Questions and thoughts on grading coins

silverdollarfitz

Being new to coin collecting, i have read little on actually grading coins. So I just had a few questions...

Coin grading is all done based on opinions correct? There is no formula or anything to make a grade?

How much grading is done by computers? I read an article on some that PCGS does, but how much of it do they do that way?

So my thought, as a statistics major, would be to create a way to grade coins based on some formula, ie x amount of scratches = y grade. plus all the other things taken into account. Or there could be a way to make statistical predictions on coins based on other coins graded in the past.
This is called data mining. So get a database of all, for example, ms 62 graded 1887 P morgans, get pictures of them, and you can make a prediction on the next coin based on the data collected already, if it will be a ms 62.

This is just a thought, i doubt many statisticians are on here so i dunno if many people will follow what i'm saying, but its just an idea.

InternetJunky

<< <i>
How much grading is done by computers? I read an article on some that PCGS does, but how much of it do they do that way?
>>



I'm 100% sure coins aren't graded by computers, but are graded by people at PCGS. If you mean the coins information is entered into computers during processing, then yes they are done by computers.



<< <i>So my thought, as a statistics major, would be to create a way to grade coins based on some formula, ie x amount of scratches = y grade. plus all the other things taken into account. Or there could be a way to make statistical predictions on coins based on other coins graded in the past.
This is called data mining. So get a database of all, for example, ms 62 graded 1887 P morgans, get pictures of them, and you can make a prediction on the next coin based on the data collected already, if it will be a ms 62.
>>



This wouldn't work unless each coin with the same grade is photographed with the same camera, lighting, angle, distance, etc. The slightest angle or lighting adjustment can hide or show many different details/scratches/corrosion/color/etc. Also, the magnification of the picture of the coin also would play a roll.

Edit to add: But, if you owned the coins and photographed them yourself, it would play a significant roll on determining that grade or the next grade up.

bidask

I like the way you think..........one day you may revolutionize grading.

In the meantime by a book on coin grading, there are many good ones around.

ColonialCoinUnion

It's an art, not a science.

Tip

Grading is not done by computers but by humans, who depending on which grading company they work for, have tougher or looser standards. But because many people will resend the same coin in to the same service up to a dozen times to try and get a higher grade, and many succeed, goes to show you that this is a human mental as well as emotional decision. Usually the coin is inspected by 2 or 3 graders who have to agree on the final grade before it is certified.
In general the grade is determined by wear and marks etc. made at the mint ,and with the lower grades during circulation. In the higher grades the placement of these marks plays the important role. The cleaner the field the better the grade, for instance. Too bad there is not a computer program that would do this because the grading would certainly be more fair, in my opinion.
Stratches, gouges, rim dings, ben etc. are listed separately with the grade on the holder.
There are many books on grading, some fairly general, like "Photograde" that cover most coins, to more detailed publications. There are also some good descriptions if you search online for certain coins. If you find good large photos of certified coins you will be able to see some of the variation between MS60 and MS66 but for you to say this is a 62 or looks like a 64 will be difficult. Professionals use magnification, special light, and in house criteria. I think grading from AU50 and below is the easiest for most laymen.
Tip

lcoopie

welcome,
you may want to start reading

The Official Guide to Coin Grading by PCGS

yellowkid

<< <i>It's an art, not a science. >>




My thoughts exactly

LotsoLuck

<< <i>welcome,
you may want to start reading

The Official Guide to Coin Grading by PCGS >>

ambro51

Sure, its already done by statistics.


You look at coins, leaning about them, mentally comparing them to others you have seen and then you work grading them for XX number of years.

You grade XXXXX coins per year

You are paid XXXXXX$ to do it.

Thats the statistics........

golfer72

Eventually computers will grade coins

Aegis3

<< <i>Eventually computers will grade coins >>



Eventually humans will realize that computer grades tell us nothing about human grades.

silverdollarfitz

I actually just bought the pcgs book, but haven't read it yet. this was just me spitballing a potentially "statistical way" to grade coins. I think it could be graded scientifically and not based on opinions of "highly skilled professionals"

TomB

Computer grading has been touted for many years as the next thing, but has never successfully been brought to the hobby-industry. You may be able to write exquisite code relying on powerful statistical models, but eye appeal is an enormous factor in the grade of a coin and I don't know how one can write that into code.

