Son of Deep Stops *or* Waiting to be merged with the mother thread...

[dmaziuk]

There's a fundamental difference between a sound model incorrectly implemented, and a fundamentally flawed model. The former should fairly easily be fixed, and an incorrect implementation shouldn't reflect on the soundness of a model.

Uhm... in theory. Also in theory: theory and practice are the same.

Confucius say even a wrong teacher can lead one to enlightenment and even a righteous teacher can fail to do so.

 

[Storker]

Uhm... in theory. Also in theory: theory and practice are the same.

Confucius say even a wrong teacher can lead one to enlightenment and even a righteous teacher can fail to do so.

I don't understand what you're trying to say and I'm not sure if you get my point. Let me try to illustrate my point with one of my favorite analogies:

Once upon a time, the established view of the solar system was the geocentric model. In the beginning, it was a decent model: it fitted the available data, and it had a certain predictive power. As data accumulated, more and more fudges had to be added to that model to keep it consistent with the data. The heliocentric model was a much less "faulty" model (it had its serious flaws, e.g. that the planets went around the sun in circular orbits, but its main principles were more in accordance with the reality as we know it now), but at that time its predictive power wasn't much better than the fudged geocentric model. However, as that model was further refined with new data, its predictive power increased.

The geocentric model is an example of a fundamentally flawed model which, with the right parameter adjustments, could give pretty good predictions within the dataset it had been fitted. The heliocentric model was a less fundamentally flawed model and thus much better to build upon, but at its inception its predictive power wasn't much better than its predecessor's. Do you see the difference between a "fundamentally flawed" model (the sun orbits the earth) and a "fundamentally sound" model with errors in the implementation (the earth orbits the sun in a circular orbit)?

 

[UWSojourner]

Gradient Factors, as currently implemented, are a disaster. The current and extended use that you are now proposing is grossly beyond the patch work / add-on method they employ. GF simply cannot be consistent, and are no longer relative to the model they started from.

It's not hard to "fix' GF. All that's needed is to put the adjustment into the model parameters, before it calculates the plan.

I already built a test model ZHL-D.

It seems you're having trouble implementing gradient factors. Just use the flowchart below and you shouldn't experience the problems troubling you.

 

[dmaziuk]

The geocentric model is an example of a fundamentally flawed model which, with the right parameter adjustments, could give pretty good predictions within the dataset it had been fitted. The heliocentric model was a less fundamentally flawed model and thus much better to build upon, but at its inception its predictive power wasn't much better than its predecessor's. Do you see the difference between a "fundamentally flawed" model (the sun orbits the earth) and a "fundamentally sound" model with errors in the implementation (the earth orbits the sun in a circular orbit)?

Yes, and when my operational range is covered by either model equally well, I'll take the simpler one. My impression is @lowviz is asking what is the range in which model X performs adequately, and where does it stop giving "pretty good predictions". It's not which one is more fundamentally flawed or who implemented what how wrong. ICBW and all that.

Edit: all analogies suck and yours is very apt in that heliocentric model is also fundamentally flawed. Compared to black-hole-centric, big-bang-centric, membrane-centric, or whatever-they'll-think-up-next-centric one.

 

[lowviz]

You are definitely on to me. I'm working on a smoothed version of the USNAT5 suitable for *simpler* modeling. It might take a few days.

I'm doing this just for the purpose of discussion and seeing what happens...

 

[rossh]

I've coded GF in both desktop and real-time systems and VPM-B as well. Over the years you've posted misleading or completely false statements about GF. Your statement above is the latter.

The actual coding is like Storker posted. Here's an article describing GF.

Your statement that it "gets into exponential runaway conditions" implies that you have an independent standard used to make that assessment. What is it? I suspect you're just looking at VPM-B as the standard .

We all know you have fallen in love with GF Kevin. But it has warts, including the faults I have described.

If you want to keep using it to make grossly over-inflated plans, and expect them to be taken seriously, then its time to fix the mistakes in the current GF method.

Don't try to turn the attention to VPM-B. VPM-B is still the most current, the most accurate, and the most reliable planning model we have. By comparison, ZHL-C has been soundly stepped over and patched up with GF a hundred different ways.

.

 

[PfcAJ]

We all know you have fallen in love with GF Kevin. But it has warts, including the faults I have described.

If you want to keep using it to make grossly over-inflated plans, and expect them to be taken seriously, then its time to fix the mistakes in the current GF method.

Don't try to turn the attention to VPM-B. VPM-B is still the most current, the most accurate, and the most reliable planning model we have. By comparison, ZHL-C has been soundly stepped over and patched up with GF a hundred different ways.

.

What leads you to believe that there overinflated?

 

[Patoux01]

Does it matter which? We have only one implementation of GF, and all programs need to agree with it.

Well, the programs need to agree with the underlying algorithm, if they don't, they're simply not implementing ZHL-GF, as simple as that.

 

[victorzamora]

We all know you have fallen in love with GF Kevin. But it has warts, including the faults I have described.

If you want to keep using it to make grossly over-inflated plans, and expect them to be taken seriously, then its time to fix the mistakes in the current GF method.

Don't try to turn the attention to VPM-B. VPM-B is still the most current, the most accurate, and the most reliable planning model we have. By comparison, ZHL-C has been soundly stepped over and patched up with GF a hundred different ways.

Here's what I don't understand, Ross: VPM-B is based off of zero man-testing, has provably flawed theories it's based on, and causes a marked increase in VGE counts. You've said before that as dives get bigger, you have to add conservatism with VPM. You keep calling GF "over-inflated"....but I think it's incredibly fundamentally flawed to make such an objective statement. Some GF settings will produce longer deco times than some VPM-B conservatism settings. Many of the recommended GFs cause longer deco times than similarly recommended VPM ascents. Calling it "over-inflated" just because the deco it produces is longer is astonishing to me. You're arbitrarily calling one good and one "grossly over-inflated" without any data supporting you.

And then you call VPM-B "the most accurate...planning model we have." I honestly can't see how to agree with you. You have an abundance of evidence against that. There's an abundance of evidence that the fundamental principles are flawed.

 

[rossh]

but aren't exponentials also a issue with VPM? Thus why you don't agree with anything past 5 rossh?

The problem we are discussing is how GF is patched onto the end of a ZHL, and in the current method, it becomes an exponential growth error, that obviously looses contact with what its supposed to represent.


In VPM-B anything past +5 is equally garbage. That's why VPM-B stops at +5 ,and even that has gotten to the end of useful range.

There is no such thing a VPM-B +7. You cannot buy a +7, you cannot make a +7, there is no planning program that allows us to view a +7. It does not exist. And yet by magic, it turns up in someones pretty colored, devoid of dimension, biased to shallow side, comparison chart. I wonder why they cooked up a non-existent data point to compare?


.