16-commuting-time.Rmd

# Commuting time {#commuting-time}

```{r commuting-time, include=FALSE}
chap <- 17
lc <- 0
rq <- 0
# **`r paste0("(LC", chap, ".", (lc <- lc + 1), ")")`**
# **`r paste0("(RQ", chap, ".", (rq <- rq + 1), ")")`**

knitr::opts_chunk$set(
  tidy = FALSE, 
  out.width = '\\textwidth', 
  fig.height = 4,
  warning = FALSE
  )

options(scipen = 99, digits = 3)

# Set random number generator see value for replicable pseudorandomness. Why 76?
# https://www.youtube.com/watch?v=xjJ7FheCkCU
set.seed(76)
```


```{r}
do_eval=file.exists("commute_times.csv")
```

Commuting time is often mentioned as a valid reason for leaving one employer for another. 


You are asked to calculate the commuting time for the employee working at your company. You decide that hyphotetically they must all arrive in the office at 8.30 am. In the Google you identified that there is google api for doing this, called [Distance Matrix API](https://developers.google.com/maps/documentation/distance-matrix/start?csw=1), which works for a matrix of origins and destinations. 

The information returned is based on the recommended route between start and end points and consists of rows containing duration and distance values for each pair." To use the Distance Matrix API, you must first activate the API in the Google Cloud Platform Console and obtain the proper authentication credentials. You need to provide your own API key in each request.

The documentation on how to do this is located here: https://developers.google.com/maps/documentation/distance-matrix/intro

A file with 200 fake names and addresses has been put together for you. The task is to add the commuting time next to each staff member. Normally google will charge us one dollar for checking 200 commuting times.

Please note that the departure_time specifies the desired time of departure must be in POSIXct. format and must be in the future (i.e.greater than sys.time()). If no value is specified it defaults to Sys.time().

Please note you can only specify one of arrival_time or departure_time, not both. If both are supplied, departure_time will be used.

```{r}
library(googleway)
library(tidyverse)
library(lubridate)

# Set up here your api key key=
key = 'AIzaSyCOly69PDrlPlM42I378p2lmvNs8I2w'

d <- dmy_hms("10/09/2018 08:30")

arrival <- as.numeric(d)


from <- c("SE3+8UQ,UK")
to <- c("E14+5EU,UK")

test <- google_distance(origins = from,
                        destinations = to,
                        mode = "walking",
                        arrival_time = as.POSIXct(arrival, origin = "1970-01-01", tz = "UTC"),
                        key=key)

test$rows$elements[[1]]$distance

```

To make all this simpler, we put together an R function which can be used to analyse many results. Note this calls all modes of transport for every employee making it more costly than just running one mode per employee.

```{r}
#' Get distance data between two points based on all the travel mode options. Works for many origin points.
#'
#' @param x A vector of origins in address or postcode format
#' @param dest A single destinationin address or postocde format
#' @param arrival_time A POSIXct datetime that folks need to arrive by
#' @param key A google distance API key
#' @param ... Additional options to pass to `google_distance()`
#'
#' @return Data.frame containing (typically) 4 rows per input element

google_distance_all =  function(x, dest, arrival_time, key, ...){
  
  # simple hygeine stuff
  gd = purrr::possibly(
    memoise::memoise(
      google_distance)
    , "Fail"
  )
  
  # Prep dataset
   interested_in = expand.grid(from=x, 
     mode=c("driving", "walking", "bicycling", "transit"), 
      stringsAsFactors = FALSE)
   # Perform google_distance calls for all combos
  purrr::map2(interested_in$from,interested_in$mode, 
     ~gd(.x, dest, mode=.y,
                        arrival_time = arrival_time,
                        key=key)
  ) %>% 
    # Extract relevant section
    purrr::map("rows") %>% 
    purrr::map("elements") %>% 
    purrr::flatten() %>% 
    # Simplify the data.frames
    purrr::map(unclass) %>% 
    purrr::map_df(purrr::flatten) %>% 
    # Add original lookup values
    cbind(interested_in)
}
```

This needs some packages installed to work
```{r}
if(!require(purrr)) install.packages("purrr")
if(!require(memoise)) install.packages("memoise")
if(!require(googleway)) install.packages("googleway")
```

We need some information about our google API account, where our office is, and when we're testing commute times for.

```{r}
office = "E14 5EU"
monday_9am = as.POSIXct("2018-12-03 09:00")
```

Then we can use this function to get data for our 200 employees

```{r eval=FALSE}
the_200 <- read_csv("https://hranalytics.netlify.com/data/200_staff_members.csv")
```

```{r read_data_commuting, echo=FALSE, warning=FALSE, message=FALSE}
the_200 <- read_csv("data/200_staff_members.csv")
```


```{r eval=do_eval}

results = google_distance_all(
  the_200$Postcode,
  office,
  arrival_time = monday_9am,
  key = key
)

write_csv(results, file.path("data"), "commute_times.csv")
```

```{r echo=FALSE, message=FALSE, warning=FALSE}
results = read_csv("data/commute_times.csv")
```

```{r}
results
```


Now that we have this data we can answer questions about our employee's commute times.

Who has the longest commutes by car?
```{r}
results %>% 
  filter(mode == "driving") %>% 
  mutate(hours=value/60^2) %>% 
  top_n(10, hours) %>% 
  arrange(desc(hours))
```

We can look at the distributions overall too.
```{r}
results %>% 
  mutate(hours=value/60^2) %>% 
  ggplot() +
  aes(x=hours) +
  geom_density() +
  scale_x_log10() +
  facet_wrap(~mode) +
  theme_minimal()
```