DivePro.me online Decompression Planner

[stuartv]

The point is to get the safest profile using machine learning. The idea of this method is to analyze most of the popular software products with same parameters (GF, Last Stop, Mix Type etc.) and pick the safest profile. By analyzing and comparing data from different software products in order to select the safest plan we minimize the probability of an error in a particular software. In the process of analyzing DiveProMe finds errors in profiles (zero ascent time, zero stop time, zero time to surface, value converting errors, zero deco stop time, etc.) and ignores these profiles.

There is still a chance that all the software products might have the same error in the same plan which will result in all the data being wrong. But the probability of suchlike result is severely diminished by comparing data from similar plans. If DiveProMe indicates a certain “leap” of errors, it will use calculations from similar plans and get the average values using the safest method.

User part of DiveProMe contains about 2,500 strings of code. It is not that many and only shows data calculated previously using machine learning. Which makes the probability of an error in the code and your plan even less.

Surely, the plan DiveProMe will give you won't be the most effective, but it will definitely be several hundred times safer than a plan calculated with a separate software.

Okay, this time without being flippant....

How does the machine learning process decide what profile is safest?

I don't see how this can be done without comparing a large number of actual dive results (i.e. did the diver get bent?) against the planned profiles. E.g. one source gives 2 minutes at 50' and 1 minute at 40'. Another source gives 1 minute at 50' and 2 minutes at 40'. How does the machine learning determine which one is "safer"?

And, when the software yields the "safest" profile, is it just picking a profile that was generated by one of the specific sources it consulted? Or is it merging the profiles that were produced by various sources into one "safest" profile? E.g. one source has 2 minutes at 50' and 1 minute at 40'. Another source has 1 minute at 50' and 2 minutes at 40'. So, the machine learning outputs a profile that has 2 minutes at 50' and 2 minutes at 40'?

If the output profile is simply verbatim one of the ones that came from one of the source applications, then it seems that the result is EXACTLY AS SAFE as the plan calculated by "a separate software".

 

[stuartv]

@stuartv , that's not what you said. In this case, the experts are apparently deco algorithms. Maybe wait for confirmation on that before making claims?

And does the software tell you, the user, which actual implementation and which algorithm determined the result it gives you? If not the the committee members are "anonymous" aren't they?

I haven't gone and looked at the actual software because I don't generally trust links to websites that I know nothing about. Too many weird viruses running around the webz these days. I got a security notice the other day about a new one that infect you just by hovering over a link or something like that.

 

[tursiops]

foul on defensive side during play, logical fallacy, Argument to moderation / fallacy of the golden mean, 10 yard penalty, restart the clock.

You said this like you thought I believed it. Ha. Far from it.

By the way, all of you who think a knowledgable committee usually arrives at a better answer than one of the individual members, might want to look up the Abilene Paradox.

 

[northernone]

foul on defensive side during play, logical fallacy, Argument to moderation / fallacy of the golden mean, 10 yard penalty, restart the clock.

as for the OP,
1) online only dive plans are worthless when you're not online. y'know, like on a boat?
2) no way to save / link to a plan
3) there's a very large amount of 'black boxing' going on here. 'mixing' various decompression algorithms does not result in a 'better' plan.

A little defensive of an unknown free volunteer based service to help divers.
:
1. Surely any serious technical dive planning where gas mixes, decompression algorithms and cns limits are needing attention isn't made right on the boat? What if we didn't happen to have the right amount of a mix along?
2. Download or save is an option. I accessed the website on my cellphone and just took screen shots.

I entirely agree 3 is a great concern and as a result I won't be using the programming until I understand it.

I'm impressed to see such a well thought out project in the creative commons so hoping we'll be gracious in our feedback and thoughtful in our criticism.

Regards,
Cameron

 

[stuartv]

Just in case any of you don't know, Subsurface (in which I have no vested interest) is also free and open source and, while it is known as a dive logging application, it has a very nice dive planner module built into it.

And it runs on Windows, Mac, and Linux.

 

[northernone]

Just in case any of you don't know, Subsurface (in which I have no vested interest) is also free and open source and, while it is known as a dive logging application, it has a very nice dive planner module built into it.

And it runs on Windows, Mac, and Linux.

Appreciated! Just realised I missed the android app launch. Being 100% mobile I hadn't been able to enjoy subsurface prior. Great mention.

 

[stuartv]

Appreciated! Just realised I missed the android app launch. Being 100% mobile I hadn't been able to enjoy subsurface prior. Great mention.

I wish they had the resources to get the dive planner and the dive computer download added into the mobile app. But, it's really great, just as it is!

 

[Alexandros]

Hi,

the website has has a really nice outfit, seems promising.

talking about open source, where could we find the source code of the project? Is it publicly available?

 

[doctormike]

The interface is irrelevant - the premise of the project to me seems seriously flawed.

If you use, for example, MultiDeco to plan your dive you know that the algorithm is GF 30/70, or VPM-B +2, or whatever. You make a decision about dive planning by picking an algorithm, enter your mixes, depth and time, and come up with an ascent plan.

In this case, you have no idea what is generating your ascent profile, other than the assurance of some unknown person that it will be several hundreds of times safer than a plan generated by the conventional methods.

Here's my method for generating a dive plan that is several thousands of times safer than a plan generated by other software: Stay on the boat.

 

[DevonDiver]

What determines 'safer'? Define?