The Ultimate "Why Orion Is A Failure" thread......

[Poop Head]

Brute forcing the problem is not the answer. 150 stops and only visiting each stop once allows for 57133839564458545904789328652610540031895535786011264182548375833179829124845398393126574488675311145377107878746854204162666250198684504466355949195922066574942592095735778929325357290444962472405416790722118445437122269675520000000000000000000000000000000000000 possible combinations (let's just call this 6*10^262 for convenience). Maybe you have a supercomputer that can do 200 trillion calculations per second (2*10^14) and assuming that it only takes one calculation to find one combination (which is obviously not the case, but let's just go with it). Literally the world's fastest supercomputer. So it takes that supercomputer about 3*10^248 seconds to do the calculation. Or about 9.5*10^240 years. One trillion years is practically nothing compared to this span of time.

Even speeding up our supercomputer by several orders of magnitude barely helps. We just can't do it with our current level of technology. For that matter, we may NEVER be able to do it in a reasonable time frame at any level of technology humans will achieve.



First, there is not necessarily bound to be a set of possibilities with identical outcomes. It's possible, but not guaranteed. Second, ORION isn't something that would be considered AI. Something like optical character recognition isn't even commonly considered AI. Third, creating some rules to evaluate n=50 different (already computed) solutions to a problem like this where miles and time (or cost, or whatever else you're looking at) are equivalent isn't difficult. Just evaluate those 50 best solutions based on some other criteria (or set of criteria) that's pre-determined. If somehow there's *still* a tie and you can't find any more parameters to evaluate and make a decision, just pick one at random.

This isn't that far off from how people make complicated decisions between very similar options. Look at other criteria you weren't initially looking at before (or weren't initially presented with the information on). If all options or just some options are still basically equivalent, pick between those options at random if forced to make a choice. And the reason it's similar is because a person is evaluating such a decision in our example case.

ancient astronaut theorists say.. Yes.

 

[JustAnotherSup]

The math you and the other guy is using is flawed. It isnt 150x149x148x147..... Because you have variables of business and pickup times. So if you have 60 business it should be figured out starting with 60 x down to 2. Then however many pickups say 20 x down to 2. Then 70x down to 2. The final number shrinks exponentially.

If we assume that the optimal order for delivering a route will be running it in segments of straight business deliveries, straight pickups, and straight residential deliveries, you are correct (and to be extra fair to your idea, maybe we should add straight next day air to the start there).

Even if we don't throw anything 'interesting" into that mix like placard pickups that are earlier in the day than you'd finish your last business delivery...

That solution is clearly not the optimal solution for delivering a route in almost all cases. It isn't even close. Have next day air in an area that you'd normally spend a lot of time and miles getting to, and it can be your last air stop, and you can knock out all the nearby deliveries after making that air stop so you don't need to backtrack 20 minutes later in your day? Sorry, those additional deliveries are mixed business and commercial. So uhh, we'll just deliver the business and then backtrack later for 4 residential stops. Because we're running straight businesses. Or maybe that area is all residential. So we go to the first business, bypassing those 4 residential stops, and then spend 20 minutes later in the day to back track for no reason.

So maybe you improve on the idea and find an approach to when it makes sense to insert resis into that order of businesses, and insert pickups in there in a smarter way, et cetera. And you do it in a way that is, computationally speaking, reasonable based on our current level of technology. That approach is literally what ORION already is (except, worse at solving the problem we really want to solve, because the framework behind ORION is way more robust than what we're describing here).

 

[BadIdeaGuy]

Brute forcing the problem is not the answer. 150 stops and only visiting each stop once allows for 57133839564458545904789328652610540031895535786011264182548375833179829124845398393126574488675311145377107878746854204162666250198684504466355949195922066574942592095735778929325357290444962472405416790722118445437122269675520000000000000000000000000000000000000 possible combinations (let's just call this 6*10^262 for convenience). Maybe you have a supercomputer that can do 200 trillion calculations per second (2*10^14) and assuming that it only takes one calculation to find one combination (which is obviously not the case, but let's just go with it). Literally the world's fastest supercomputer. So it takes that supercomputer about 3*10^248 seconds to do the calculation. Or about 9.5*10^240 years. One trillion years is practically nothing compared to this span of time.

Even speeding up our supercomputer by several orders of magnitude barely helps. We just can't do it with our current level of technology. For that matter, we may NEVER be able to do it in a reasonable time frame at any level of technology humans will achieve.



First, there is not necessarily bound to be a set of possibilities with identical outcomes. It's possible, but not guaranteed. Second, ORION isn't something that would be considered AI. Something like optical character recognition isn't even commonly considered AI. Third, creating some rules to evaluate n=50 different (already computed) solutions to a problem like this where miles and time (or cost, or whatever else you're looking at) are equivalent isn't difficult. Just evaluate those 50 best solutions based on some other criteria (or set of criteria) that's pre-determined. If somehow there's *still* a tie and you can't find any more parameters to evaluate and make a decision, just pick one at random.

