This is the R script/materials repository of the "Mastering R Skills" course in the 2012/2021 Spring term, part of the MSc in Business Analytics at CEU. For the previous edition, see 2018/2019 Spring, 2019/2020 Spring.
Table of Contents
- Schedule
- Syllabus
- Location
- Syllabus
- Technical Prerequisites
- Class materials
- API ingest and data transformation exercises
- Report on the current price of 0.42 BTC
- Report on the current price of 0.42 BTC in HUF
- Move helpers to a new R package
- Report on the price of 0.42 BTC in the past 30 days
- Report on the price of 0.42 BTC and 1.2 ETH in the past 30 days
- Report on the price of cryptocurrency assets read from a database
- Report on the price of cryptocurrency assets based on the transaction history read from a database
- Profiling, benchmarks
- Reporting exercises
- API ingest and data transformation exercises
- Home assignment
- References
Schedule
3 x 200 mins on May 5, 12 and 19:
- 15:30 - 17:00 session 1
- 17:00 - 17:30 break
- 17:30 - 19:00 session 2
Location
This class will take place online. Find the Zoom URL shared in Moodle.
Syllabus
Please find in the syllabus
folder of this repository.
Technical Prerequisites
-
Bookmark, watch or star this repository so that you can easily find it later
-
Please bring your own laptop and make sure to install R and RStudio before attending the first class!
💪 R packages to be installed from CRAN viainstall.packages
:data.table
httr
jsonlite
lubridate
ggplot2
scales
zoo
RMySQL
RSQLite
openxlsx
googlesheets
devtools
roxygen2
pander
logger
botor
(requires Python andboto3
Python module)
💪 R packages to be installed from GitHub viaremotes::install_github
:daroczig/binancer
daroczig/logger
daroczig/dbr
If you get stuck, feel free to use the preconfigured, shared RStudio Server at http://mr.ceudata.net (I will share the usernames and passwords at the start of the class). In such case, you can skip all the steps prefixed with "
💪 " as the server already have that configured. -
Join the #ba-mastering-4-2020 Slack channel in the
ceu-bizanalytics
Slack group. -
If you do not already have a GitHub account, create one
-
Create a new GitHub repository called
mastering-r
-
💪 Installgit
from https://git-scm.com/ -
💪 Verify that in RStudio, you can see the path of thegit
executable binary in the Tools/Global Options menu's "Git/Svn" tab -- if not, then you might have to restart RStudio (if you installed git after starting RStudio) or installed git by not adding that to the PATH on Windows. Either way, browse the "git executable" manually (in somebin
folder look for theegit
executable file). -
Create an RSA key via Tools/Global options/Git/Create RSA Key button (optionally with a passphrase for increased security -- that you have to enter every time you push and pull to and from GitHub), then copy the public key (from
~/.ssh/id_rsa.pub
) and add that to you SSH keys on your GitHub profile. -
Create a new project in RStudio choosing "version control", then "git" and paste the SSH version of the repo URL copied from GitHub (from point 4) in the pop-up -- now RStudio should be able to download the repo. If it asks you to accept GitHub's fingerprint, say "Yes".
-
If RStudio/git is complaining that you have to set your identity, click on the "Git" tab in the top-right panel, then click on the Gear icon and then "Shell" -- here you can set your username and e-mail address in the command line, so that RStudio/git integration can work. Use the following commands:
$ git config --global user.name "Your Name" $ git config --global user.email "Your e-mail address"
Close this window, commit, push changes, all set.
Find more resources in Jenny Bryan's "Happy Git and GitHub for the useR" tutorial if in doubt or contact me.
Class materials
Report on the current price of 0.42 BTC
We have 0.42 Bitcoin. Let's write an R script reporting on the current value of this asset in USD.
Click here for a potential solution ...
library(devtools)
install_github('daroczig/binancer')
library(binancer)
coin_prices <- binance_ticker_all_prices()
library(data.table)
coin_prices[from == 'BTC' & to == 'USDT', to_usd]
## alternative solution
coin_prices <- binance_coins_prices()
coin_prices[symbol == 'BTC', usd]
## don't forget that we need to report on the price of 0.42 BTC instead of 1 BTC
coin_prices[symbol == 'BTC', usd * 0.42]
Report on the current price of 0.42 BTC in HUF
Let's do the same report as above, but instead of USD, now let's report in Hungarian Forints.
Click here for a potential solution ...
