15th May 2001
Busy busy busy.
Tried out ScreenWriter 2000 demo - really impressed. It did just about all the things I wanted it to without really having to think too hard. Two niggles:-
(1) CTRL-HOME and CTRL-END delete lines!!!!!, not go to the top and bottom of the document (as you'd expect). This is fairly irritating, as (for short screenplays especially) you want to go to both ends quickly. Pressing them once should go to the top/bottom of the scene, and twice to the top/bottom of the screenplay. Or something.
(2) It seems that if you perform lots of actions at speed, you can crash SW2000 - this has happened a couple of times, and lost me a pagefull-plus each time. A little bit worrying.
Still, I think I really ought to pony up for this - in every other sense it's really very good. :-)
13th May 2001
Laid out in the sun again - having finished reading the book on Brunelleschi yesterday, moved on to Wired today. On balance, I think I prefer Brunelleschi's machines. :-)
Overheard from our (Arsenal-supporting) neighbour yesterday: "F*&kin' referee! Pass me the phone!". And that was when it was still nil-nil... :-/
12th May 2001
Ouch! Thought my pitas page had died, but it turned out I'd included "-->" in my page without HTMLising it, which had messed up the validity of the page, and hence turned it into a narrow vertical black strip. All sorted now!
Trying out embedding HTML into my diary entries... should help me format them a little more (than none at all)!
11th May 2001
SWT train strike scuppered meal out at the Fire Station with cousin Steve & wife Kate. Bah!
Tried loads of variations on sprite -> sprites algorithm, but settled on a really simple approach, which just happened to work out the best. Typical, eh?
To end the day, fish'n'chips, a bottle of Picpoul, & Dom Joly's Trigger Happiness. V nice! :-)
10th May 2001
Yuk - felt 'orrible all day. Finished the purple prose for the 3dpp press release, scratched my head till it bled figuring out how best to convert a sprite image into a series of AGB hardware sprites... sounds a *lot* like the Dressmakers' Problem, which is NP-incomplete (IIRC).
Put up my PWP page - http://www.nickpelling.pwp.blueyonder.co.uk/ -and some recent chess games. Still, the weather's been great and my hay fever hasn't *really* kicked in yet... so... it's time for the first barbecue of the summer! Where did those beers go...?
9th May 2001
New chair day - it had the wrong assembly instructions in (thanks Viking) but I figured it out anyway. It'll take a while for my back to settle down - thinking about a sports massage or something. It's curious - I can feel the misalignment all the way from from my right knee, through my hip, through my back, and into my shoulder & neck & jaw.
Rooting around, I found a paper I really wanted to read - Xiaoji "Jimmy" Wang, 1980, "New Results on the costs of Huffman Trees", in the Australian Journal of Combinatorics. But **1980**, that's about 10BP (before PDFs), so the poor guy probably had to employ monks to copy it up from his papyrus original. :-/ So: not a huge chance of finding it anywhere outside the British Library. :-/
8th May 2001
*sigh* Spent the day evaluating run-length compression techniques. Spent 45 minutes writing a particularly nice generalisation of the RLE-0 algorithm... before spending two hours finding the *incredibly* subtle bug I'd managed to build in.
Back's still killing me - good job my new chair is coming tomorrow. Also: wrote a new baby card for Ian & Sarah.
Heyyyy - incoming mail from tha Natestah. Gotta fly.
7th May 2001
Took a day off yesterday from everything. Since my back chair died a death, I'm using my 'plan B' cheapo Viking office chair, which is very nearly as dead... but it's *killing* my back.
Had lunch with Julie and my grandmother at "The Dove" in Dargate, Kent. Excellent as usual, though I wish they'd have more than a single bitter (Shepherd Neame) on.
On the way home, took the A20/A25 to avoid the perennial Bank Holiday Clacketts Lane snarlup on the M25. Westerham looked very nice - Julie had a great-aunt who lived there during the war. Perhaps should take another look at the town on a day when we're less hurried.
Implemented a cool idea for a compressor that didn't even *slightly* work (bah):- (1) Send the freq counts for a large buffer, (2) split the buffer, (3) predict the freq counts for each half, (4) encode the differences, (5) recurse until each sub-block contains only a run of a single number.
5th May 2001, continued...
Final thought: Deodorowicz (2000) notes that the structure of x(BWT) isn't really understood. The problem is actually that the appropriate *metaphor* isn't in place. IMO, x(BWT) has a "foam-like" structure, comprising bubbles of wildly different entropies juxtaposed. This explains why passing x(MTF) through a simple adaptive entropy coder doesn't achieve the best results, and why BW94 (which used a blocking Huffman encoder (IIRC, 16K blocks?) achieved such good results. My multi-band arithmetic encoder (MBA) squeezes better results out than ARI, but it's still not optimal for foam encoding.
That's the exact term I'm reaching for: what's needed is a "foam encoder". I should write a paper about this...
5th May 2001, morning...
Aching all over - what a lousy start to the day. I wouldn't mind but I didn't do anything yesterday... apart from overwork a tad on Crawfish's compression.
In my dream, I was talking to Charles Bloom on the phone: he asked "What do you know that I don't?". I asked him to hold for 30 secs while I got my lab-books to read out my notes, but when I got back, he was on hold & the line was breaking up. Ah well.
Thought for the morning: all the decent context-based predictors I've tried out match Charles Bloom's observation that the top-ranking prediction has approx probability (2/3). Further, experience shows that subsequent high-ranking predictions taper off in an inverse power relationship, roughly inverse square. But is this universal? Are the only coding gains beyond that to be found in back-end coding and tricksy run-encoding?
Or could *substantially* better predictors be built within this same framework?
3rd May 2001, morning
Woohoo! Telewest finally got my cable modem hooked up to the pavement, so now I have a reasonably fast connection. Which is nice. Also: booked up for the Hotel Ibis in Birmingham for the Back In Time 3 retro knees-up/party thing on May 16th. Which should also be nice.
Reminder to self: download cracked version of Bangkok Knights C64 so can answer fan-stylequestions on same with larger degree of authority than none at all. :-/
2nd May, PS
Also had fun beating Arason 5.4 (one minute per move) in 25 moves with my trusty anti-English Borg. Pity I didn't manage the same trick against Wimbledon last night. :-(
2nd May 2001
Loooong day. The 4-bit gfx compressor I wrote for Crawfish gives excellent compression, but even after all the speedups I put in place, it's *still* too slow for that king of conversions Keith B... so I've been thinking about alternatives that would have a faster run-time decompressor.
Shame - it was a nice example of (what Peter Fenwick would call) a static Shannon51 compressor. So... I've come up with a cooler version of the same, only more byte-oriented. Hope this works...
It's interesting: Fenwick's Shannon all-guess compressor ("srank.c") and Bloom's LZP (which is actually a Shannon 1-guess compressor) are at either end of the scale, and this would be an example of a Shannon n-guess compressor (although static, rather than dynamic).
1st May 2001
Smoked fish pie for lunch at the "King William IV" in Mickleham - highly recommended. An excellent pint of Ringwood Fortyniner, too. I try not to drink more than a pint at lunchtime or I lose the afternoon... though I do occasionally forget... :-)