RStudio anounced pins 1.0.0. I like the changes, but it’s taking a bit of getting used to.
Find data and put it somewhere sensible is a good first step in a project. In the old API it was difficult to distinguish between “here is a dataset I have that will never change, put it somewhere sensible” and “I would like the latest data from this server, but I’d like to cache it so I don’t hit the server every time I render this report”.
For example, I’ve handwaved “2020 is a weird year for Powerlifting so I’m excluding it” in a few posts so I don’t have to update my OpenPowerlifting pin.
If this was as simple as “I pointed board_url at the file” then I wouldn’t be posting. The file is a zip. {readr} can read a csv inside a zip but I can only find it working when the zip consists of only a csv.
So I hacked together a simple script that pointed board_url at the zip, extracted the csv and pinned it.
I don’t want the same snippet at the start of any OPL analysis, so I took a few additional steps to make it a package - openpoweRlifting. 1
It’s a minimal product at this point, you can throw a PR up for that sweet Hacktoberfest swag if you want.
RStudio Cloud
RStudio server on my home box is out of date, and I’ve given that box a little bit too much work. I’m testing RStudio Cloud. So far my only problem is that I want certain settings by default in a new workspace. It’s probably here somewhere.
As this blog works via GitHub I only needed to clone the project to work over here.
Similarly, I have a fresh environment to test my new pinboard!
Pinning the OPL
library("pins")
remotes::install_github("jimr1603/openpoweRlifting")
library("openpoweRlifting")
board = board_folder("~/pins/")
#pin_opl(board)
opl = pin_read(board, "opl-ipf") %>%
filter(equipment == "Raw") %>% #free up a little RAM
select(sex, age_class, bodyweight_kg, weight_class_kg, best3squat_kg,
best3bench_kg, best3deadlift_kg) %>% #free up more RAM
filter((sex=="M" & weight_class_kg %in% c('59', '66', '74', '83', '93', '105', '120', '120+')) |
(sex=="F" & weight_class_kg %in% c('47', '52', '57', '63', '69', '76', '84', '84+'))) # filter for current wt classes. There’s two expensive computations there - downloading the zip and extracting the zip. {pins} ensures that the zip doesn’t get downloaded again unnecessarily. I may or may not get around to fixing the other one. (Looks at so many unfinished projects on GH.)
Analysis
A recent episode of Iron Culture Podcast floated the suggestion that some weight categories suit different lifts more. So I’m going to do a simple analysis on max lifts by weight class & sex.
opl %>%
mutate(id = row_number()) %>% #cheaper on RAM than name strings
pivot_longer(c(best3squat_kg, best3bench_kg, best3deadlift_kg)) %>%
filter(!is.na(value)) %>%
group_by(name, weight_class_kg, sex) %>%
summarise(max_lift = max(value), bodyweight = mean(bodyweight_kg, na.rm=T)) %>% #Hack to make the graph look right
ungroup() %>%
rename(lift=name) %>%
ggplot(aes(x=bodyweight, y=max_lift, colour=lift)) + geom_point() + facet_wrap("sex") + ggthemes::scale_colour_few()
While the general trend is increasing for all 3 lifts X 2 sexes 2, there are a few cases where the lighter weight class is doing better than the next class up, on absolute numbers.
Ideas Pile
Somewhere in-and-among all this I found a couple of bugs in the package. (There’s 1 function, and I’ve fixed 2 bugs already…) and updated this post.
The graph of absolute top lifts by weight category is interesting, but it’d be nice to come back to this and look at Allometric Scalling. The free tier of RS Cloud is limited to 1GB of RAM. This was not a good dataset to work with. I’m going to create an issue where someone can pin the dataset as a SQLite Database to make analysis easier on the free machine.
Look, there’s two hard problems in CompSci - naming things and cache invalidation. And I’ve delegated the 2nd problem to {pins}.↩︎
A very small number of lifters are recorded as Mx. The 2021 Rulebook only has weight classes for M/F and otherwise only references Men and Women. Today the easiest thing to do is filter for sex %in% c(M, F), but digging further into how this data happened is on the ideas pile.↩︎