## How to get USD/HUF rate?
## See eg https://exchangeratesapi.io for free API access
## Loading data without any dependencies
## https://api.exchangerate.host/latest
## https://api.exchangerate.host/latest?base=USD
readLines('https://api.exchangerate.host/latest?base=USD')
## Parse JSON
library(jsonlite)
fromJSON(readLines('https://api.exchangerate.host/latest?base=USD'))
fromJSON('https://api.exchangerate.host/latest?base=USD')
## Extract the USD/HUF exchange rate from the list
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
coin_prices[symbol == 'BTC', 0.42 * usd * usdhuf]
Click here for a potential solution ... after cleaning up
## loading requires packages on the top of the script
library(binancer)
library(httr)
## constants
BITCOINS <- 0.42
## get Bitcoin price in USD
coin_prices <- binance_coins_prices()
btcusdt <- coin_prices[symbol == 'BTC', usd]
## get USD/HUF exchange rate
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
## report
BITCOINS * btcusdt * usdhuf
Click here for a potential solution ... with logging
library(binancer)
library(httr)
library(data.table)
library(logger)
BITCOINS <- 0.42
coin_prices <- binance_coins_prices()
log_info('Found {coin_prices[, .N]} coins on Binance')
btcusdt <- coin_prices[symbol == 'BTC', usd]
log_info('The current Bitcoin price is ${btcusdt}')
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
log_info('1 USD currently costs {usdhuf} Hungarian Forints')
log_eval(forint(BITCOINS * btcusdt * usdhuf), level = INFO)
log_info('{BITCOINS} Bitcoins now worth {round(btcusdt * usdhuf * BITCOINS)} HUF')
Click here for a potential solution ... with validating values received from the API
library(binancer)
library(httr)
library(data.table)
library(logger)
library(checkmate)
BITCOINS <- 0.42
coin_prices <- binance_coins_prices()
log_info('Found {coin_prices[, .N]} coins on Binance')
btcusdt <- coin_prices[symbol == 'BTC', usd]
log_info('The current Bitcoin price is ${btcusdt}')
assert_number(btcusdt, lower = 1000)
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
log_info('1 USD currently costs {usdhuf} Hungarian Forints')
assert_number(usdhuf, lower = 250, upper = 500)
log_info('{BITCOINS} Bitcoins now worth {round(btcusdt * usdhuf * BITCOINS)} HUF')
Click here for a potential solution ... with auto-retries for API errors
library(binancer)
library(httr)
library(data.table)
library(logger)
library(checkmate)
BITCOINS <- 0.42
get_usdhuf <- function() {
tryCatch({
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
assert_number(usdhuf, lower = 250, upper = 400)
}, error = function(e) {
## str(e)
log_error(e$message)
Sys.sleep(1)
get_usdhuf()
})
log_info('1 USD={usdhuf} HUF')
usdhuf
}
get_bitcoin_price <- function() {
tryCatch({
btcusdt <- binance_coins_prices()[symbol == 'BTC', usd]
assert_number(btcusdt, lower = 1000)
log_info('The current Bitcoin price is ${btcusdt}')
btcusdt
},
error = function(e) {
log_error(e$message)
Sys.sleep(1)
get_bitcoin_price()
})
}
log_info('{BITCOINS} Bitcoins now worth {round(get_bitcoin_price() * get_usdhuf() * BITCOINS)} HUF')
Click here for a potential solution ... with auto-retries for API errors with exponential backoff
get_usdhuf <- function(retried = 0) {
tryCatch({
## httr
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
assert_number(usdhuf, lower = 250, upper = 400)
}, error = function(e) {
## str(e)
log_error(e$message)
Sys.sleep(1 + retried ^ 2)
get_usdhuf(retried = retried + 1)
})
log_info('1 USD={usdhuf} HUF')
usdhuf
}
Click here for a potential solution ... with better currency formatter
round(btcusdt * usdhuf * BITCOINS)
format(btcusdt * usdhuf * BITCOINS, big.mark = ',', digits = 10)
format(btcusdt * usdhuf * BITCOINS, big.mark = ',', digits = 6)
library(scales)
dollar(btcusdt * usdhuf * BITCOINS)
dollar(btcusdt * usdhuf * BITCOINS, prefix = '', suffix = ' HUF')
forint <- function(x) {
dollar(x, prefix = '', suffix = ' HUF')
}
forint(get_bitcoin_price() * get_usdhuf() * BITCOINS)
Move helpers to a new R package
initial pkg version for saying hello
git config --global user.email "you@example.com" git config --global user.name "Your Name"
-
Click File / New Project / New folder and create a new R package (maybe call it
mr
, also create a git repo for it) -- that will fill in your newly created folder with a package skeleton delivering thehello
function in thehello.R
file. -
Get familiar with:
-
the
DESCRIPTION
file- semantic versioning: https://semver.org
- open-source license, see eg http://r-pkgs.had.co.nz/description.html#license or https://rstats-pkgs.readthedocs.io/en/latest/licensing.html
-
the
R
subfolder -
the
man
subfolder -
the
NAMESPACE
file
-
-
Install the package (in the Build menu), load it and try
hello()
, then?hello
-
Create a git repo (if not done that already) and add/commit this package skeleton
