zx8754
LightNews LightNews
banner
zx8754.bsky.social
zx8754
@zx8754.bsky.social
120 followers 950 following 140 posts
Analysis - Data - Analysis - Data - Volleyball - Father - #rstats #STEM #ProstateCancer
Posts Replies Media Videos
zx8754.bsky.social
#rstats plot of the day - "Listogram" by r2evans

stackoverflow.com/a/79808261/6...
listogram plot - Histogram of three-letter words, with the words themselves shown on the plot.
November 4, 2025 at 7:34 AM
zx8754.bsky.social
"my default assumption is nothing is going to work" - @clauswilke.com
quote from the linked post: "my default assumption is nothing is going to work" https://blog.genesmindsmachines.com/p/we-still-cant-predict-much-of-anything
October 23, 2025 at 2:04 PM
zx8754.bsky.social
#rstats easier to compare the ratios
Barplots of population change ratios in "Most populous countries in Europe" 2000 vs 2025. R code for the plot: library(data.table) library(ggplot2) library(scales) d <- fread(" Rank,Country,y2000,y2025 1,Russia,146000000,144000000 2,Germany,82800000,84080000 3,Britain,59510000,69550000 4,France,59330000,66660000 5,Italy,57640000,59150000 6,Ukraine,49430000,38980000 7,Spain,40000000,49320000 8,Poland,38650000,38140000") d[, rate := y2025/y2000 ] d[, Country := factor(Country, levels = d[ order(rate), Country]) ] d[, lab := paste0(round(rate, 4), "\n", label_number(accuracy = 0.1, scale_cut = cut_short_scale())(y2025 - y2000)) ] d[, col := as.factor(sign(y2025-y2000)) ] ggplot(d, aes(Country, rate, label = lab, colour = col)) + geom_col(fill = "white", show.legend = FALSE, linewidth = 1.1) + geom_hline(yintercept = 1, colour = "blue", linetype = "dashed") + geom_text(vjust = 1.4, colour = "black") + scale_x_discrete(name = NULL) + scale_y_continuous(name = NULL, breaks = c(0, 1), limits = c(0, 1.3), expand = c(0, 0)) + scale_color_manual(values = c("-1"="red", "1" = "darkgreen")) + labs(title = "Most populous countries in Europe", subtitle = "2000 vs 2025", caption = "Source: @GEOMAPAS.GR | https://bsky.app/profile/simongerman600.bsky.social/post/3m2nz34kzyi27") + theme_minimal() + theme(panel.grid = element_blank()) ggsave("tmp.plot.jpeg", width = 8, height = 6)
October 8, 2025 at 10:15 AM
zx8754.bsky.social
It had been gradually declining by about 5 million before the war, then dropped by another 5 million between 2022 and 2025.
October 8, 2025 at 9:22 AM
zx8754.bsky.social
Blame these guys 🤪

blog.r-project.org/2020/02/16/s...
"Hence, in the R Core meetings in Toulouse in 2019, it was decided to move towards using stringsAsFactors = FALSE by default, ideally starting with the 4.0.0 release."
September 5, 2025 at 7:58 AM
zx8754.bsky.social
Ah, thought it was a hidden joke from devs, but it isn't

unix.stackexchange.com/a/153662/16560
Screenshot of the answer at: https://unix.stackexchange.com/a/153662/16560
September 5, 2025 at 7:50 AM
zx8754.bsky.social
I use Google as calculator, too. But recently discovered that the iPhone Notes app does the same:
September 4, 2025 at 8:08 PM
zx8754.bsky.social
#rstats #TIL I knew about tibble::tribble, but just discovered data.table::rowwiseDT.