This isn't that far off from how people make complicated decisions between very similar options. Look at other criteria you weren't initially looking at before (or weren't initially presented with the information on). If all options or just some options are still basically equivalent, pick between those options at random if forced to make a choice. And the reason it's similar is because a person is evaluating such a decision in our example case.

No.
Just. No.

There's a lot wrong here.

First off, this is @zubenelgenubi 's actual question:

First question is, do the computers really have the power to come up, in a timely manner, with the best way to run a route with so many possibilities. I'm willing to believe that could be possible with enough equipment.

And you went off on a tangent about brute forcing.

Let's look at chess.
Computers can play chess at rating higher than 3000 Elo (Higher than any human player).
Chess suffers from the same type of combinatoric explosion that ORION does. And yet they have it down to a science. Almost no errors.

The reason for this is because of a process known as "pruning". Basically, this is what an engine will do when it encounters a "branch" on the search tree that do not have a significant reason to be looked through.

The exact same method should be usable with ORION. You aren't going to deliver one stop on one side of your route, and then drive to the other side of the route. That will almost never make sense. And so the algorithm could cull it, and reduce the number of branches that it must traverse by a massive amount.

200 trillion calculations per second ... Literally the world's fastest supercomputer.

You clearly aren't familiar with ASICS.
Just look at the cryptocurrency mining scene. You can get a computer capable of running 67 trillion SHA256 hashes (not a small feat, for the mathematically challenged here) per second for less than 2K USD.

Second, ORION isn't something that would be considered AI.

Completely wrong.

First off, ORION has basically been around for decades, in the form of the travelling salesman problem.
That is typically one of the first things you learn when you take artificial intelligence classes, and it absolutely qualifies as AI.

the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.

Something like optical character recognition isn't even commonly considered AI.

Neural nets are a process of taking statistics and using them to come to decisions about inputted data. Yes. OCR is AI.

You clearly aren't a programmer.

 

[Poop Head]

You clearly aren't a programmer

 

[MECH-lift]

ancient astronaut theorists say.. Yes.

ALIENS

 

[burrheadd]

No.
Just. No.

There's a lot wrong here.

First off, this is @zubenelgenubi 's actual question:


And you went off on a tangent about brute forcing.

Let's look at chess.
Computers can play chess at rating higher than 3000 Elo (Higher than any human player).
Chess suffers from the same type of combinatoric explosion that ORION does. And yet they have it down to a science. Almost no errors.

The reason for this is because of a process known as "pruning". Basically, this is what an engine will do when it encounters a "branch" on the search tree that do not have a significant reason to be looked through.

The exact same method should be usable with ORION. You aren't going to deliver one stop on one side of your route, and then drive to the other side of the route. That will almost never make sense. And so the algorithm could cull it, and reduce the number of branches that it must traverse by a massive amount.


You clearly aren't familiar with ASICS.
Just look at the cryptocurrency mining scene. You can get a computer capable of running 67 trillion SHA256 hashes (not a small feat, for the mathematically challenged here) per second for less than 2K USD.


Completely wrong.

First off, ORION has basically been around for decades, in the form of the travelling salesman problem.
That is typically one of the first things you learn when you take artificial intelligence classes, and it absolutely qualifies as AI.




Neural nets are a process of taking statistics and using them to come to decisions about inputted data. Yes. OCR is AI.

You clearly aren't a programmer.

No, he’s just another sup

 

[zubenelgenubi]

No.
Just. No.

There's a lot wrong here.

First off, this is @zubenelgenubi 's actual question:


And you went off on a tangent about brute forcing.

Let's look at chess.
Computers can play chess at rating higher than 3000 Elo (Higher than any human player).
Chess suffers from the same type of combinatoric explosion that ORION does. And yet they have it down to a science. Almost no errors.

The reason for this is because of a process known as "pruning". Basically, this is what an engine will do when it encounters a "branch" on the search tree that do not have a significant reason to be looked through.

The exact same method should be usable with ORION. You aren't going to deliver one stop on one side of your route, and then drive to the other side of the route. That will almost never make sense. And so the algorithm could cull it, and reduce the number of branches that it must traverse by a massive amount.


You clearly aren't familiar with ASICS.
Just look at the cryptocurrency mining scene. You can get a computer capable of running 67 trillion SHA256 hashes (not a small feat, for the mathematically challenged here) per second for less than 2K USD.


Completely wrong.

First off, ORION has basically been around for decades, in the form of the travelling salesman problem.
That is typically one of the first things you learn when you take artificial intelligence classes, and it absolutely qualifies as AI.




Neural nets are a process of taking statistics and using them to come to decisions about inputted data. Yes. OCR is AI.

You clearly aren't a programmer.

Thanks for helping to sort that out. Glad someone could understand what I was getting at, especially since I'm not a programmer. The issue with my second question is that, as I understand it, it is actually very difficult to program randomness.