-
Add a new function called
forint
in theR
subfolder:forint.R
forint <- function(x) { dollar(x, prefix = '', suffix = ' HUF') }
-
Install the package, re-load it, and try running
forint
eg calling on42
-- realize it's failing -
After loading the
scales
package (that delivers thedollar
function), it works ... we need to prepare our package to loadscales::dollar
without user interventation -
Also, look at the docs of
forint
-- realize it's missing, so let's learn aboutroxygen2
and update theforint.R
file to explicitely list the function to be exported and note thatdollar
is to be imported from thescales
package:forint.R
#' Formats Hungarian Forint #' @param x number #' @return string #' @export #' @importFrom scales dollar #' @examples #' forint(100000) #' forint(10.3241245125125) forint <- function(x) { dollar(x, prefix = '', suffix = ' HUF') }
-
Run
roxygen2
on the package by enabling it in the "Build" menu's "Configure Build Tools", then "Document" it (if there's no such option, probably you need to install theroxygen2
package first), and make sure to check what changes happened in theman
,NAMESPACE
(you might need to delete the original one) andDESCRIPTION
files. It's also a good idea to automatically runroxygen2
before each install, so I'd suggests marking that option as well. The resulting files should look something like:DESCRIPTION
Package: mr Type: Package Title: Demo R package for the Mastering R class Version: 0.1.0 Author: Gergely <***@***.***> Maintainer: Gergely <***@***.***> Description: Demo R package for the Mastering R class License: AGPL Encoding: UTF-8 LazyData: true RoxygenNote: 7.1.0 Imports: scales
NAMESPACE
# Generated by roxygen2: do not edit by hand export(forint) importFrom(scales,dollar)
-
Keep committing to the git repo
-
Delete
hello.R
and rerunroxygen2
/ reinstall the package -
Add a new function that gets the most exchange rate for USD/HUF:
converter.R
#' Look up the value of a US Dollar in Hungarian Forints #' @param retried number of times the function already failed #' @return number #' @export #' @importFrom jsonlite fromJSON #' @importFrom logger log_error log_info #' @importFrom checkmate assert_number get_usdhuf <- function(retried = 0) { tryCatch({ ## httr usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF assert_number(usdhuf, lower = 250, upper = 400) }, error = function(e) { ## str(e) log_error(e$message) Sys.sleep(1 + retried ^ 2) get_usdhuf(retried = retried + 1) }) log_info('1 USD={usdhuf} HUF') usdhuf }
-
Now you can run the original R script hitting the Binance and ExchangeRatesAPI by using these helper functions:
library(binancer)
library(logger)
log_threshold(TRACE)
library(scales)
library(mr)
BITCOINS <- 0.42
log_info('Number of Bitcoins: {BITCOINS}')
usdhuf <- get_usdhuf()
btcusd <- binance_coins_prices()[symbol == 'BTC', usd]
log_info('1 BTC={dollar(btcusd)}')
log_info('My crypto fortune is {forint(BITCOINS * btcusd * usdhuf)}')
Report on the price of 0.42 BTC in the past 30 days
If you have missed the above steps on creating the R package with the required helpers, you can install the above version of mr
via:
devtools::install_github('daroczig/CEU-R-mastering-demo-pkg')
Let's do the same report as above, but instead of reporting the most recent value of the asset, let's report on the daily values from the past 30 days.
Click here for a potential solution ... with fixed USD/HUF exchange rate
library(binancer)
library(httr)
library(data.table)
library(logger)
library(ggplot2)
library(mr)
## ########################################################
## CONSTANTS
BITCOINS <- 0.42
## ########################################################
## Loading data
## USD in HUF
usdhuf <- get_usdhuf()
## Bitcoin price in USD
btcusdt <- binance_klines('BTCUSDT', interval = '1d', limit = 30)
str(btcusdt)
balance <- btcusdt[, .(date = as.Date(close_time), btcusd = close)]
str(balance)
balance[, btchuf := btcusd * usdhuf]
balance[, btc := BITCOINS]
balance[, value := btc * btchuf]
str(balance)
## ########################################################
## Report
ggplot(balance, aes(date, value)) +
geom_line() +
xlab('') +
ylab('') +
scale_y_continuous(labels = forint) +
theme_bw() +
ggtitle('My crypto fortune',
subtitle = paste(BITCOINS, 'BTC'))
Click here for a potential solution ... with daily corrected USD/HUF exchange rate
library(binancer)
library(httr)
library(data.table)
library(logger)
library(scales)
library(ggplot2)
library(mr)
## ########################################################
## CONSTANTS
BITCOINS <- 0.42
## ########################################################
## Loading data
## USD in HUF
usdhuf <- fromJSON('https://api.exchangerate.host/latest?base=USD&symbols=HUF')$rates$HUF
