Strange Behavior of group_by and summarize

[adambickford]

I have a straight forward chunk of code:

f.fullattendee <- read_excel("Full Attendee List.xlsx") %>%
mutate(start_time = ymd_hms(starttime),
end_time = ymd_hms(endtime))
f.uniqatt <- f.fullattendee %>%
group_by(fullname) %>%
summarize(minStart = min(start_time),
maxEnd = max(end_time)) %>%
mutate(durationMin = round((maxEnd - minStart)/60, digits=2))

When running it (using tidyverse 1.3.0 and NOT running plyr), f.uniqatt contained a single record instead of the number of unique names (474 in my data).

This behavior persisted through multiple attempts until I quit and then restarted the R-studio session. After the restart, this code worked properly.

This is less of a question and more of a package stability issue. Why wouldn't this very basic tidyverse code work without a restart?
Sorry for the rant.
AB

[StatSteph]

How many levels of fullname are there? Are you sure it is 474? It is hard for us to verify without looking at the data. Either share the data or a subset of it.

f.fullattendee  %>%
  pull(gear) %>%
  unique() %>%
  length()

[adambickford]

Thanks for the reply @StatSteph . There are 474 unique names in the dataset. Here is a redacted sample:

fullname	start_time	end_time
Name1	2020/10/28 13:25:09	2020/10/28 15:00:21
Name1	2020/10/28 08:34:08	2020/10/28 09:29:44
Name1	2020/10/28 10:48:53	2020/10/28 11:30:23
Name1	2020/10/28 10:33:56	2020/10/28 10:47:49
Name1	2020/10/28 11:34:32	2020/10/28 11:56:44
Name2	2020/10/28 13:00:06	2020/10/28 13:14:50
Name2	2020/10/28 08:24:25	2020/10/28 15:27:37
Name2	2020/10/28 08:24:52	2020/10/28 12:04:29
Name2	2020/10/28 14:43:40	2020/10/28 15:27:37
Name3	2020/10/28 08:24:35	2020/10/28 08:59:29
Name3	2020/10/28 09:45:39	2020/10/28 15:27:37

I suppose my question is really about the reliability of the package. As I was developing the code, things worked during the initial coding on a sample data set. Then it didn't work on the full data set. After I restatred my R session things worked again.

The only change I made between the initial coding and the runs where things didn't work was use a longer dataset. This is a confusing performance issue, certainly one that shouldn't exist. If the underlying code remains stable, shouldn't the result?

AB

[nirgrahamuk]

When it runs with an error the first time you try it, but then after a restart, its correct every time, then the betting money is your having done something interactive in your environment that spoiled your data the first time, which you are not doing subsequently.