Omit.

Many thanks for this lovely example. I hadn't understood geom_bar's NA behavior with factor variables until your explanation.

Is this "buyer's psychology?" The opposite of $4.99 being cheaper than $5.00?

Nice! Your intuition seems correct. But how do you account for the large number of 1.01 carat diamonds?

print(n = Inf) will print all rows of a tibble.

see

More precisely: there's a $90 gap: the closed interval [1455,1545].

I wonder if this includes identifying outliers for possible data errors. E.g. x = 0 for 8 diamonds and z = 0 for 20. Also z = 31.8 for one. y = 0 for 7 diamonds and y = 31.8 and y = 58.9 for one each. I believe all are data errors.

I believe the plots can be generated without id:

diamonds %>%
  select(x, y, z) %>%
  gather(variable, value) %>%
  ggplot(aes(x = value)) +
  geom_density() +
  geom_rug() +
  facet_grid(vars(variable))

I think this is a lovely use of logicals and cumsum. But I believe that it omits planes whose first flight is delayed by more than an hour. There are 234 of these:

(zero  <- flights %>%
   filter(!is.na(dep_delay)) %>%
   arrange(tailnum,month,day) %>%
   group_by(tailnum) %>%
   mutate(delay_gt1hr = dep_delay > 60) %>%
   filter(row_number() == 1,delay_gt1hr) %>%
   select(tailnum) %>%
   mutate(n = 0)
)

Then bind_rows could concatenate these to the tibble produced by the posted solution.

This is a "two-part" solution. Is there a "one-part" solution?

Since the year column contains only 2013, you could arrange chronologically without this argument.

Or, the two parts could be combined:

rank <- flights %>%
   group_by(dest) %>%
   mutate(n_carriers = n_distinct(carrier)) %>%
   filter(n_carriers > 1) %>%
   group_by(carrier) %>%
   summarize(n_dest = n_distinct(dest)) %>%
   arrange(desc(n_dest))

Here width = Inf will print all columns.

I believe that print(width = Inf) will reliably print all selected columns.

below the

observations of each

Omit.

If you wish, this could be eliminated since the only columns that survive are those in the summarise.

omit

I believe you don't need ungroup().

"and standard deviation. The following"

I believe the problem asks for "fast flights per destination". If so, this code produces that list, including ties. Sadly, it uses slice.

fast <- flights %>%
  filter(!is.na(air_time)) %>%
  group_by(dest) %>%
  mutate(z = (air_time - mean(air_time)) / sd(air_time)) %>%
  slice(which(z == min(z))) %>%
select(origin,dest,month,day,carrier,flight,air_time,z) %>%
  arrange(z)

This is a nice plot, but the image below is not it. It appears to have been cut-and-pasted from a previous plot.

Somewhat more succinctly:

standardized_flights <- flights %>%
  filter(!is.na(air_time)) %>%
  group_by(dest,origin) %>%
  mutate(air_time_standard = (air_time - mean(air_time)) / sd(air_time)) %>%
select(origin,dest,month,day,carrier,flight,air_time,air_time_standard)

Omit, since you've already filtered.

English is inherently ambiguous while only math (vectors, subscripts) and algorithms are not. But then you'd lose your reader.

Here's a slight rephrasing where rank refers to a component of, say, min_rank(x) and value a component of x.

For each set of tied values the min_rank() function assigns a rank equal to the number of values less than that tied value plus one. In contrast, the dense_rank() function assigns a rank equal to the number of distinct values less than that tied value plus one.

English is inherently ambiguous while only math (vectors, subscripts) and algorithms are not. But then you'd lose your reader.

Here's a slight rephrasing where rank refers to a component of, say, min_rank(x) and value a component of x.

