Finding the consensus classification using anonymous functions (CC284)

[@caesar6128]

I hope this does not come as pedantic. A small suggestion is to replace "1:length(xxx)" with seq_along(xxx) or seq_len(length(xxx)) in loops and lapply functions. This is recommended mainly for edge cases where the input is an empty object, i.e, length is 0: > for (i in 1:0) print(i) [1] 1 [1] 0 > for (i in seq_len(0)) print(i) #nothing is printed > for (i in seq_along(NULL)) print(i) #nothing is printed > for (i in seq_along(c())) print(i) #nothing is printed > for (i in seq_along(list())) print(i) #nothing is printed

[@Riffomonas]

Thanks - yeah I know I'm bad at this :) Hope to address in a future episode

[@user-sb9oc3bm7u]

Hey Pat. as always, great vid. I am wondering, why using apply if we already saw in previous vids that a for loop might be faster? I tested some dummy code and found that for loop is indeed faster, and with the added benefit of readability (i think its a bit more clear than map/apply), would not hesitate to using it. # my code: library(stringi) library(purrr) library(microbenchmark) set.seed(123) myipsum

[@user-sb9oc3bm7u]

Another thought? instead of removing, why not leaving the confidence and let user decide what to do with it? Other though (which you probably implemented already) is to allow a "threshold" attribute where the user can adjust the threshold-confidence for each use-case.

[@Riffomonas]

Thanks! I may come back and refactor this part of the code, but it's pretty low intensity data processing compared to the other steps. It was easier to code in this instance. Speed isn't always the most important factor

[@Riffomonas]

I think I did that, right? The output of classifying is the plurality consensus. Only on filtering and printing would the low confidence scores get dropped