Refreshing R skills.

You use cat(mystring) no new line is added. To get printf then you need sprintf

run a bash command via R. Critically, uses a bash expansion, so not entirely trivial.

cxxfunction that's not from Rccp but from inline

using stop() with a function, and also, how to escape quotations.

Further adventures in using stop(), this time showing how stop(-1) is worse than useless.

scripts are some way of debugging R when arguments are being called in

static variables in R. Upshot is you need a closure of a function within a function with a <<- local assignment.

Printing out a beta distribution, ran into a few fundamental problems here. uses dbeta(). ./becu 5 2 is the right biased one, and 2 5 is the left-sided one.

Simple 2 genes interaction example from the DESeq2 vignette.

grDevices:color() will get the list of 600 or R colors which actually have names. it's handy until you want opacity, named colors won;t do that, you need a function for it ... here it is.

Diff expression of one gene, though more an exercise in ggplot.

Using the pheatmap package because it seems to be well-known for its annotation abilities. It does have limitations, the legend positions area bit hard-coded.

degenerate matrices.

experiments with the scale() function.

What geom_bar, it may reorder

I've often done this in C, somehow, nto for R. It's actually dead easy, but it's also easy to get mixed up, it's just a basic chunkifier.

Linear models often add an extra term (multiplied by a coeeficient), but how much does that affect variation? Well only incremental is would seem. A single effect with a higher SD, say 4, would have a much bigger effect than 4 additions of sd=1 rnorm.

there are 657 of these, rcol1.R was used to generate most of rcolp.h (which means rcolor proportional RGB vlaues givien in [0,1] value format). I then editted th

These are the seqgendiff trials. In the beginning it held promise, but I wrote G for Gerard, G for garbled. for a reason, his terminology seems weird.

• uses null for zero
• seems obsessed with thinning all the time. This may have something to do with variance reduction, but is that right? Because he's pushing it all the time
• "databased" is the idea of using existing (real) rnaseq datasets, and adding a simulated signal to it. This is the approach of seggendiff.
• thinning, actually is binomial thinning and it means subsampling from the bionmial distribution and this is actually the method of actuall applying the artificial signal. Perhaps isntead of just adding? Maybe.

there is more than one version it appears. v2 must be run via the rmarkdown package. see rmd2h.R script. dia.Rmd was the input file for it.

playing with basic R graphics and the shape package. They are good at adding to previous plots, you can use the scale of the first plot as the canvas and position relative to it.

Kind of got a bit of hadle on this, see relevant rjotting

ggrepel's label_repel will put a back ground

Finally I worked out how to embolden part of a character vaiable that servers as a text label