"creates a data.table object by specifying a row-by-row layout. This is convenient and highly readable for small tables."
Create a data.table row-wise Aliases: rowwiseDT Keywords: ### ** Examples rowwiseDT( A=,B=, C=, 1, "a",2:3, 2, "b",list(5) ) A B C <num> <char> <list> 1: 1 a 2,3 2: 2 b 5 You can define a tibble row-by-row with tribble(): tribble( ~x, ~y, ~z, "a", 2, 3.6, "b", 1, 8.5 ) #> # A tibble: 2 × 3 #> x y z #> <chr> <dbl> <dbl> #> 1 a 2 3.6 #> 2 b 1 8.5
September 2, 2025 at 6:49 AM
zx8754.bsky.social
update: Searching GitHub R repos gives 1.5K hits. Looking at couple of pages, it looks like it is from ggfun - cran.r-project.org/web/packages...
September 1, 2025 at 1:30 PM
zx8754.bsky.social
# for loop f <- function(x, y){ out <- vector("integer", length(x)) yy <- y for(i in seq_along(x)) { mm <- match(x[ i ], yy) out[ i ] <- mm yy[ mm ] <- NA } out } # recursive version fr <- function(x, y, out = integer(length(x)), i = 1) { if (i > length(x)) return(out) mm <- match(x[ i ], y) out[ i ] <- mm if (!is.na(mm)) y[ mm ] <- NA fr(x, y, out, i + 1) } f(c("a", "b", "a"), c("a", "b", "c", "a")) # [1] 1 2 4 fr(c("a", "b", "a"), c("a", "b", "c", "a")) # [1] 1 2 4
July 24, 2025 at 9:01 AM
zx8754.bsky.social
I think they are called Earthy colours
paletteer_c("grDevices::Earth", 30) https://r-charts.com/color-palettes/
July 24, 2025 at 7:34 AM
zx8754.bsky.social
Image source: https://blogs.sas.com/content/sastraining/2019/01/24/re-creating-a-hans-rosling-graph-animation-with-sas/ Hans Rosling bubble chart presentation TED Talks: The best stats you've ever seen | Hans Rosling https://www.youtube.com/watch?v=hVimVzgtD6w
July 22, 2025 at 8:50 AM
zx8754.bsky.social
Similarly, long long time ago, I realised mutate can be called just once 🤦‍♂️
library(dplyr) df1 |> mutate(mycol1 = "too") |> mutate(mycol2 = "much") |> mutate(mycol3 = "typing") df1 |> mutate(mycol1 = "this", mycol2 = "is", mycol3 = "less")
July 21, 2025 at 7:43 AM
zx8754.bsky.social
As base::rbind also takes more than two dataframes as input, I expected at least the same from dplyr::bind_rows. And bind_rows handles rownames, factor levels, and mismatched column names better. #rstats
#list of dfs set.seed(1); listOfDfs <- lapply(1:3, \(i){ mtcars[ sample(1:nrow(mtcars), 4), ]}) #input as list do.call(rbind, listOfDfs) bind_rows(listOfDfs) #input as dataframe rbind(listOfDfs[[ 1 ]], listOfDfs[[ 2 ]], listOfDfs[[ 3 ]]) bind_rows(listOfDfs[[ 1 ]], listOfDfs[[ 2 ]], listOfDfs[[ 3 ]])
July 21, 2025 at 7:37 AM
zx8754.bsky.social
> usedRegexN == hadToReLearnRegexN [1] TRUE
July 18, 2025 at 8:40 AM
zx8754.bsky.social
In genetics we use Mahalanobis to assign/guess the population, e.g.:

