November 2019
dplyr / purrr for efficient data manipulationbroomtibbleggplot using relevant aesthetics
interactive session
Tutorial based on the great conference by Hadley Wickham
nest()
mtcars %>%
group_nest(cyl) %>%
mutate(model = map(data, ~lm(mpg ~ wt, data = .x)),
summary = map(model, summary),
r_squared = map_dbl(summary, "r.squared"))
# A tibble: 3 x 5
cyl data model summary r_squared
<dbl> <list> <list> <list> <dbl>
1 4 <tibble [11 × 10]> <lm> <smmry.lm> 0.509
2 6 <tibble [7 × 10]> <lm> <smmry.lm> 0.465
3 8 <tibble [14 × 10]> <lm> <smmry.lm> 0.423
gapminder
05:00
gapminder packagegapminder and tidyverse packages%>% to pass gapminder to ggplot()life expectency (lifeExp in y) ~ year (x)geom_line()
library(gapminder) gapminder %>% ggplot(aes(x = year, y = lifeExp, group = country)) + geom_line()
04:00
mutate() named year1950 which is:year - 1950group_nest() the tibble by country and continentby_countrygroup_nest()by_country <- gapminder %>% mutate(year1950 = year - 1950) %>% group_nest(continent, country) by_country
# A tibble: 142 x 3 continent country data <fct> <fct> <list> 1 Africa Algeria <tibble [12 × 5]> 2 Africa Angola <tibble [12 × 5]> 3 Africa Benin <tibble [12 × 5]> 4 Africa Botswana <tibble [12 × 5]> 5 Africa Burkina Faso <tibble [12 × 5]> 6 Africa Burundi <tibble [12 × 5]> 7 Africa Cameroon <tibble [12 × 5]> 8 Africa Central African Republic <tibble [12 × 5]> 9 Africa Chad <tibble [12 × 5]> 10 Africa Comoros <tibble [12 × 5]> # … with 132 more rows
datayear1950 will help to get meaningful interceptscontinent to keep it along with countrygapminder %>% filter(country == "Germany") %>% select(-country, -continent, -pop)
# A tibble: 12 x 3
year lifeExp gdpPercap
<int> <dbl> <dbl>
1 1952 67.5 7144.
2 1957 69.1 10188.
3 1962 70.3 12902.
4 1967 70.8 14746.
5 1972 71 18016.
6 1977 72.5 20513.
7 1982 73.8 22032.
8 1987 74.8 24639.
9 1992 76.1 26505.
10 1997 77.3 27789.
11 2002 78.7 30036.
12 2007 79.4 32170.by_country %>% filter(country == "Germany")
# A tibble: 1 x 3 continent country data <fct> <fct> <list> 1 Europe Germany <tibble [12 × 5]>
by_country %>% filter(country == "Germany") %>% unnest(data) %>% select(-country, -continent, -pop)
# A tibble: 12 x 4
year lifeExp gdpPercap year1950
<int> <dbl> <dbl> <dbl>
1 1952 67.5 7144. 2
2 1957 69.1 10188. 7
3 1962 70.3 12902. 12
4 1967 70.8 14746. 17
5 1972 71 18016. 22
6 1977 72.5 20513. 27
7 1982 73.8 22032. 32
8 1987 74.8 24639. 37
9 1992 76.1 26505. 42
10 1997 77.3 27789. 47
11 2002 78.7 30036. 52
12 2007 79.4 32170. 5706:00
by_countrymodel with linear regressions of lifeExp on year1950by_country_lm
mutate or summarise?data), do you need to use map?by_country_lm <- by_country %>% mutate(model = map(data, ~ lm(lifeExp ~ year1950, data = .x))) by_country_lm
# A tibble: 142 x 4 continent country data model <fct> <fct> <list> <list> 1 Africa Algeria <tibble [12 × 5]> <lm> 2 Africa Angola <tibble [12 × 5]> <lm> 3 Africa Benin <tibble [12 × 5]> <lm> 4 Africa Botswana <tibble [12 × 5]> <lm> 5 Africa Burkina Faso <tibble [12 × 5]> <lm> 6 Africa Burundi <tibble [12 × 5]> <lm> 7 Africa Cameroon <tibble [12 × 5]> <lm> 8 Africa Central African Republic <tibble [12 × 5]> <lm> 9 Africa Chad <tibble [12 × 5]> <lm> 10 Africa Comoros <tibble [12 × 5]> <lm> # … with 132 more rows
03:00
data columnby_country_lm %>% mutate(n = map_int(data, nrow)) %>% select(country, n)
# A tibble: 142 x 2 country n <fct> <int> 1 Algeria 12 2 Angola 12 3 Benin 12 4 Botswana 12 5 Burkina Faso 12 6 Burundi 12 7 Cameroon 12 8 Central African Republic 12 9 Chad 12 10 Comoros 12 # … with 132 more rows
by_country_lm %>% mutate(n = map_int(data, nrow)) %>% distinct(n)
# A tibble: 1 x 1
n
<int>
1 1206:00
summary for the linear model of Rwandafilter() for the desired countrymap() on the model column the function summary(), name it summaryr.squared with map_dbl(summary, "r.squared")by_country_lm %>%
filter(country == "Rwanda") %>%
mutate(summary = map(model, summary),
rsq = map_dbl(summary, "r.squared"))
