Full 2023 Dataset available at no charge

[headhawg]

You didn't "parse" anything. You tried to load a large XML file into a text editor. No offense but you don't have the chops to make this usable. It's possible that someone will put it into a different format, but sharing it would be in violation of the terms of service based on what @equusvates posted earlier.

[equusvates]

Originally Posted by wiretowire68

Parsed this data into my Notepad...it all comes up in Chinese like writing? is this correct? @ndly. Again, not a software guy but my Notepad stops working like it cannot handle that amount of parsing. Is this a memory issue? I have tried to repair as I run Windows 11. Also, will this data fit automatically into a spreadsheet or do I need to code in the original XMl or whatever into my system.

Treat me like a I am beginner at this as would like to be able to run this and use it as a building block.

Not NotePad which I believe is a MicroSoft program. "NotePad++", which is an open source program which has a long history and is actively maintained. NotePad++ uses plugins to correctly display files types and color code the text. I believe the XML plugin is standard or you will be prompted to install the plugin. NotePad++ is just a more powerful editor versus NotePad.

By the way, I am still writing code to parse the files and now into the past performances sections. I look at each element to make sure I understand the data types. So it takes a while. A typical file unzipped, may contain 166,000 lines of text.

[equusvates]

Originally Posted by Dave Schwartz

I like your style.


Reach out if you want some conversation on your AI approach. I have some degree of experience writing AI engines.

Dave, been meaning to get back with you concerning A.I. I understand convolutional networks and genetic algorithms, but a convolutional genetic algorithm I can only think that you are using the feedback as formcycle analysis, i.e. a memory of some sort. In the selection thread I place a page of selections which only displays angles and pace analysis. The odds are being set by a Clips Rules Engine. The Clips Rule Engine allows for creating composite angles and weighting of those angles. I am, as stated above, training a Linear Genetic Programming that teaches itself to write programs to handicap races. Most literature refers to tree based genetic programming which up to now have stifled the progress of genetic programming. Any complex program will require the use of many subroutines and lengthy programs. As of now, I consider the programs being created by the LGP to be at the intelligence of a two year old not even potty trained.

In the selection page there are many angles which may confuse. A good thing about machine learning is, if given enough, it may learn what to ignore in some situations. My target goal for my A.I. is 60% correct on training data, and 30% correct on testing data. At 30%, independent of the public odds, should be very profitable.

[Dave Schwartz]

Originally Posted by equusvates

Dave, been meaning to get back with you concerning A.I. I understand convolutional networks and genetic algorithms, but a convolutional genetic algorithm I can only think that you are using the feedback as formcycle analysis, i.e. a memory of some sort. In the selection thread I place a page of selections which only displays angles and pace analysis. The odds are being set by a Clips Rules Engine. The Clips Rule Engine allows for creating composite angles and weighting of those angles. I am, as stated above, training a Linear Genetic Programming that teaches itself to write programs to handicap races. Most literature refers to tree based genetic programming which up to now have stifled the progress of genetic programming. Any complex program will require the use of many subroutines and lengthy programs. As of now, I consider the programs being created by the LGP to be at the intelligence of a two year old not even potty trained.

In the selection page there are many angles which may confuse. A good thing about machine learning is, if given enough, it may learn what to ignore in some situations. My target goal for my A.I. is 60% correct on training data, and 30% correct on testing data. At 30%, independent of the public odds, should be very profitable.

Interesting - the clips IDEA.
(For those who have interest but don't know, that's an open source, rule-based engine.)

My challenge with angles is that, with the whales having such a huge impact and modifying their system every January, the value of data going too far back may be significantly diminished.
__________
About The DeTerminator...
There is an internal section in deTerminator that is rule-based but not in the way you might think.

It is convolutional in the sense that it develops/learns moving left-to-right, similar to layers. That is, the leftmost layers are of greater importance than the rightmost.

I call them PHASES, but they serve the same purpose. Sort of like you have to understand favorites before you can understand longshots.

or... You must understand who's likely to go to the front before you can decide who will win in the stretch. And before any of that, you must understand pace pressure.

It is not DIRECT feedback, like a neural net.
The feedback is indirect.