ModernDollarNut

SDF, photograde pretty much bombed except as a an aid to learn grading and no one has come up with a valid computer grading system. As you know better than I computers are a whiz at statistics so that's not the problem.

Rather, the problem is the way the human mind can synthesize and build a qualitative composit of the data coming from one's eyeballs.

The closest we'll come to computer grading would have to involve rotating lights around the coin (perhaps a dense array of LED'slit up in sequence?) while optic censors pick up every stray reflection, coupled with a scanning microscope pick-up to render a 3-D image of the entire face of the coin being examined.

Run a few thousand graded-by-human coins through the system to build its data base and you might have something. Rob

InternetJunky

<< <i>SDF, photograde pretty much bombed except as a an aid to learn grading and no one has come up with a valid computer grading system. As you know better than I computers are a whiz at statistics so that's not the problem.

Rather, the problem is the way the human mind can synthesize and build a qualitative composit of the data coming from one's eyeballs.

The closest we'll come to computer grading would have to involve rotating lights around the coin (perhaps a dense array of LED'slit up in sequence?) while optic censors pick up every stray reflection, coupled with a scanning microscope pick-up to render a 3-D image of the entire face of the coin being examined.

Run a few thousand graded-by-human coins through the system to build its data base and you might have something. Rob >>



Well, there is already something to replicate a copy of any 3D object. Jay Leno actually uses it to replicate old car parts that cannot be replaced. Jay Leno Article NextEngine Desktop 3D Scanner Dimension 3D Printer

"These incredible devices allow you to make the form you need to create almost any part. The scanner can measure about 50,000 points per second at a density of 160,000 dots per inch (dpi) to create a highly detailed digital model. The 3D printer makes an exact copy of a part in plastic, which we then send out to create a mold. Some machines can even make a replacement part in cobalt-chrome with the direct laser sintering process. Just feed a plastic wire—for a steel part you use metal wire—into the appropriate laser cutter."

I guess creating a database of the digital models would help in an exact grading process, obviously not using the machine to try and replicate coins.

Makes you wonder why the chinese haven't gotten a hold of this yet. Its only $3,000

RedTiger

<< <i>...

Run a few thousand graded-by-human coins through the system to build its data base and you might have something. Rob >>



Yes, that's what broke the "nut" with chess programs. Large databases with access to all the board positions played in top level games and whether they led to win-lose or draw catapulted computers ahead of humans in chess. For many decades, chess was the "holy grail," an impossible programming project because of the near infinite number of board positions. What changed was the approach, an approach that incorporated human expertise into the decision making.

PCGS did a large scale prototype for computer grading. It cost about $1 million for the program that only did Morgan dollars. There is an article at the link below:
http://www.coingrading.com/compgrade1.html
link

At $1 million for a limited program that had trouble with 20% of the coins, the program was pulled off the market, and never expanded to do other coins. However, that project was done in 1990, and the algorithm did not use the chess-computer style database that builds on accumulated human knowledge and experience. The old cliche of anyone can learn how to grade, all it takes is a grading guide, a loupe and 20 years of looking at coins. If a programmer can find a way to give the computer all those years of hands-on "experience," it will be able to grade as well as top human graders, just as the top chess computers can compete and beat the top human players.

The hurdles at this point are more monetary than technological. Coin grading is a mature business, with a relatively limited revenue stream. Coin collectors and dealers don't like change, especially when a change may threaten their bacon. For these reasons it is unlikely at this point, that computerized coin grading will come from one of the established grading companies. It will more likely be done by a hobbyist group (like open source software projects) or a small start up.

InternetJunky

<< <i>

<< <i>...

Run a few thousand graded-by-human coins through the system to build its data base and you might have something. Rob >>



Yes, that's what broke the "nut" with chess programs. Large databases with access to all the board positions played in top level games and whether they led to win-lose or draw catapulted computers ahead of humans in chess. For many decades, chess was the "holy grail," an impossible programming project because of the near infinite number of board positions. What changed was the approach, an approach that incorporated human expertise into the decision making.

