<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
  <channel>
    <title>TAKT notes</title>
    <link>https://taktcycles.com/notes/</link>
    <description>Technical notes from TAKT on the speed and reproducibility of numerical code. Each one shows its measurements and, where possible, a script to check them on your own machines.</description>
    <language>en</language>
    <lastBuildDate>Mon, 12 Oct 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://taktcycles.com/notes/feed.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>The official QuantLib wheel prints different last digits on different Linux systems</title>
      <link>https://taktcycles.com/notes/quantlib-glibc-last-digits.html</link>
      <guid isPermaLink="true">https://taktcycles.com/notes/quantlib-glibc-last-digits.html</guid>
      <pubDate>Mon, 12 Oct 2026 00:00:00 +0000</pubDate>
      <description>The same official QuantLib 1.43 wheel prints different last digits on glibc 2.28 than on 2.39 and 2.41, because it calls the system's exp, log and pow.</description>
    </item>
    <item>
      <title>Why LightGBM permutation importance is slow, and what you can do about it without changing the result</title>
      <link>https://taktcycles.com/notes/lightgbm-permutation-importance.html</link>
      <guid isPermaLink="true">https://taktcycles.com/notes/lightgbm-permutation-importance.html</guid>
      <pubDate>Mon, 12 Oct 2026 00:00:00 +0000</pubDate>
      <description>Permutation importance runs one full prediction per feature and repeat. Which settings make it faster with the same result, and which ones change the result.</description>
    </item>
    <item>
      <title>The same Python backtest prints a different Sharpe ratio on a newer CPU</title>
      <link>https://taktcycles.com/notes/numpy-avx512-backtest.html</link>
      <guid isPermaLink="true">https://taktcycles.com/notes/numpy-avx512-backtest.html</guid>
      <pubDate>Fri, 09 Oct 2026 00:00:00 +0000</pubDate>
      <description>On AVX-512 CPUs numpy rounds about one log or exp in twenty differently, and the widely quoted way to switch that off stopped working in numpy 2.4.</description>
    </item>
    <item>
      <title>TA-Lib performance in Python: measure the native computation first</title>
      <link>https://taktcycles.com/notes/python-talib-performance.html</link>
      <guid isPermaLink="true">https://taktcycles.com/notes/python-talib-performance.html</guid>
      <pubDate>Thu, 08 Oct 2026 00:00:00 +0000</pubDate>
      <description>Before speeding up TA-Lib in a Python backtest, measure how much of the run it takes, and read what published speed-ups actually cover.</description>
    </item>
    <item>
      <title>Measure a Rust program with interleaved runs and output checks</title>
      <link>https://taktcycles.com/notes/harness-rust.html</link>
      <guid isPermaLink="true">https://taktcycles.com/notes/harness-rust.html</guid>
      <pubDate>Thu, 08 Oct 2026 00:00:00 +0000</pubDate>
      <description>Compare two builds of a Rust program with interleaved runs, hardware counters and an output check, using the open-source takt-harness.</description>
    </item>
  </channel>
</rss>