## try with a single date?
fromJSON('https://api.exchangerate.host/2021-05-01?base=USD&symbols=HUF')
## no, it's just a single day
# fromJSON('https://api.exchangerate.host/timeseries?start_date=2021-05-01&base=USD&symbols=HUF')
## need end
fromJSON('https://api.exchangerate.host/timeseries?start_date=2021-05-01&end_date=2021-05-05&base=USD&symbols=HUF')
## we can do a much better job!
library(httr)
response <- GET(
'https://api.exchangerate.host/timeseries',
query = list(
start_date = Sys.Date() - 30,
end_date = Sys.Date(),
base = 'USD',
symbols = 'HUF'
))
exchange_rates <- content(response)
str(exchange_rates)
exchange_rates <- exchange_rates$rates
library(data.table)
usdhuf <- data.table(
date = as.Date(names(exchange_rates)),
usdhuf = as.numeric(unlist(exchange_rates)))
str(usdhuf)
## Bitcoin price in USD
btcusdt <- binance_klines('BTCUSDT', interval = '1d', limit = 30)
str(btcusdt)
balance <- btcusdt[, .(date = as.Date(close_time), btcusd = close)]
str(balance)
str(usdhuf)
balance <- merge(balance, usdhufs, by = 'date')
balance[, btchuf := btcusdt * usdhuf]
balance[, btc := 0.42]
balance[, value := btc * btchuf]
## ########################################################
## Report
ggplot(balance, aes(date, value)) +
geom_line() +
xlab('') +
ylab('') +
scale_y_continuous(labels = forint) +
theme_bw() +
ggtitle('My crypto fortune',
subtitle = paste(BITCOINS, 'BTC'))
Now let's update the get_usdhuf
function to take start and end dates!
converter.R
#' Look up the value of a US Dollar in Hungarian Forints
#' @param start_date date
#' @param end_date date
#' @param retried number of times the function already failed
#' @return \code{data.table} object with dates and values
#' @export
#' @importFrom httr GET content
#' @importFrom logger log_error log_info
#' @importFrom checkmate assert_number
#' @importFrom data.table data.table
get_usdhufs <- function(start_date = Sys.Date(), end_date = Sys.Date(), retried = 0) {
tryCatch({
response <- response <- GET(
'https://api.exchangerate.host/timeseries',
query = list(
start_date = start_date,
end_date = end_date,
base = 'USD',
symbols = 'HUF'
))
exchange_rates <- content(response)$rates
usdhuf <- data.table(
date = as.Date(names(exchange_rates)),
usdhuf = as.numeric(unlist(exchange_rates)))
assert_numeric(usdhuf$usdhuf, lower = 250, upper = 400)
}, error = function(e) {
## str(e)
log_error(e$message)
Sys.sleep(1 + retried ^ 2)
get_usdhufs(retried = retried + 1)
})
usdhuf
}
Report on the price of 0.42 BTC and 1.2 ETH in the past 30 days
Let's do the same report as above, but now we not only have 0.42 Bitcoin, but 1.2 Ethereum as well.
Click here for a potential solution ...
library(binancer)
library(httr)
library(data.table)
library(logger)
library(scales)
library(ggplot2)
library(mr)
## ########################################################
## CONSTANTS
BITCOINS <- 0.42
ETHEREUMS <- 1.2
## ########################################################
## Loading data
## USD in HUF
usdhufs <- get_usdhufs(start_date = Sys.Date() - 40, end_date = Sys.Date())
## Cryptocurrency prices in USD
btcusdt <- binance_klines('BTCUSDT', interval = '1d', limit = 30)
ethusdt <- binance_klines('ETHUSDT', interval = '1d', limit = 30)
coinusdt <- rbind(btcusdt, ethusdt)
str(coinusdt)
## oh no, how to keep the symbol??
balance <- coinusdt[, .(date = as.Date(close_time), btcusd = close, symbol = ???)]
