Technical writing, without the clutter
A demonstration site for the DataLeaf Hugo theme.
Articles on mathematics, code, and research methods, set for long reading. Each is also a working example: equations render at build time, code keeps its line numbers, and statements read as they would in a paper. The documentation shows how to publish your own.
Posts
The normal equations, QR, and the SVD solve the same problem with very different accuracy. The difference is the square of a condition number.
Polynomial fits make the difference visible: with twelve columns, the normal equations lose every digit while QR and the SVD keep eight or nine. One identity about condition numbers explains it, and also says when the fast method is good enough.
What it took to reproduce an old bootstrap analysis, and why its confidence interval moved in the third decimal place.
Recursive summation in single precision loses four digits over ten million terms. Pairwise and compensated summation keep them for little extra work.