library(nycflights13)
library(tidyverse)
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr 1.2.0 ✔ readr 2.1.6
## ✔ forcats 1.0.1 ✔ stringr 1.6.0
## ✔ ggplot2 4.0.1 ✔ tibble 3.3.1
## ✔ lubridate 1.9.5 ✔ tidyr 1.3.2
## ✔ purrr 1.2.1
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
Question 1
q1_lga_xna_planes <- flights %>%
filter(origin == "LGA", dest == "XNA") %>%
select(tailnum) %>%
distinct() %>%
inner_join(planes, by = "tailnum") %>%
arrange(tailnum)
head(q1_lga_xna_planes)
## # A tibble: 4 × 9
## tailnum year type manufacturer model engines seats speed engine
## <chr> <int> <chr> <chr> <chr> <int> <int> <int> <chr>
## 1 N711MQ 1976 Fixed wing multi … GULFSTREAM … G115… 2 22 NA Turbo…
## 2 N713EV 2003 Fixed wing multi … BOMBARDIER … CL-6… 2 80 NA Turbo…
## 3 N737MQ 1977 Fixed wing single… CESSNA 172N 1 4 105 Recip…
## 4 N840MQ 1974 Fixed wing multi … CANADAIR LTD CF-5D 4 2 NA Turbo…
Question 2
q2_flights_with_airline <- flights %>%
left_join(airlines, by = "carrier") %>%
select(year, month, day, carrier, name, flight, origin, dest)
head(q2_flights_with_airline)
## # A tibble: 6 × 8
## year month day carrier name flight origin dest
## <int> <int> <int> <chr> <chr> <int> <chr> <chr>
## 1 2013 1 1 UA United Air Lines Inc. 1545 EWR IAH
## 2 2013 1 1 UA United Air Lines Inc. 1714 LGA IAH
## 3 2013 1 1 AA American Airlines Inc. 1141 JFK MIA
## 4 2013 1 1 B6 JetBlue Airways 725 JFK BQN
## 5 2013 1 1 DL Delta Air Lines Inc. 461 LGA ATL
## 6 2013 1 1 UA United Air Lines Inc. 1696 EWR ORD
Question 3
airports_with_flights <- flights %>%
select(faa = origin) %>%
bind_rows(flights %>% select(faa = dest)) %>%
distinct()
q3_airports_no_commercial_flights <- airports %>%
anti_join(airports_with_flights, by = "faa") %>%
select(faa, name, lat, lon)
head(q3_airports_no_commercial_flights)
## # A tibble: 6 × 4
## faa name lat lon
## <chr> <chr> <dbl> <dbl>
## 1 04G Lansdowne Airport 41.1 -80.6
## 2 06A Moton Field Municipal Airport 32.5 -85.7
## 3 06C Schaumburg Regional 42.0 -88.1
## 4 06N Randall Airport 41.4 -74.4
## 5 09J Jekyll Island Airport 31.1 -81.4
## 6 0A9 Elizabethton Municipal Airport 36.4 -82.2
Question 4
q4_airports_most_high_winds <- weather %>%
filter(!is.na(wind_speed), wind_speed > 30) %>%
count(origin, name = "high_wind_events") %>%
filter(high_wind_events == max(high_wind_events)) %>%
left_join(airports %>% select(faa, name), by = c("origin" = "faa")) %>%
select(name) %>%
distinct()
head(q4_airports_most_high_winds)
## # A tibble: 1 × 1
## name
## <chr>
## 1 John F Kennedy Intl