The really powerful part is the Topological stuff: the segmentation of the handicapping.

It addresses the issue that a 4f dash on the dirt for 2-year olds must be handicapped far differently than a 9f graded stakes on the turf. Yet it is all one giant, global system.

But the most exciting part is the part I am coding over the next 3-4 days. It's an informational series called NHT.

That stands for "NOBODY HAS THIS."
It's all factors that nobody has ever seen.

[equusvates]

Originally Posted by Dave Schwartz

Interesting - the clips IDEA.
(For those who have interest but don't know, that's an open source, rule-based engine.)

My challenge with angles is that, with the whales having such a huge impact and modifying their system every January, the value of data going too far back may be significantly diminished.
__________
About The DeTerminator...
There is an internal section in deTerminator that is rule-based but not in the way you might think.

It is convolutional in the sense that it develops/learns moving left-to-right, similar to layers. That is, the leftmost layers are of greater importance than the rightmost.

I call them PHASES, but they serve the same purpose. Sort of like you have to understand favorites before you can understand longshots.

or... You must understand who's likely to go to the front before you can decide who will win in the stretch. And before any of that, you must understand pace pressure.

It is not DIRECT feedback, like a neural net.
The feedback is indirect.

The really powerful part is the Topological stuff: the segmentation of the handicapping.

It addresses the issue that a 4f dash on the dirt for 2-year olds must be handicapped far differently than a 9f graded stakes on the turf. Yet it is all one giant, global system.

But the most exciting part is the part I am coding over the next 3-4 days. It's an informational series called NHT.

That stands for "NOBODY HAS THIS."
It's all factors that nobody has ever seen.

Okay I somewhat get it. I fully understand the need for segmentation. To make my problem worst, the LGP is started up many times to create a program for class / distance / surface / number of entries. Because of this, you quickly run out of training samples. There are many races at 6, 7 and 8 runners. As the number of entries goes lower or higher, there are fewer races to learn from. I will see what occurs in the Equibase data files. I typically learn with a population of 100 to 500 chromosomes and limit the length of the created programs to 2000 instructions. The generated program output is the order of finish. In doing so some entries are not shown in the output, while others are dulplicated.

[equusvates]

Here a screenshot with the a.i. shown. I consider this nowhere near my goal.

[equusvates]

Forgot image.

[Dave Schwartz]

Originally Posted by equusvates

Okay I somewhat get it. I fully understand the need for segmentation. To make my problem worst, the LGP is started up many times to create a program for class / distance / surface / number of entries. Because of this, you quickly run out of training samples. There are many races at 6, 7 and 8 runners. As the number of entries goes lower or higher, there are fewer races to learn from. I will see what occurs in the Equibase data files. I typically learn with a population of 100 to 500 chromosomes and limit the length of the created programs to 2000 instructions. The generated program output is the order of finish. In doing so some entries are not shown in the output, while others are dulplicated.

Reach out and schedule an appointment.

I can give you some insight.

[equusvates]

Originally Posted by Dave Schwartz

Reach out and schedule an appointment.

I can give you some insight.

Okay, how does making an appointment work?

[Dave Schwartz]

Originally Posted by equusvates

Okay, how does making an appointment work?

You email me, we connect, then we schedule a Zoom meeting.

[wiretowire68]

Interesting Dave, never heard back but Canadians get caught up when it comes to the Stanley Cup chase. Especially, when you have been waiting your whole life just to see your team possibly make it to the finals and they continue to get knocked out in the first round in the hockey capital of the world. Looks like your work is coming along.... hurry up and as Mr. Cable Guy says.. get er done.

[Dave Schwartz]

Originally Posted by wiretowire68

Interesting Dave, never heard back but Canadians get caught up when it comes to the Stanley Cup chase. Especially, when you have been waiting your whole life just to see your team possibly make it to the finals and they continue to get knocked out in the first round in the hockey capital of the world. Looks like your work is coming along.... hurry up and as Mr. Cable Guy says.. get er done.

Never heard back from who?

[wiretowire68]

Chuck

[Dave Schwartz]

You mean to sell his whatever-it's-called?

[wiretowire68]

the nice guy.... with yo's yo's on here who think you should sell it for nothing or do not have e-transfer