## DRY (don't repeat yourself)
balance <- rbindlist(lapply(c('BTC', 'ETH'), function(s) {
binance_klines(paste0(s, 'USDT'), interval = '1d', limit = 30)[, .(
date = as.Date(close_time),
usdt = close,
symbol = s
)]
}))
balance[, amount := switch(
symbol,
'BTC' = BITCOINS,
'ETH' = ETHEREUMS,
stop('Unsupported coin')),
by = symbol]
str(balance)
## rolling join
setkey(balance, date)
setkey(usdhufs, date)
balance <- usdhufs[balance, roll = TRUE]
str(balance)
balance[, value := amount * usdt * usdhuf]
str(balance)
## ########################################################
## Report
ggplot(balance, aes(date, value, fill = symbol)) +
geom_col() +
xlab('') +
ylab('') +
scale_y_continuous(labels = forint) +
theme_bw() +
ggtitle(
'My crypto fortune',
subtitle = balance[date == max(date), paste(paste(amount, symbol), collapse = ' + ')])
Report on the price of cryptocurrency assets read from a database
-
💪 Create a new MySQL account and database at Amazon AWS and don't forget to set an "inital database name". -
Log in and give a try with MySQL client:
mysql -h mastering-r.cjue7wnqdgs4.eu-west-1.rds.amazonaws.com -u admin -p
-
💪 Installdbr
from GitHub:library(devtools) install_github('daroczig/logger') install_github('daroczig/dbr')
-
💪 Installbotor
as well to be able to use encrypted credentials (note that this requires you to install Python first and thenpip install boto3
as well):install_github('daroczig/botor')
-
Set up a YAML file (menu: new file/text file, save as
databases.yml
) for the database connection, something like:remotemysql: host: ... port: 3306 dbname: ... user: ... drv: !expr RMySQL::MySQL() password: ...
-
Set up
dbr
to use that YAML file:options('dbr.db_config_path' = '/path/to/databases.yml')
-
Create a table for the balances and insert some records:
library(dbr) db_config('remotemysql') db_query('CREATE TABLE coins (symbol VARCHAR(3) NOT NULL, amount DOUBLE NOT NULL DEFAULT 0)', 'remotemysql') db_query('TRUNCATE TABLE coins', 'remotemysql') db_query('INSERT INTO coins VALUES ("BTC", 0.42)', 'remotemysql') db_query('INSERT INTO coins VALUES ("ETH", 1.2)', 'remotemysql')
-
Write the reporting script, something like:
Click here for a potential solution ...
library(binancer) library(httr) library(data.table) library(logger) library(scales) library(ggplot2) library(mr) library(dbr) options('dbr.db_config_path' = '/path/to/databases.yml') options('dbr.output_format' = 'data.table') ## ######################################################## ## Loading data ## Read actual balances from the DB balance <- db_query('SELECT * FROM coins', 'remotemysql') ## Look up cryptocurrency prices in USD and merge balances balance <- rbindlist(lapply(balance$symbol, function(s) { binance_klines(paste0(s, 'USDT'), interval = '1d', limit = 30)[, .( date = as.Date(close_time), usdt = close, symbol = s, amount = balance[symbol == s, amount] )] })) ## USD in HUF usdhufs <- get_usdhufs(start_date = Sys.Date() - 40, end_date = Sys.Date()) ## rolling join USD/HUF exchange rate to balances setkey(balance, date) setkey(usdhufs, date) balance <- usdhufs[balance, roll = TRUE] ## DT[i, j, by = ...] ## compute daily values in HUF balance[, value := amount * usdt * usdhuf] ## ######################################################## ## Report ggplot(balance, aes(date, value, fill = symbol)) + geom_col() + xlab('') + ylab('') + #scale_y_continuous(labels = forint) + theme_bw() + ggtitle( 'My crypto fortune', subtitle = balance[date == max(date), paste(paste(amount, symbol), collapse = ' + ')])
-
Rerun the above report after inserting two new records to the table:
db_query("INSERT INTO coins VALUES ('NEO', 100)", 'remotemysql') db_query("INSERT INTO coins VALUES ('LTC', 25)", 'remotemysql')
Report on the price of cryptocurrency assets based on the transaction history read from a database
Let's prepare the transactions table:
library(dbr)
options('dbr.db_config_path' = '/path/to/database.yml')
options('dbr.output_format' = 'data.table')
db_query('
CREATE TABLE transactions (
date TIMESTAMP NOT NULL,
symbol VARCHAR(3) NOT NULL,
amount DOUBLE NOT NULL DEFAULT 0)',
db = 'remotemysql')
db_query('TRUNCATE TABLE transactions', 'remotemysql')
db_query('INSERT INTO transactions VALUES ("2020-01-01 10:42:02", "BTC", 1.42)', 'remotemysql')
db_query('INSERT INTO transactions VALUES ("2020-01-01 10:45:20", "ETH", 1.2)', 'remotemysql')
db_query('INSERT INTO transactions VALUES ("2020-02-28", "BTC", -1)', 'remotemysql')
db_query('INSERT INTO transactions VALUES ("2020-04-13", "NEO", 100)', 'remotemysql')
db_query('INSERT INTO transactions VALUES ("2020-04-20 12:12:21", "LTC", 25)', 'remotemysql')
Click here for a potential solution for the report ...
