OK, thanks for the extensive beta-testing of the ka72.com site.

I have now facelifted the site, and re-published it. Go to www.ka72.com/ to take a look. I have removed all previously uploaded files from the site, and replaced them with four of Kellie's (Astrid to her fans, and yes the domain name is her sail number) files for others to see how the View function works.

The version now live allows you to upload and analyse a file, and also to run analysis on other uploaded files. It does not allow you to download any other files, but it does show you the file names of the other files.

I am confident enough to recommend to people that they could use the site to analyse their data files with enough certainty to post the results in the team challenge, with the proviso that you should probably mention in the GPSTC comments that the results come from a Beta5 release of the site engine.

The analysis engine is unchanged from Beta5. My algorithm for detecting "runs" in the 5x10 remains different from that used in other packages. It is good, but not identical, to the others. Basically, I define a run end solely in terms of average angle of the run, whereas RealSpeed also takes into account speed drops. So if you slow down (under 7 kts) and speed up again, Realspeed treats that as two runs, whereas I don't. There are probably other subtle differences. Stay tuned for updates.

All my other caveats remain, if you want to compare the ka72 results with Realspeed, make sure you are using doppler settings and using the default.rsp for the GPS Team Challenge.

Thanks for the emphatic caveat, Ben.

Once you get me those 20 files, I will start fine-tuning.

I would also emphasise that the present B5 version has no filtering apart from the 200 second slow point, and that, as with any other software, all results are subject to verification.

Hopefully B6 will be online by the end of the weekend, with improved filtering and slightly tighter calcs.

I am very hopeful that I will be able to do a better job of detecting and filtering out anomolous alphas. There is no filtering at all at present.

I plan to opensource the engine code once it's out of beta, but if anyone would like to have a look at it in the meantime (it's C#.NET 2008) please drop me a line via the site.

Wow, that's dedication!

Excellent work Dylan, I bet that the Red Bull share price just bounced up a couple of points

I'll put together some test files and send them over tonight (got to go to work now).

sausage said...

Dylan,
Fabulous work - just uploaded file and everything matches with Realspeed results bar the Alpha 500. Realspeed shows 21.248kn (trackpoint) and KA72 SR shows 20.82kn. No real problem my end just thought you may like to know.

The problem is that my calculator only uses doppler for Alphas. I *know* that this is not ideal for GT-11 users, but I'm hoping that the main users of the site will be GT-31 users, since it's really targetting new speed sailors, not existing ones, and the GT-11 is no longer available.

One thing is that the doppler speed is generally slightly lower than the trackpoints one, so that is at least an incentive for users to move on and buy Realspeed

Also, you may have noticed the "Verbose" checkbox on the site. I will be extending the verbose data somewhat in the next few days as all it shows now is the top 10 sec speed. I will also start storing the calculated results on the site so I can compare different versions of the calculation engine over all the submitted data.

Thanks, yes that is the idea. Kellie's got a team mate that is having trouble processing gpx files, so I'm working on that next. Just got to find some spare time!!!

hardpole said...

Just ran one of my files through it - only figures that varied from Realspeed was the 10 sec realspeed gave 26.87 and ka72 25.46 and Im not sure about the distance - my realspeed one was 39.8 Km ka72 19.16 ?

I often have to remove rubbish from my SBN files in realspeed, especially when ocean sailing. The file I posted was from Hardies so should have been reasonable. I sometimes get a few points at the beginning or end that are off in space somewhere - I just delete them until it all seems to be in the right locality.

Dylan checked my the numbers and said they looked good.

My apologies

As soon as I went to check my calcs I realised that on that day (at Hardies) I had two sessions and had manually added the distances together and also the 10 sec averages were manually calculated in excel although I suspect the second session had all the good runs. I took the numbers from my posting rather than recalculating them.

There wasn’t enough wind early on and then I had a break while waiting and turned the GPS of, then the second session the wind picked up enough I could get going. So the 2 sec peek was the 2nd run and the 10 sec must have included a good run from the first session.