pmc.ncbi.nlm.nih.gov/articles/PMC...
1KGP super-population assignment. The first three PCs plotted for each of the 5 1KGP super-populations with Mahalanobis distance outliers marked in black. Figure 1b from https://pmc.ncbi.nlm.nih.gov/articles/PMC5646819/
July 15, 2025 at 12:55 PM
zx8754.bsky.social
July 15, 2025 at 7:23 AM
zx8754.bsky.social
Rowwise is heavy, 50K rows 500 cols takes about 9 seconds, see alt text for codes and benchmark. Idea is to replace NAs with -6 in rows where NAs are continuous till the end.
#input x <- fread("c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 1 2 3 4 5 6 7 8 9 10 1 2 3 NA NA 6 7 NA NA 10 1 2 3 4 5 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA 1 2 3 NA NA 6 7 8 NA NA ") x[, { i <- as.numeric(.SD) if (is.na(i[ length(i) ])) { ix <- which(is.na(nafill(i, "nocb"))) i[ix] <- -6 } as.list(i) }, by = .I ][, I := NULL][] # V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 # <num> <num> <num> <num> <num> <num> <num> <num> <num> <num> # 1: 1 2 3 4 5 6 7 8 9 10 # 2: 1 2 3 NA NA 6 7 NA NA 10 # 3: 1 2 3 4 5 -6 -6 -6 -6 -6 # 4: -6 -6 -6 -6 -6 -6 -6 -6 -6 -6 # 5: 1 2 3 NA NA 6 7 8 -6 -6 xbig <- data.table(matrix(sample(c(1:10, NA), 50000 * 500, replace = TRUE), ncol = 500)) #dim(xbig) #[1] 50000 500 bench::mark( zx8754 = { xbig[, { i <- as.numeric(.SD) if (is.na(i[ length(i) ])) { ix <- which(is.na(nafill(i, "nocb"))) i[ix] <- -6 } as.list(i) }, by = .I ][, I := NULL][] } ) # A tibble: 1 × 13 #expression min median `itr/sec` mem_alloc `gc/sec` n_itr n_gc total_time #<bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl> <int> <dbl> <bch:tm> #1 zx8754 9.93s 9.93s 0.101 701MB 1.41 1 14 9.93s
July 9, 2025 at 2:30 PM
zx8754.bsky.social
June 22, 2025 at 7:36 AM
zx8754.bsky.social
I love waffle charts, but this can be done with points.
#rstats #ggplot
set.seed(2025) d1 <- data.frame(expand.grid(1:50, 1:50)) d1$hot <- factor("no", levels = c("no", "yes")) d1[ sample(seq(nrow(d1)), 1), "hot" ] <- "yes" d1$time <- factor("before", levels = c("before", "today")) d2 <- data.frame(expand.grid(1:50, 1:50)) d2$hot <- factor("no", levels = c("no", "yes")) d2[ sample(seq(nrow(d2)), 100), "hot" ] <- "yes" d2$time <- factor("today", levels = levels(d1$time)) library(ggplot2) ggplot(rbind(d2, d1), aes(Var1, Var2, colour = hot)) + geom_point() + facet_wrap(vars(time)) + scale_colour_manual(values = c("no" = "grey90", "yes" = "red"), guide = "none") + theme_void()
June 20, 2025 at 7:13 PM
zx8754.bsky.social
What if it is really really really small 2E-28539?
#rstats #gwas
library(data.table) #https://www.ebi.ac.uk/gwas/api/search/downloads/full d <- fread("gwas_catalog_v1.0-associations_e114_r2025-06-10.tsv", check.names = TRUE) d[ which.max(PVALUE_MLOG), .(LINK, SNPS, P.VALUE, PVALUE_MLOG), ] # LINK SNPS P.VALUE PVALUE_MLOG # 1: www.ncbi.nlm.nih.gov/pubmed/33692100 rs1129038 2E-28539 28538.7
June 17, 2025 at 8:37 AM
zx8754.bsky.social
Had a quick chat with ChatGPT, and I now present with pride the definitive ranking of machine learning algorithms by gayness. Data-driven, obviously. 🌈🤖✨
A horizontal bar chart titled "How Gay Are These Algorithms? 🌈". Algorithms are ranked from least to most "gay" on a scale of 1 to 10, using rainbow-coloured bars. Linear Regression scores lowest (1), while Dimensionality Reduction and Gradient Boosting top the list (10). The tone is humorous and tongue-in-cheek, not technical. | Algorithm | Gay | Reasoning | Gayness (1–10) | | ----------------------------------------- | --- | --------------------------------------------- | -------------- | | Linear regression | No | Too straight. Literally. | 1 | | Logistic regression | Yes | Binary decisions with flair. | 7 | | Decision tree | No | Traditional values, likes rules. | 2 | | SVM algorithm | Yes | Lives for maximal separation. Iconic. | 9 | | Naive Bayes algorithm | No | Makes assumptions. Clueless ally. | 3 | | KNN algorithm | Yes | Loyal to the nearest gays. | 8 | | K-means | Yes | Loves a clique and self-discovery. | 8 | | Random forest algorithm | No | Chaotic neutral. Too much beard. | 4 | | Dimensionality reduction algorithms | Yes | Slays by simplifying and revealing the core. | 10 | | Gradient boosting & AdaBoosting algorithm | Yes | Layered, dramatic, and gets better with time. | 10 |
June 2, 2025 at 7:45 AM
zx8754.bsky.social
#rstats base ?`$`
library(dplyr) library(purrr) library(bench) set.seed(42) create_df <- function(rows, cols) { out <- replicate(cols, runif(rows, 1, 100), simplify = FALSE) out <- setNames(out, rep_len(letters, cols)) as.data.frame(out) } results <- press( rows = c(1000, 10000, 100000), cols = c(100, 1000, 10000), { d <- create_df(rows, cols) mark( d$f, d %>% .$f, d %>% pull(f), d %>% pluck("f"), min_iterations = 100 ) } ) ggplot2::autoplot(results) bench::mark output plot
May 29, 2025 at 10:58 AM
zx8754.bsky.social
In case you are wondering about #rstats

base: 1.0.0 = 1195 vs 4.5.0 = 1286
May 27, 2025 at 8:04 AM
zx8754.bsky.social
#rstats base and purrr alternatives
df <- read.table(text = "stu_id race1 race2 race3 race4 race5 race6 100 0 0 0 0 0 1 101 1 0 0 0 0 0 102 0 1 0 0 0 1", header = TRUE) f <- function(...){ ix <- which(c(...) == 1) if(length(ix) == 1) ix else 7L } #base apply(df[, -1], 1, f) # [1] 6 1 7 #to create new column #df$race <- apply(df[, -1], 1, f) #purrr pmap_int(df[, -1], f) # [1] 6 1 7 #to create new column # df <- df %>% # mutate(race = pmap_int(select(., -1), f))
May 22, 2025 at 8:42 PM