library(binancer)
library(httr)
library(data.table)
library(logger)
library(scales)
library(ggplot2)
library(zoo)
library(mr)
## ########################################################
## Loading data
## Read transactions from the DB
transactions <- db_query('SELECT * FROM transactions', 'remotemysql')
## Prepare daily balance sheets
balance <- transactions[, .(date = as.Date(date), amount = cumsum(amount)), by = symbol]
balance
## Transform long table into wide
balance <- dcast(balance, date ~ symbol)
balance
## Add missing dates
dates <- data.table(date = seq(from = Sys.Date() - 30, to = Sys.Date(), by = '1 day'))
balance <- merge(balance, dates, by = 'date', all.x = TRUE, all.y = TRUE)
balance
## Fill in missing values between actual balances
balance <- na.locf(balance)
## Fill in remaining missing values with zero
balance[is.na(balance)] <- 0
## Transform wide table back to long format
balance <- melt(balance, id.vars = 'date', variable.name = 'symbol', value.name = 'amount')
balance
## Get crypto prices
prices <- rbindlist(lapply(as.character(unique(balance$symbol)), function(s) {
binance_klines(paste0(s, 'USDT'), interval = '1d', limit = 30)[
, .(date = as.Date(close_time), symbol = s, usdt = close)]
}))
balance <- merge(balance, prices, by = c('date', 'symbol'), all.x = TRUE, all.y = FALSE)
## Merge USD/HUF rate
response <- GET(
'https://api.exchangerate.host/timeseries',
query = list(start_date = Sys.Date() - 30, end_date = Sys.Date(),
base = 'USD', symbols = 'HUF'))
exchange_rates <- content(response)$rates
usdhufs <- data.table(
date = as.Date(names(exchange_rates)),
usdhuf = as.numeric(unlist(exchange_rates)))
setkey(balance, date)
setkey(usdhufs, date)
balance <- usdhufs[balance, roll = TRUE]
## compute daily values in HUF
balance[, value := amount * usdt * usdhuf]
## ########################################################
## Report
ggplot(balance, aes(date, value, fill = symbol)) +
geom_col() +
ylab('') + scale_y_continuous(labels = forint) +
xlab('') +
theme_bw() +
ggtitle(
'My crypto fortune',
subtitle = balance[date == max(date), paste(paste(amount, symbol), collapse = ' + ')])
Profiling, benchmarks
Breaking down the a single run of the get_usdhuf
function to see which component is slow and taking up resources:
## devtools::install_github('daroczig/CEU-R-mastering-demo-pkg')
library(mr)
library(profvis)
profvis({
get_usdhuf()
})
profvis({
get_usdhuf()
}, interval = 0.005)
A more realistic example: is ggplot2
indeed slow when generating scatter plots on a dataset with larger number of observations?
NOTE first run for library call
profvis({
library(ggplot2)
x <- ggplot(diamonds, aes(price, carat)) + geom_point()
print(x)
})
system.time(x <- ggplot(diamonds, aes(price, carat)) + geom_point())
Pipe VS Bracket:
library(data.table)
library(dplyr)
dt <- data.table(diamonds)
profvis({
dt[, sum(carat), by = color][order(color)]
group_by(dt, color) %>% summarise(price = sum(carat))
})
## run too quickly for profiling ...
library(microbenchmark)
results <- microbenchmark(
aggregate(dt$carat, by = list(dt$color), FUN = sum),
dt[, sum(carat), by = color][order(color)],
group_by(dt, color) %>% summarise(price = sum(carat)),
times = 100)
results
plot(results)
autoplot(results)
library(bench)
## needs to make sure that resulting objects are the same
results <- bench::mark(
as.data.frame(dt[, .(price = sum(carat)), by = color][order(color)]),
as.data.frame(group_by(dt, color) %>% summarize(price = sum(carat)))
)
results
autoplot(results)
## revisit benchmarking creating and printing ggplot
results <- microbenchmark(
x <- ggplot(diamonds, aes(price, carat)) + geom_point(),
print(x),
times = 10)
Also check out dtplyr
!
More examples at https://rstudio.github.io/profvis/examples.html
Reporting exercises
Connecting to and exploring the SQLite database
Download and extract the database file:
## download database file
download.file('http://bit.ly/CEU-R-ecommerce', 'ecommerce.zip', mode = 'wb')
unzip('ecommerce.zip')
Install the SQLite client on your operating system and then use the sqlite3 ecommerce.sqlite3
command to enter the command-line SQLite client to browse the database:
-- list tables in the database
.tables
-- show the structure of the sales table
.schema sales
-- show the first 5 rows of the table
select * from sales limit 5
-- tweak how the rows are shown
.headers on
.mode column
select * from sales limit 5
-- count number of rows in the table
SELECT COUNT(*) FROM sales;
-- count number of rows in January 2011 (lack of proper date/time handling in SQLite)
SELECT COUNT(*)
FROM sales
WHERE SUBSTR(InvoiceDate, 7, 4) || SUBSTR(InvoiceDate, 1, 2) || SUBSTR(InvoiceDate, 4, 2)
BETWEEN '20110101' AND '20110131'
-- check on the date format
SELECT InvoiceDate FROM sales ORDER BY random() LIMIT 25;
-- count the number of rows per month
SELECT
SUBSTR(InvoiceDate, 7, 4) || SUBSTR(InvoiceDate, 1, 2) AS month,
COUNT(*)
FROM sales
GROUP BY month
ORDER BY month;
Let's switch to R!
