satRday-paris notes https://paris2019.satrdays.org/

satrday-paris.R

## /////////////////////////////////
##       satRday Paris notes 
##            (untidy)
## /////////////////////////////////

## Lionel Henry: Programming in the tidyverse
## tidyverse (data analysis, data manipulation, data cleaning) <----> r-lib (production prgramming)
##                                       dplyr, ggplot2, tidyr <----> vctrs, rland, devtools
## reprocibillity by few users                                 <----> reusability by many users

## TIDYVERSE
## https://principles.tidyverse.org/
## data (data masking), domain oriented (e.g. data cleaning), language like (verbs like mutate, arrange)

## rlang
## =tidyeval
## !! and !!! and enquo()
## requires new concepts
## delayed computation: loops or functions
## datamasking is not transitive! 

## options
## - fixed columns > no problem
## - mapping with purrr or *_all, *_if, *_at
## - tidyeval

## tidyeval:
## 1) pass the dots

library(margittr)
library(dplyr)

## group_by_at
my_count_by <- function(data, ...)
{
  data %>% group_by_at(vars(...)) %>% summarize(n = n())
}

my_count_by(data = fivethirtyeight::bechdel, c("year", "binary"))


mean_freq <- function(data, ...)
{
  data %>% summarize_at(vars(contains("alcohol")), mean)
}

mean_freq(data = fivethirtyeight::drug_use, c("alcohol"))


## ! select is the only verb that does a selection and not an action (adding a new column)
## selections have special properties: c(), - and :
## group_by also creates actions
## if you want selections at group_by, use group_by_at

my_wrap <- function(...) {
  
  facet_wrap(vars(...), labeller = label_both)
}


ggplot(data = fivethirtyeight::bechdel) + 
  geom_point(aes(year, budget_2013, color = binary)) + 
  my_wrap(Budget = cut_number(intgross_2013, 3)) + 
  hrbrthemes::theme_ipsum()
  
## 2) subsetting .data
## use .data as a pronoun , e.g. %>% group_by(.data$gender)

group_by_summarise <- function(.data, var1, var2)
{
 .data %>% group_by(.data[[var1]]) %>% summarise(mean = mean(.data[[var2]], na.rm = TRUE))
}

group_by_summarise(bechdel, "binary", "domgross")

fivethirtyeight::bechdel %>% group_by_summarise("binary", "domgross")


## 3) interpolation = tidyeval
## delay a blueprint with enquo() and insert it back with !! 
## e.g.   group_by(!!enquo(var1) %>% summarise(!!enquo(var2))

## example

#
group_by_summarise <- function(.data, var1, var2)
{
  .data %>% group_by(!!enquo(var1)) %>% summarise(mean = mean(!!enquo(var2), na.rm = TRUE))
}

group_by_summarise(bechdel, binary, domgross)

fivethirtyeight::bechdel %>% group_by_summarise(binary, domgross)


## 
drug <- fivethirtyeight::drug_use %>% select_at(vars("age", contains("use"))) %>% tidyr::gather(drug, use, -age)
heavy <- drug %>% group_by(drug) %>% summarize(max_larger_20 = max(use, na.rm = TRUE) > 20)
drug <- drug %>% full_join(heavy)

ggplot(data = drug) + 
  geom_path(aes(age, use, color = drug, group = drug)) + 
  facet_wrap(~max_larger_20, labeller = label_both)

ggplot(data = drug) + 
  geom_path(aes(age, use, color = drug, group = drug)) + 
  facet_wrap(~drug, scales = "free", labeller = label_both) + 
  theme_ipsum()


## https://twitter.com/dreamRs_fr
## etienne sanchez
## @fanny and @victor from dreamRs show the 
## shiny application making job search more fun with
## - text mining
## - hierarchical clustering
## - shiny application


## bea: football and graph

## graph
## cypher is a syntax for pattern recognition , not like SQL




## Henrik Bengtson
## -------------------------------------
## lapply
## parallel : not working for windows
## foreach: not perfect either

## ???
## large data
## actions done

library(future)
plan(multiprocess)
fa <- future(sum(1:50))
fa

fb <- future(sum(51:100))
value(fb)

fa %<-% prod(1:1e2) ##%<-% ## future syntax
fa

## microbenchmarking

plan(sequential)
#system.time(value(future(sum(1:1e2))))
system.time(x %<-% sum(1:1e6))

plan(multiprocess)
system.time(x %<-% sum(1:1e6))


## custom function

slow_sum <- function(x)
{
  SUM <- 0
  for (i in x)
  {
    SUM <- SUM + i
    Sys.sleep(0.5)  
  }
  return(SUM)
  
}

library(future)
plan(multicore)
fa <- future(slow_sum(1:50))
value(fa)
resolved(fa)








pryr::ast
resolved()

## other options
future_lapply
furrr::future_map


#- how to use bash tools within R: wrapper
#- distributing data to nodes 
#- using scheduler without script



## Suzanne Baert's theme ---------------------------------------------------------------
# points semitransparent
library(ggplot2)
## nice: theme_linedraw, theme_minimal

x <- c("theme_ipsum", "theme_ipsum_rc")

for (THEME in x)
{
  
  qp <- ggplot(data = fivethirtyeight::bechdel) + 
    geom_point(aes(domgross, intgross, color = binary)) + 
    my_wrap(c(budget = cut_number(budget, 3))) + 
    eval(call(THEME)) + labs(
      title="Seminal ggplot2 scatterplot example",
      subtitle="A plot that is only useful for demonstration purposes",
      caption="Brought to you by the letter 'g'") 
  print(qp)
}