 

[Brownslave688]

I can’t follow an Orion “solution” the tells me to go into cul-de-sacs or dead end streets multiple times a day to deliver stops a few houses away.

Easy money

 

[Brownslave688]

I see some new drivers wearing them all day. Meanwhile, they’re blowing through stop signs, running, not taking breaks, etc.

It’s mostly older drivers I see wearing them all day

 

[Overpaid Union Thug]

Easy money

Another driver and I have decided to only say CHA CHING! whenever we get the impulse to bitch about ORION to each other.

 

[MangoMango]

No.
Just. No.

There's a lot wrong here.

First off, this is @zubenelgenubi 's actual question:


And you went off on a tangent about brute forcing.

Let's look at chess.
Computers can play chess at rating higher than 3000 Elo (Higher than any human player).
Chess suffers from the same type of combinatoric explosion that ORION does. And yet they have it down to a science. Almost no errors.

The reason for this is because of a process known as "pruning". Basically, this is what an engine will do when it encounters a "branch" on the search tree that do not have a significant reason to be looked through.

The exact same method should be usable with ORION. You aren't going to deliver one stop on one side of your route, and then drive to the other side of the route. That will almost never make sense. And so the algorithm could cull it, and reduce the number of branches that it must traverse by a massive amount.


You clearly aren't familiar with ASICS.
Just look at the cryptocurrency mining scene. You can get a computer capable of running 67 trillion SHA256 hashes (not a small feat, for the mathematically challenged here) per second for less than 2K USD.


Completely wrong.

First off, ORION has basically been around for decades, in the form of the travelling salesman problem.
That is typically one of the first things you learn when you take artificial intelligence classes, and it absolutely qualifies as AI.




Neural nets are a process of taking statistics and using them to come to decisions about inputted data. Yes. OCR is AI.

You clearly aren't a programmer.

I am. You want to have a COBAL code off??

 

[pkgdriver]

I am. You want to have a COBAL code off??

COBOL

 

[no_map_needed]

Orion is very flawed. It is obvious the programming isn't done in-house. The variables missing and the mistakes it makes actually cost the company more money. And, there shouldn't be more ways to do the route than the number of stops the route goes out with x5 (air,schools,business,pickups,OCAs). 1 billion ways my ass.

 

[SafetyFirst]

So ... The part about Orion culling the option where it has you deliver one stop on one end of your route and then drive all the way to the opposite end of your route is definitely false.

It 100% will do that and frequently on at least 25% of the routes in my center.

 

[Poop Head]

1 billion ways my ass.

I'm not a math guy... But, after a little Google research. A route of 200 stops would have this many different ways to run it,

51476250067707216486487940160200993378605462690538824117424529787961666186325979299168297759488246475782024298753387060





That's not a joke. That's from some calculations website. So it must be true..

 

[JustDeliverIt]

Mathmatically, I believe everyone is right with the absurdly high number of possible outcomes. But I think that the number is way overblown. Someone mentioned "pruning" before and I believe this is exactly what is happening.

If you have a route with 200 stops, it's primarily a resi route (at least it would be in my center). You can pretty much count one street as one stop in most cases. You have 10 stops on a side street, most times you are doing all 10 at once. I know, we probably have all had days where you do some and then come back. But for the most part, I would think that it identifies these 10 stops as being a cluster together. You have streets and neighborhoods that you almost always do together, that drastically eliminates the "possible solutions" that ORION is going to spit out.

By no means am I saying it works, quite frankly it doesn't work well in the real world no matter how good it looks on paper. But someone high up was sold a bill of goods and now anyone in a truck is stuck with it. You either bitch about it or make money off of it. Your choice.

 

[zubenelgenubi]

So ... The part about Orion culling the option where it has you deliver one stop on one end of your route and then drive all the way to the opposite end of your route is definitely false.

It 100% will do that and frequently on at least 25% of the routes in my center.

He was talking about what it should do rather than what it actually does.

With RDO, you would know pretty much where all your stops are, especially on your own route. Right off the bat a decent driver can eliminate most of the nearly infinite possibilities, which makes Orion basically useless, even if it worked as intended. It might be good for making suggestions for unintuitive options, if it actually worked, but not having the stops listed in a predictable order hinders a driver's ability to come up with their own solution.

 

[JustDeliverIt]

but not having the stops listed in a predictable order hinders a driver's ability to come up with their own solution.

That's exactly what they want, eliminate driver knowledge and have mindless clones behind the wheel

 

[Brownslave688]

Orion is very flawed. It is obvious the programming isn't done in-house. The variables missing and the mistakes it makes actually cost the company more money. And, there shouldn't be more ways to do the route than the number of stops the route goes out with x5 (air,schools,business,pickups,OCAs). 1 billion ways my ass.

This has long been a flaw with the traveling Salesman problem. Humans will immediately eliminate almost all scenarios with common sense.

 

[Brownslave688]

That's exactly what they want, eliminate driver knowledge and have mindless clones behind the wheel

I will happily oblige.