Connect to SQLite from R
Create a database config file for the dbr
package:
ecommerce:
drv: !expr RSQLite::SQLite()
dbname: /path/to/ecommerce.sqlite3
Update your dbr
settings to use the config file:
library(dbr)
options('dbr.db_config_path' = '/path/to/database.yml')
options('dbr.output_format' = 'data.table')
sales <- db_query('SELECT * FROM sales', 'ecommerce')
str(sales)
## explore and fix the invoice date column
sales[, sample(InvoiceDate, 25)]
sales[, InvoiceDate := as.POSIXct(InvoiceDate, format = '%m/%d/%Y %H:%M')]
## see fasttime::fastPOSIXct
## number of sales per month like in SQL
library(lubridate)
sales[, .N, by = month(InvoiceDate)]
sales[, .N, by = year(InvoiceDate)]
sales[, .N, by = paste(year(InvoiceDate), month(InvoiceDate))]
# slow
sales[, .N, by = as.character(InvoiceDate, format = '%Y %m')]
# smart
sales[, .N, by = floor_date(InvoiceDate, 'month')]
system.time(sales[, .N, by = as.character(InvoiceDate, format = '%Y %m')])
system.time(sales[, .N, by = floor_date(InvoiceDate, 'month')])
library(microbenchmark)
microbenchmark(
sales[, .N, by = as.character(InvoiceDate, format = '%Y %m')],
sales[, .N, by = floor_date(InvoiceDate, 'month')],
times = 10)
## number of items per country
sales[, .N, by = Country]
sales[, .N, by = Country][order(-N)]
Aggregate transaction items into invoice summary
invoices <- sales[, .(date = min(as.Date(InvoiceDate)),
value = sum(Quantity * UnitPrice)),
by = .(invoice = InvoiceNo, customer = CustomerID, country = Country)]
db_insert(invoices, 'invoices', 'ecommerce')
Check the structure of the newly (and automatically) created table using the command-line SQLite client:
.schema invoices
Check the date column after reading back from the database:
invoices <- db_query('SELECT * FROM invoices', 'ecommerce')
str(invoices)
invoices[, date := as.Date(date, origin = '1970-01-01')]
Report the daily revenue in Excel
revenue <- invoices[, .(revenue = sum(value)), by = date]
library(openxlsx)
wb <- createWorkbook()
sheet <- 'Revenue'
addWorksheet(wb, sheet)
writeData(wb, sheet, revenue)
## open for quick check
openXL(wb)
## write to a file to be sent in an e-mail, uploaded to Slack or as a Google Spreasheet etc
filename <- tempfile(fileext = '.xlsx')
saveWorkbook(wb, filename)
unlink(filename)
## static file name
filename <- 'report.xlsx'
saveWorkbook(wb, filename)
Tweak that spreadsheet:
freezePane(wb, sheet, firstRow = TRUE)
setColWidths(wb, sheet, 1:ncol(revenue), 'auto')
poundStyle <- createStyle(numFmt = '£0,000.00')
addStyle(wb, sheet = sheet, poundStyle,
gridExpand = TRUE, cols = 2, rows = (1:nrow(revenue)) + 1, stack = TRUE)
greenStyle <- createStyle(fontColour = "#00FF00") # previously? fgFill = "#00FF00"
conditionalFormatting(wb, sheet, cols = 2,
rows = 2:(nrow(revenue) + 1),
rule = '$B2>66788.35', style = greenStyle)
standardStyle <- createStyle()
conditionalFormatting(wb, sheet, cols = 2,
rows = 2:(nrow(revenue) + 1),
rule = '$B2<=66788.35', style = standardStyle)
Add a plot:
addWorksheet(wb, 'Plot')
library(ggplot2)
library(ggthemes)
ggplot(revenue, aes(date, revenue)) + geom_line() + theme_excel()
insertPlot(wb, 'Plot')
saveWorkbook(wb, filename)
saveWorkbook(wb, filename, overwrite = TRUE)
Report the monthly revenue and daily breakdowns in Excel
library(lubridate)
monthly <- invoices[, .(value = sum(value)), by = .(month = floor_date(date, 'month'))]
library(openxlsx)
wb <- createWorkbook()
sheet <- 'Summary'
addWorksheet(wb, sheet)
writeData(wb, sheet, monthly)
for (month in as.character(monthly$month)) {
revenue <- invoices[floor_date(date, 'month') == month,
.(revenue = sum(value)), by = date]
addWorksheet(wb, as.character(month))
writeData(wb, month, revenue)
}
saveWorkbook(wb, 'monthly-report.xlsx')
Report on the top 10 customers in a Google Spreadsheet
top10 <- sales[!is.na(CustomerID),
.(revenue = sum(UnitPrice * Quantity)), by = CustomerID][order(-revenue)][1:10]
library(openxlsx)
