Skip to content

Latest commit

 

History

History
135 lines (126 loc) · 6.32 KB

r-notes.md

File metadata and controls

135 lines (126 loc) · 6.32 KB
Raw

R Notes

Notes on the powerful but often confusing statistical computing package

  • Only & and | are vectorized logical operators; attempting to use && or || on a vector will likely not result in the desired outcome.

  • The code at the bottom of page 178 in Practical Machine Learning in R didn't work. Here's what did:

    donors <- donors %>%
mutate(incomeRating = as.character(incomeRating)) %>%
mutate(incomeRating = as.factor(str_replace(incomeRating, 'NA', 'UNK')))

  • install.packages didn't work for DMwR. Here's what did:

    library(devtools)
remotes::install_github('cran/DMwR')

    (I had equal success installing ggtheme in a similar fashion.)

  • attrition is no longer in rsample. To gain access to attrition, install modeldata, then do:

    library(modeldata)
data(attrition, package = 'modeldata')

  • verboseIter = TRUE as an argument to trainControl will enable progress tracking during runs of train.

  • Tidyverse packages include glimpse() which gives a quick rundown of a tibble (or presumably a regular dataframe too).

  • I still don't fully understand, but refer to page 11 of R for Data Science, 2nd Edition (O'Reilly) for how to get single vs. multiple estimated lines for geom_smooth() being used with color or the like.

  • geom_density() with color, fill and alpha can be very nice.

  • geom_bar() with position = 'fill' can be used to show relative frequency well.

  • Consider using facet_wrap() and maybe even facet_grid().

  • "count() lets you quickly count the unique values of one or more variables: df %>% count(a, b) is roughly equivalent to df %>% group_by(a, b) %>% summarise(n = n()).".

  • Consider using .groups = 'keep' with summarize(). (I refuse to use the limey spelling.)

  • Better to use a common prefix than a common suffix in variable naming.

  • Configure RStudio never to use .RData.

  • Concerning the difference between unique() and distinct(): https://stackoverflow.com/questions/45198738/difference-between-distinct-vs-unique

  • The ColorBrewer scales have been hand-tuned to be effective with people having various types of color blindness and there are online color blindness simulators etc. to test things out.

  • Use is.na() and near() as needed.

  • (NOTE TO SELF: The following code examples until the time series stuff came from, IIRC, R for Data Science, 2nd Edition and Statistical Inference Via Data Science: A ModernDive Into R and the Tidyverse. Properly attributing them will be considered a TODO.)

  • flights |> filter(month == 1, day == 1) |> arrange(desc(is.na(dep_time)), dep_time) makes rows with dep_time equal to NA appear first.

  • if_else() and the more advanced case_when() are in dplyr and they rock.

  • n_distinct() is also quite useful.

  • pmin() and pmax() return parallel extrema among two or more vectors.

  • forcats::fct() as opposed to factor() won't silently turn values not given in levels to NAs.

  • help(package = 'nycflights13').

  • flights |> ggplot(aes(x = carrier, fill = origin)) + geom_bar(position=position_dodge(preserve = 'single')).

  • flights |> ggplot(aes(x = carrier)) + geom_bar() + facet_wrap(~origin, ncol = 1).

  • ncol = 1 is especially important for the last one because not doing so results in a very crowded x-axis.

  • , can be used instead of & with filter().

  • libR.so needs to be available for RStudio to run. This can be made available during compiliation from source. (I did so with ./configure --enable-R-shlib --enable-BLAS-shlib --enable-memory-profiling --with-blas --with-lapack.)

  • The enhanced R shell radian also requires libR.so.

  • View() works outside of RStudio but it uses an ugly Xaw interface. Oh well. Note that print() has parameters n (rows) and width (columns) and both can be supplied with Inf.

  • flights |> select(year, month, day, hour, minute, time_hour, everything()).

  • flights %>% select(starts_with("a")).

  • flights %>% select(ends_with("delay")).

  • flights %>% select(contains("time")).

  • named_dests %>% top_n(n = 10, wt = num_flights).

  • flights |>
inner_join(planes, by = c('tailnum')) |>
select(carrier, distance, seats) |>
mutate(available_seat_miles = distance * seats) |>
group_by(carrier) |>
summarize(available_seat_miles = sum(available_seat_miles)) |>
arrange(desc(available_seat_miles))

  • guat_dem |> pivot_longer(names_to = 'year', values_to = 'democracy_score', cols = -country, names_transform = list(year=as.integer)).

  • life_expectancy |> pivot_longer(names_to = 'year', values_to = 'life_expectancy', cols = -country) |> mutate(year = as.integer(year)).

  • sample_n().

  • skim() from skimr is really very good!

  • geom_jitter() can be used to prevent overplotting.

  • Consider using boundary with geom_histogram().

  • options(pillar.sigfig = n) sets precision of tibble printouts.

  • A tsibble will retain its key even after a group_by() operation but will lose its initial key when summarize() is carried out, which will then be replaced with another key reflected by the group_by() operation.

  • Following along with Forecasting: Principles and Practice, 3rd Edition and with wesanderson installed and loaded:

    vic_elec |>
mutate(Year = factor(year(Time), ordered = TRUE)) |>
ggplot(aes(x = Temperature, y = Demand, color = Year)) +
geom_point(alpha =  0.25) +
scale_color_manual(values = wes_palette(name = 'Zissou1', n = 3))

  • In tsibbledata, aus_production is in wide format. One can convert it to tidy long format like so:

    aus_production |>
  pivot_longer(!Quarter, names_to = 'Product', values_to = 'Volume')

Note that it is not necessary to supply a key to the tsibble produced thereby. An appropriate one will be created automatically!

  • The lag plot example in the forecasting book can be made properly "tidy":

     aus_production_tidy <- aus_production |>
  pivot_longer(!Quarter, names_to = 'Product', values_to = 'Volume')
recent_production_tidy <- aus_production_tidy |>
  filter(year(Quarter) >= 2000)
recent_production_tidy |>
  filter(Product == 'Beer') |>
  gg_lag(Volume, geom = 'point') |>
  labs(x = 'lag(Beer, k)')

  • Consider using geom_freqpoly(), especially with stat = 'density'.

  • Package naniar can be used to visualize missing data.

  • Use plotly_json() to examine R plotly objects in the browser.