Predictive performance optimizer
A method, apparatus and program product are provided for optimizing a training regimen to achieve performance goals. Historical training data is provided. At least one training regimen is defined. A training objective is selected for at least one training regimen to optimize. The training regimen is optimized by computing an initial training regimen solution and computing a neighbor solution at a distance from the initial training regimen solution. The neighbor solution is compared to the initial training regimen solution. If the neighbor solution is determined to be a better solution than the initial training regimen solution, the initial training regimen solution is replaced with the neighbor solution. The distance is updated per a schedule to compute a next neighbor solution.
1. A method of optimizing a training regimen to achieve performance goals, the method comprising:
providing historical training data;
defining at least one training regimen;
selecting a training objective for the at least one training regimen to optimize; and
optimizing the training regimen by:
computing with a hardware based processor an initial training regimen solution;
computing with the hardware based processor a neighbor solution at a distance from the initial training regimen solution;
comparing with the hardware based processor the neighbor solution to the initial training regimen solution;
determining if the neighbor solution is a better solution than the initial training regimen solution;
replacing the initial training regimen solution with the neighbor solution if the neighbor solution is a better solution; and
updating the distance per a schedule to compute a next neighbor solution,
wherein the determining of the neighbor solution is a better solution occurs using Performance=S·St·N c ·T −d , and
St
=
⌊
∑
lag
P
·
P
i
T
i
·
∑
i
j
(
lag
max
i
,
j
-
lag
min
i
,
j
)
N
i
⌋
,
wherein S is a scalar, c is a learning rate, d is a decay rate, T is true time passed since training began, N is a discreet number of training events over a training period, lag is an amount of true time passed between training events, and P is a true amount of time amassed in practice.
2. The method of claim 1 , wherein the neighbor solution is a better solution if a difference between a predictive model variable of the solution and training objective decreases.
3. The method of claim 1 , wherein optimizing the training regimen further comprises:
computing neighbor solutions until a local minima or local maxima is found.
4. The method of claim 3 , wherein optimizing the training regimen further comprises:
utilizing Monte Carlo methods to compute the neighbor solutions.
5. The method of claim 1 , wherein the training objective is selected from a group consisting of: performance objectives, cost objectives, schedule objectives, and combinations thereof.
6. The method of claim 1 , further comprising:
graphically overlaying predicted training performance with optimized training performance.
7. The method of claim 1 , further comprising:
graphically displaying historical training performance with at least one of predicted training performance and optimized training performance.
8. The method of claim 1 , further comprising:
updating the provided historical training data.
9. An apparatus comprising:
a processor; and
program code configured to be executed by the processor to optimize a training regimen to achieve performance goals, the program code further configured to provide historical training data, define at least one training regimen, select a training objective for the at least one training regimen to optimize, and optimize the training regimen by:
computing an initial training regimen solution;
computing a neighbor solution at a distance from the initial training regimen solution;
comparing the neighbor solution to the initial training regimen solution;
determining if the neighbor solution is a better solution than the initial training regimen solution;
replacing the initial training regimen solution with the neighbor solution if the neighbor solution is a better solution; and
updating the distance per a schedule to compute a next neighbor solution,
wherein the determining of the neighbor solution is a better solution occurs using Performance=S·St·N c ·T −d , and
St
=
⌊
∑
lag
P
·
P
i
T
i
·
∑
i
j
(
lag
max
i
,
j
-
lag
min
i
,
j
)
N
i
⌋
,
wherein S is a scalar, c is a learning rate, d is a decay rate, T is true time passed since training began, N is a discreet number of training events over a training period, lag is an amount of true time passed between training events, and P is a true amount of time amassed in practice.
10. The apparatus of claim 9 , wherein the program code is further configured to optimize the training regimen by computing neighbor solutions until a local minima or local maxima is found.
11. The apparatus of claim 10 , wherein the program code is further configured to optimize the training regimen utilizing Monte Carlo methods to compute the neighbor solutions.
12. The apparatus of claim 9 , wherein the training objective is selected from a group consisting of: performance objectives, cost objectives, schedule objectives, and combinations thereof.
13. The apparatus of claim 9 , further comprising:
a display; and
the program code is further configured to graphically overlay predicted training performance with optimized training performance.
14. The apparatus of claim 9 , further comprising:
a display; and
the program code is further configured to graphically display historical training performance with at least one of predicted training performance and optimized training performance.
15. A program product, comprising:
a computer recordable type medium; and
a program code configured to optimize a training regimen to achieve performance goals, the program code resident on the computer recordable type medium and further configured, when executed on a hardware implemented processor to provide historical training data, define at least one training regimen, select a training objective for the at least one training regimen to optimize, and optimize the training regimen by:
computing an initial training regimen solution;
computing a neighbor solution at a distance from the initial training regimen solution;
comparing the neighbor solution to the initial training regimen solution;
determining if the neighbor solution is a better solution than the initial training regimen solution;
replacing the initial training regimen solution with the neighbor solution if the neighbor solution is a better solution; and
updating the distance per a schedule to compute a next neighbor solution,
wherein the determining of the neighbor solution is a better solution occurs using Performance=S·St·N c ·T −d , and
St
=
⌊
∑
lag
P
·
P
i
T
i
·
∑
i
j
(
lag
max
i
,
j
-
lag
min
i
,
j
)
N
i
⌋
,
wherein S is a scalar, c is a learning rate, d is a decay rate, T is true time passed since training began, N is a discreet number of training events over a training period, lag is an amount of true time passed between training events, and P is a true amount of time amassed in practice.
16. The program product of claim 15 , wherein the program code is further configured to optimize the training regimen by computing neighbor solutions until a local minima or local maxima is found.
17. The program product of claim 16 , wherein the program code is further configured to optimize the training regimen utilizing Monte Carlo methods to compute the neighbor solutions.
18. The program product of claim 15 , wherein the training objective is selected from a group consisting of: performance objectives, cost objectives, schedule objectives, and combinations thereof.
19. The program product of claim 15 , wherein the program code is further configured to graphically overlay predicted training performance with optimized training performance.
20. The program product of claim 15 , the program code is further configured to graphically display historical training performance with at least one of predicted training performance and optimized training performance.