wb <- createWorkbook()
sheet <- 'Top Customers'
addWorksheet(wb, sheet)
writeData(wb, sheet, top10)
t <- tempfile(fileext = '.xlsx')
saveWorkbook(wb, t)
## upload file
library(googledrive)
drive_auth()
## NOTE you can clean up credentials in ~/.R/gargle/gargle-oauth
drive_upload(media = t, name = 'top customers', path = 'ceu')
drive_update(media = t, file = 'top customers')
## instead of top10, let's do top25 ... so appending a few rows to an already existing spreadsheet
top25 <- sales[
!is.na(CustomerID),
.(revenue = sum(UnitPrice * Quantity)), by = CustomerID][order(-revenue)][1:25]
library(googlesheets4)
gs4_auth()
for (i in 11:25) {
sheet_append('your.spreadsheet.id', data = top25[i])
}
Homeworks
Week 1
Create the mr
R package described above with the forint
and get_bitcoin_price
functions, and push to a new repo in your GitHub account, so that you can install the package on any computer via remotes::install_github
. Submit the URL to your GitHub repo in Moodle.
Week 2
Create a new git branch in your (above created) git repo for the mr
package, and add the newly created get_usdhufs
function there. Then create a more generalized version of the function called get_exchange_rates
that queries historical exchange rates for any currency pair (so configurable symbol
and base
currency) for the provided time interval. Example run:
> get_exchange_rates('EUR', 'USD', start_date = '2020-05-12', end_date = '2020-05-13')
date rate
1: 2020-05-12 1.0813
2: 2020-05-13 1.0847
Don't forget about documenting the function!
Then push your changes (either in one or multiple commits) as a new branch to GitHub and create a pull request to merge to your master
branch. Share the URL to your pull request on Moodle!
Preparations for Week 3
Let's plan to use RStudio Desktop on your laptop next week due to working with Excel and some sensitive data (Google Drive credentials) that is better to handle outside of a shared environment ... so please make sure that the below R packages are installed and working in your local R environment:
data.table
lubridate
RSQLite
openxlsx
googlesheets
logger
profvis
microbenchmark
daroczig/binancer
(install from GitHub)daroczig/dbr
(install from GitHub)
And download this file (12Mb) to your computer before the next class http://bit.ly/CEU-R-ecommerce with the ecommerce.sqlite3.zip
file name.
Home Assignment
For pass:
- Merge your PR from the second week's homework!
- Read the "Testing" chapter from Hadley's "R Packages" book at http://r-pkgs.had.co.nz/tests.html
- Create a new branch in your R package, write a unit test for the
forint
function to make sure thatforint(42)
returns42 HUF
, open a pull request and share the PR URL on Moodle!
For grade:
- Set up CI to automatically run your unit tests when pushing to GitHub (see https://r-pkgs.org/r-cmd-check.html)
- Set up a webpage for your package using
pkgdown
and GitHub Pages (see https://pkgdown.r-lib.org)
Deadline: June 6, 2021 (midnight by CET)
References
- AWS Console: https://ceu.signin.aws.amazon.com/console
- Binance (cryptocurrency exchange) API: https://github.com/binance/binance-spot-api-docs/blob/master/rest-api.md (R implementation available at https://github.com/daroczig/binancer)
- Foreign exchange rates API, eg https://exchangerate.host
- "Writing R Extensions" docs: https://cran.r-project.org/doc/manuals/r-release/R-exts.html
- Hadley Wickham's "R packages" book: http://r-pkgs.had.co.nz
- Interesting thread on open-source licenses in the R community: tidyverse/ggplot2#4236
- Hadley Wickham's "Advanced R" book (1st edition): http://adv-r.had.co.nz/
- R package tests with GitHub Actions instead of Travis: https://github.com/r-lib/actions/tree/master/examples#quickstart-ci-workflow
- The
tidyverse
style guide: https://style.tidyverse.org/ pkgdown
package: https://pkgdown.r-lib.org/index.htmldbr
package: https://github.com/daroczig/dbr
Contact
File a GitHub ticket.