# A tibble: 1 x 6 continent country data model summary rsq <fct> <fct> <list> <list> <list> <dbl> 1 Africa Rwanda <tibble [12 × 5]> <lm> <smmry.lm> 0.0172
\(r^2\) is close to 0, linearity sounds broken
broom will cleanup lm elements into tibbles
06:00
by_country_lm, add 4 new columns:
glance, using the broom function on the model columntidy, using the broom function on the model columnaugment, using the broom function on the model columnrsq from the glance columnmodelsmap when dealing with a list columnmap, shortcut with quotes (like "r.squared") extract the specified variablemap takes and returns a list. Use map_dbl() to coerce output to doubleslibrary(broom)
models <- by_country_lm %>%
mutate(glance = map(model, glance),
tidy = map(model, tidy),
augment = map(model, augment),
rsq = map_dbl(glance, "r.squared"))
models
# A tibble: 142 x 8 continent country data model glance tidy augment rsq <fct> <fct> <list> <list> <list> <list> <list> <dbl> 1 Africa Algeria <tibble… <lm> <tibble… <tibbl… <tibble … 0.985 2 Africa Angola <tibble… <lm> <tibble… <tibbl… <tibble … 0.888 3 Africa Benin <tibble… <lm> <tibble… <tibbl… <tibble … 0.967 4 Africa Botswana <tibble… <lm> <tibble… <tibbl… <tibble … 0.0340 5 Africa Burkina Faso <tibble… <lm> <tibble… <tibbl… <tibble … 0.919 6 Africa Burundi <tibble… <lm> <tibble… <tibbl… <tibble … 0.766 7 Africa Cameroon <tibble… <lm> <tibble… <tibbl… <tibble … 0.680 8 Africa Central Afr… <tibble… <lm> <tibble… <tibbl… <tibble … 0.493 9 Africa Chad <tibble… <lm> <tibble… <tibbl… <tibble … 0.872 10 Africa Comoros <tibble… <lm> <tibble… <tibbl… <tibble … 0.997 # … with 132 more rows
05:00
country ~ rsqrsq): snake plotto reorder discrete values:
factorforcats packagefct_reorder() to reorder according to a continuous variablemodels %>%
ggplot(aes(x = rsq,
y = fct_reorder(country,
rsq))) +
geom_point(aes(colour = continent),
alpha = 0.5) +
theme_classic(18) +
theme(axis.text.y = element_blank(),
axis.ticks.y = element_blank(),
legend.position = c(0.25, 0.75)) +
guides(color = guide_legend(
override.aes = list(alpha = 1))) +
labs(x = "r square",
y = "Country") 
04:00
unnest column datalifeExp ~ year with linesarrange(col) will sort according to coltop_n(x, col) not only sort col but return only x top entriestop_n(x, desc(col)) same but sort from lowest valuesmodels %>%
top_n(20, desc(rsq)) %>%
unnest(data) %>%
ggplot(aes(x = year, y = lifeExp)) +
geom_line(aes(colour = continent)) +
facet_wrap(~ country) +
theme(axis.text.x = element_text(angle = 45,
hjust = 1),
legend.position = "bottom")
models %>%
top_n(20, rsq) %>%
unnest(data) %>%
ggplot(aes(x = year, y = lifeExp)) +
geom_line(aes(colour = continent)) +
facet_wrap(~ country) +
theme(axis.text.x = element_text(angle = 45,
hjust = 1),
legend.position = "bottom")
year1950?year?
filter(models, country == "Germany") %>% unnest(tidy) %>% select(term, estimate, statistic)
# A tibble: 2 x 3 term estimate statistic <chr> <dbl> <dbl> 1 (Intercept) 67.1 282. 2 year1950 0.214 30.7
# A tibble: 2 x 3 term estimate statistic <chr> <dbl> <dbl> 1 (Intercept) -350. -25.4 2 year 0.214 30.7
04:00
tidy column)
continent, country and rsq columns(Intercept) name which needs to be called between backsticks ‘`’)scale_size_area() for lisibility)geom_smooth(method = "loess")models %>% unnest(tidy) %>% select(continent, country, rsq, term, estimate) %>% pivot_wider(names_from = term, values_from = estimate) %>% ggplot(aes(x = `(Intercept)`, y = year1950)) + geom_point(aes(colour = continent, size = rsq)) + geom_smooth(se = FALSE, method = "loess") + scale_size_area() + labs(x = "Life expectancy (1950)", y = "Yearly improvement")
library(gganimate)
gapminder %>%
ggplot(aes(x = gdpPercap,
y = lifeExp,
size = pop,
color = continent)) +
transition_time(year) +
ease_aes("linear") +
scale_size(range = c(2, 12)) +
geom_point() +
theme_bw(16) +
labs(title = "Year: {frame_time}",
x = "GDP per capita",
y = "life expectancy") +
scale_x_log10() -> p
animate(p)
anim_save("gapminder2.gif")