For each set of tied values the min_rank() function assigns a rank equal to the number of values less than that tied value plus one. The dense_rank() function assigns a rank equal to the number of distinct values less than that tied value plus one.

the ranking functions handle

functions

Is it "rough" or "exact"?

a vector

I might omit this since you give an explanation of row_number berlow which is accurate. Also, tibble adds its own rownumbers and rownumber(x) appears to be "the rownumber after x is sorted", which I don't find confusing.

ties

This probably is intended to be an argument of mean. It's not necessary, however, since neither dep_delay nor dep_delay_lag have any NAs. Also, could you provide an interpretation of delay_diff? For example, what does it mean that JFK's is negative?

Could be eliminated since the dataset is for one specific year.

Another definition of flight could be the quad (carrier,flight,origin,dest) which, say, might fly daily. Here's a solution with this definition.

dest_delay <- flights %>%
  group_by(dest) %>%
  filter(arr_delay > 0) %>%
  summarize(dest_ad = sum(arr_delay,na.rm=T),dest_ct = n())
flight_delay <- flights %>%
  group_by(carrier,flight,origin,dest) %>%
  filter(arr_delay > 0) %>%
  summarize(flight_ad = sum(arr_delay,na.rm=T),flight_ct = n())
prop <- as_tibble(merge(dest_delay,flight_delay,by="dest")) %>%
  mutate(prop = flight_ad / dest_ad) %>%
select(carrier,flight,origin,dest,dest_ct,flight_ct,prop)

In general selecting just a few columns (often the mutate variables) makes the result clearer since "irrelevant" cols are eliminated.

This could be omitted but keeping it in redundantly may add clarity.

Omit

Since this minimum rank group has an on_time of 0.0, I believe that this tibble is all planes which have never flown an on-time flight.

==

I believe min_rank cannot be 0.

(arr_delay > 0)

Of course precedence "does the right thing" but since R will use coercion to give lgl & dbl a value, putting in the redundant parentheses prevents the not-careful reader (me, for example) from "thinking the wrong thing."

In this dataset the presence of arr_delay implies the presence of arr_time so the boolean cancelled could be eliminated. Of course, the reader hasn't verified this relation so keeping it in makes sense. The definition of cancelled as "having an arrival delay" and "not having an arrival time" seems a bit odd.

Do all functions on grouped tibbles operate only per group and not the entire tibble? E.g.

flights %>% 
  group_by(month,day) %>%
  arrange(desc(dep_delay))

Will the arrange function order both by group and by the entire tibble?

Perhaps, but this is a lovely example of the power of R.

". atan(y,x) returns"

My worry is that the reader would assume that atan(x,y) returns that angle.

Hadley defines this as

not_cancelled <- flights %>% 
  filter(!is.na(dep_delay), !is.na(arr_delay))

But in

flights %>% 
  filter(is.na(dep_time) | is.na(arr_time) | is.na(dep_delay) | is.na(arr_delay)) %>% 
select(sched_arr_time,arr_time,arr_delay,sched_dep_time,dep_time,dep_delay)

the first 6 flights have an NA only in arr_delay. Aren't these likely not to have been cancelled and simply have a missing data entry? Why isn't

not_cancelled <- flights %>%
filter(!is.na(dep_time), !is.na(arr_time))

a "better" definition?

Omit.

Another solution. Now that we know we can "trust" dep_delay we could simply

md <- flights %>%
select(carrier,flight,origin,dest,sched_dep_time,dep_time,dep_delay) %>%
arrange(min_rank(desc(dep_delay)))

There are no ties in the top 10.

If you wish, you could change this to "Eastern Daylight" to agree with its previous reference in that sentence.

I understand the point of the paragraph but this sentence is a bit unclear.

It's unclear to me what is intended here.

A nice analysis, showing the "messiness" of actual datasets.

Note that

select(flights,vars)

produces the same result.

*

December

Replace , by |.

Interestingly between microbenchmarks to 143 microseconds while >=, <= is 2.3 on my box!

Another interpretation:

colnames(flights)[colSums(is.na(flights)) > 0]

Elegant, at a 2x microbenchmark cost.

Do numerical operators execute faster than, say, %in%?

Omit

There was no typo in the 2/2/19 version of r4ds.