PCGS did a large scale prototype for computer grading. It cost about $1 million for the program that only did Morgan dollars. There is an article at the link below:
http://www.coingrading.com/compgrade1.html
link

At $1 million for a limited program that had trouble with 20% of the coins, the program was pulled off the market, and never expanded to do other coins. However, that project was done in 1990, and the algorithm did not use the chess-computer style database that builds on accumulated human knowledge and experience. The old cliche of anyone can learn how to grade, all it takes is a grading guide, a loupe and 20 years of looking at coins. If a programmer can find a way to give the computer all those years of hands-on "experience," it will be able to grade as well as top human graders, just as the top chess computers can compete and beat the top human players.

The hurdles at this point are more monetary than technological. Coin grading is a mature business, with a relatively limited revenue stream. Coin collectors and dealers don't like change, especially when a change may threaten their bacon. For these reasons it is unlikely at this point, that computerized coin grading will come from one of the established grading companies. It will more likely be done by a hobbyist group (like open source software projects) or a small start up. >>



Hey, I learn something everyday on this board. Thanks for that info

astrorat

Computer grading will not be implemented by any major grading service for two primary reasons:

1. Computer software/hardware cannot determine eye appeal (which is THE major factor in grading MS coins). The software and hardware technology are available for technical grading, but not market grading.

2. The premise (and promise) of computer grading is that the grade would be "set" for a given coin. What smart company would implement a system that virtually eliminates a major future market share? In other words, it is not in the best interest of any grading company to eliminate resubmissions.

Of course, that does not mean that computer coin grading won't resurface. Compugrade tried it and the software did not work well. PCGS tried it...and probably thought ahead about the implications...and then determined it would not work. Compugrade was a trail blazer...PCGS was smart.

Of course...that is only my opinion...and a free one at that!

Lane

silverdollarfitz

<< <i>PCGS did a large scale prototype for computer grading. It cost about $1 million for the program that only did Morgan dollars. There is an article at the link below:
http://www.coingrading.com/compgrade1.html
link >>



That was one of the articles I read. You would think that within almost 20 years some advances have been made.

I realize that eye appeal is important, but what the statistical system would do is compare the coin at hand to the ones on record. How i'm thinking this would work is it could spit out a probability for being a certain grade, and every grade, making a semi normal curve (maybe some of you took basic statistics in hs or college). then the graders could look at the coins and either agree or disagree with what the statistical output is.
So i dont think going just computers is the way to go. Maybe a combined system could help regulate the grading a bit more than what it is today.

It would be a huge undertaking to analyze the coin. Maybe that 3D thing would work, i dunno. I know i don't have the computer programming skills to actually make this. Its just a basic idea.

silverdollarfitz

is anyone else on here a statistician?

TomB

<< <i>

<< <i>PCGS did a large scale prototype for computer grading. It cost about $1 million for the program that only did Morgan dollars. There is an article at the link below:
http://www.coingrading.com/compgrade1.html
link >>



That was one of the articles I read. You would think that within almost 20 years some advances have been made.

I realize that eye appeal is important, but what the statistical system would do is compare the coin at hand to the ones on record. How i'm thinking this would work is it could spit out a probability for being a certain grade, and every grade, making a semi normal curve (maybe some of you took basic statistics in hs or college). then the graders could look at the coins and either agree or disagree with what the statistical output is.
So i dont think going just computers is the way to go. Maybe a combined system could help regulate the grading a bit more than what it is today.

It would be a huge undertaking to analyze the coin. Maybe that 3D thing would work, i dunno. I know i don't have the computer programming skills to actually make this. Its just a basic idea. >>


Did you see what I highlighted in bold? This statement may explain a difference in experience, knowledge or opinion between some folks in this thread. The graders at the TPGs (NGC or PCGS) typically look at each coin for only a few seconds in order to determine the grade. Therefore, putting the computer into the system and funneling coins that way before bringing them back to the graders for confirmation of the grade adds no value to the TPG, nor does it remove time.