Two questions, and the last is probably a suggestion.
1. When I run an optimization, NT loads minute data from my broker to perform the backtests. The GA optimization actually runs quickly compared to loading the data. Usually I will see the results, change something, and rerun the optimization. Upon rerunning the optimization, NT reloads the data from scratch. Is there any way to keep that data in the buffer, and only reload if necessary?
2. My background is in optimization, and I found the whole estimated time to completion in the Control Center very odd. Only rarely can you know how long an optimization will take. NT knows because you set the maximum number of generations the algorithm performs. There should be a way to plot the optimization performance against the generation number or iteration number so you know if the optimization is converging near a solution or if you are practically picking random numbers. I know MultiCharts has this capability. Ideally, something like this would be super: http://www.vrand.com/visualDOC.html (I am in no way connected to Vanderplaats).
2.1. If, after you pass on suggestion #2 to your superiors, and better optimization methods are possibly in NT's future, your software developers should check out this: http://dakota.sandia.gov/software.html . I am not sure about the implications of using government-funded, open source software in commercial software, but at the very least it could give your developers an idea of what industrial-grade optimization looks like.
1. When I run an optimization, NT loads minute data from my broker to perform the backtests. The GA optimization actually runs quickly compared to loading the data. Usually I will see the results, change something, and rerun the optimization. Upon rerunning the optimization, NT reloads the data from scratch. Is there any way to keep that data in the buffer, and only reload if necessary?
2. My background is in optimization, and I found the whole estimated time to completion in the Control Center very odd. Only rarely can you know how long an optimization will take. NT knows because you set the maximum number of generations the algorithm performs. There should be a way to plot the optimization performance against the generation number or iteration number so you know if the optimization is converging near a solution or if you are practically picking random numbers. I know MultiCharts has this capability. Ideally, something like this would be super: http://www.vrand.com/visualDOC.html (I am in no way connected to Vanderplaats).
2.1. If, after you pass on suggestion #2 to your superiors, and better optimization methods are possibly in NT's future, your software developers should check out this: http://dakota.sandia.gov/software.html . I am not sure about the implications of using government-funded, open source software in commercial software, but at the very least it could give your developers an idea of what industrial-grade optimization looks like.
Comment