Storker,
Your post on fitting observational data to some definable measure of that data is well taken. With today’s mathematical tools, anyone can play the fitting game and come out with a right proper result.
Raw datasets are gold.
“I believe in evidence. I believe in observation, measurement, and reasoning, confirmed by independent observers. I'll believe anything, no matter how wild and ridiculous, if there is evidence for it. The wilder and more ridiculous something is, however, the firmer and more solid the evidence will have to be.”
― Isaac Asimov, The Roving Mind
Raw datasets are gold.
I am trying to follow the nuances of this thread as there is much knowledge and experience sprinkled throughout these endless posts. I keep thinking back to post #997 where UWSojourner took a single point guesstimate as to where to start looking for the region where bubble models begin to have predictive problems. That is a place for me to start poking around, but I have precisely NO raw data to work with. I do have the Navy Air Tables. So with that, I can make my own assumptions by treating those tables as metadata. Someone also mentioned a particular profile as being “a big dive”. *SIGH* -First deep stops now big dives. Fuzzy and fuzzier...
I’ll take a shot at defining what is a “big dive” so I can begin to compare that to UWSojourner’s comment.
We always start with assumptions. Let’s assume that the US Navy Air Tables all provide a similar level of “threat” to the diver. In other words, let’s assume that some of them aren’t stupidly conservative while others are idiotically dangerous with respect to getting bent on any particular profile. So my metadata is assumed to be pretty much self-consistent. Second big assumption is that the stop profiles are all rooted in an (unknown to me but) common theme. Third assumption is that the stop schedules have been refined by experience over time, thus they are reasonably optimal. I'm good to go if all assumptions are fairly reasonable.
Let’s start with the simplest hypothesis: A dive’s “Bigness” with respect to a bubble model is well-defined by just the total decompression time given in the tables. Problem. That doesn’t work as well as one would expect. 70’ for 70 min gives you 14 min of deco. 160’ for 20 min gives you 14 min of deco. One of those two dives seems (to me) to be “bigger” than the other with respect to a bubble model.
I previously showed how one can accurately predict NDL’s for the tables in question using a constant divided by depth squared. Let’s assume that this is a form factor that more or less applies to deco schedules too. We can easily run this in reverse to get a score for each profile.
Square each stop depth in any given profile, divide by the constant and multiply by the time spent at that stop. Add up all the values that you get for all of the stops in the same profile and this is your single numeric score for how “big” that particular dive is. Note that the number is totally meaningless for anything but comparison to another profile’s score. Such is the nature of metadata, much is lost but we can still compare. By comparison of scores, the deeper one of the two dives mentioned above is now “bigger”.
This is what fitting those scores to a surface looks like in the region around UWSojourner’s suggestion:
Now for the obvious question, does “bigness” reliably predict the degree of difficulties that VPM-B has with a given profile?
I don’t expect much of an answer. This is more of a thought experiment for those who may wish to make adjustments to their model that is based solely on metadata.
Define and conquer the problem…
“If we knew what we were doing, we couldn’t call it research.” -apocryphal
Never be afraid to be wrong. It is a researcher’s way of life, being right once in a while is what makes it all ever so worthwhile.
