will be filled in at a time for each individual i across all S columns (simulations); records for individual i will populate rows ((i - 1)*M + 1):(i*M)

dantonnoriega

# inspired by https://stats.stackexchange.com/questions/174121/can-a-model-for-non-negative-data-with-clumping-at-zeros-tweedie-glm-zero-infl
# additions by Danton Noriega
library(statmod)
library(tweedie)
library(mgcv)

# generate fake mu (poisson count rates)
set.seed(1789)
x <- seq(1,100, by = .1)
mutrue <- exp(-1+x/25)

dantonnoriega

# a running list of really useful data.table tricks

# TOP 1,2 ... last row by some group id
## source: https://stackoverflow.com/questions/16325641/is-it-possible-to-extract-the-first-2-rows-for-each-date#comment23381259_16325932
## comment by @eddi
id <- c('date', 'userid')
dt[dt[, .I[1:2], by = id]$V1] # first 2 rows by id
dt[dt[, .I[.N], by = id]$V1] # last row by id (.N = length of group, .I = row index)

dantonnoriega

dantonnoriega

import sublime, sublime_plugin

class ExpandSelectionToEofCommand(sublime_plugin.TextCommand):
    def run(self, edit):
        v = self.view
        s = v.sel()
        eof = v.size()

        first_point = v.line(s[0]).a
        region_to_eof = sublime.Region(first_point, eof)