Rank: Newbie Handicapper
Groups: Member
Joined: 7/23/2008
Posts: 4
Points: 12
Location: Miami
|
Hi Scott,
Is it possible to make GM as a threaded application so that if you have a dual processor it will run with both cores during optimization? This will speed it up X2.
|
Rank: Administrator
Groups: Administration
Joined: 6/8/2008
Posts: 35
Points: 114
Location: Clearwater, Florida
|
bodog11 wrote:Hi Scott,
Is it possible to make GM as a threaded application so that if you have a dual processor it will run with both cores during optimization? This will speed it up X2. I started programming GM in 2001 and use Visual Basic 6. Unfortunately, VB6 does not support multi-threaded applications. I have been considering for some time upgrading the VB6 code to VB.NET, which does support multi-threaded apps. But, it is a huge undertaking... I have been looking at other ways to increase the speed in the optimization form (I assume that is what you're referring to). The last couple of days I have programmed a neural network that would take the handicapping factor inputs and predict the output (PR) based upon past racing results. It is incredibly faster than the manual optimization method which requires multiple database reads as the handicapping factor weights change. It is still a ways off from being release ready.
Good handicapping! Scott www.GreyhoundMaster.com
|
Rank: Newbie Handicapper
Groups: Member
Joined: 7/23/2008
Posts: 4
Points: 12
Location: Miami
|
This is great news if you plan to move to VB.net. It will speed things up a lot with the optimization process. As a side note, your idea of using the Neural Network is a great one. A popular program by the name of Neuralbet (not pushing it, just using it for reference) uses a basic form of BackPropagation and determines after 25000 epochs which weights are most likely to be profitable for each individual track. Similar to your optimization - it uses Speed, Break, Class, Running Style, Kennel etc. as the inputs to the NN. From what the author says - it has 4 hidden layers each with 4 neurons and its output is a "composite" weight or weights for that track. The only problem with this program.. is that it only does British Greyhound races!!
Hopefully in time, you can apply sort of the same approach (maybe even improved with a faster and more efficient NN architecture) so that GM can train the NN for each track and save its weights automatically. Looking forward to this someday! Just a recommendation for the possible future Neural Network development. You can use the FANN (Fast Artificial Neural Network) open source library - very fast, versatile and free! Great work on the 3.0 version!
|
