Comments on: datestr performance

By: Yair Altman

Yair Altman — Sat, 29 Oct 2011 15:52:21 +0000

In reply to EBS. @Eric - Thanks for the new ideas

By: EBS

EBS — Fri, 28 Oct 2011 18:22:06 +0000

Yair, good tip – what about separating out the cache creation into its own function, and putting that call into startup.m? Then you’d move the first-run time hit to startup, which is innocuous… You could next try making the cache creation one-time by saving it into an mat-file stored in the prefdir(?), using similar logic to check if it exists at the entry to the cache_creation function. This may not be any faster though 🙂

One power-user tip that used to work is to move any often-called/performance-critical ‘user library’ functions into the MATLAB library directory, and then update the function cache. This used to decrease the time required by ML to lookup the function, but recent improvements may make this untrue in recent versions? A related but little-known tip is that ‘clear all’ also deletes the internal function cache and causes it to be rebuilt, which can reduce performance esp. for user functions until it has been built over time as funcs are called.

Cheers,
EBS

By: marco

marco — Sat, 15 Oct 2011 17:05:59 +0000

Really cool idea. I normally cringe when I see persistent variables, but this seems like a great use of them. Interesting indentation scheme you have there – kind of c-like to have your function guts indented like that.

By: MJJ

MJJ — Fri, 07 Oct 2011 09:11:36 +0000

Also, logical indexing is fast, both to write and to process, but if you are only after a select few values within a large/huge array find may be faster (and less memory hungry) especially with the 'k' parameter.

By: Rory

Rory — Thu, 06 Oct 2011 21:23:33 +0000

I love this blog. This is fantastic.