Methods and systems for automatically tuning weights associated with breeding models
Systems and methods for use in identifying weights to be employed in a selection algorithm associated with plant advancement are disclosed. One example method includes identifying a start set of weights for a selection algorithm associated with a breeding program, and for each scale parameter value in a schedule, and for each of N iterations, modifying the start set of weights based on the scale parameter value, identifying a set of germplasm based on at least the modified set of weights, advancing the modified set of weights to a next iteration as the start set of weights when certain criteria are satisfied, and identifying the modified set of weights as an output when the iteration is equal to N.
1. A system for use in identifying weights to be employed in a selection algorithm associated with plant advancement, the system comprising:
a computing device configured to:
identify a start set of weights for a selection algorithm associated with a breeding program and a start set of germplasm defined by the selection algorithm and the start set of weights, the start set of germplasm associated with a performance threshold;
for each scale parameter value in a scale parameter schedule:
for each of N iterations, where Nis an integer:
modify the start set of weights based on the scale parameter value;
identify a set of germplasm based on the selection algorithm, the modified set of weights, and the scale parameter value;
advance the modified set of weights to a next iteration as the start set of weights, in response to i) an estimated performance of the identified set of germplasm satisfying the performance threshold and ii) the iteration being less than N;
advance the modified set of weights to a next iteration as the start set of weights, in response to i) the estimated performance of the identified set of germplasm failing to satisfy the performance threshold, ii) the estimated performance of the identified set of germplasm satisfying a further threshold, and iii) the iteration being less than N; and
identify the modified set of weights as an output for the N iterations in response to i) the estimated performance of the identified set of germplasm satisfying the performance threshold and ii) the iteration being equal to N; and
return one of the outputs, for one of the scale parameter values in the scale parameter schedule, as a final set of weights for the selection algorithm associated with the breeding program.
2. The system of claim 1 , wherein the computing device is configured to randomly modify the start set of weights consistent with the scale parameter value.
3. The system of claim 1 , wherein the scale parameter schedule includes multiple scale parameter values, and wherein each of the scale parameter values is indicative of variability available for modifying the set of weights.
4. The system of claim 1 , wherein the computing device is configured to identify the set of germplasm based on historical test data, via the selection algorithm.
5. The system of claim 1 , wherein the computing device is further configured, for each scale parameter value in the scale parameter schedule, and for each of the N iterations, to calculate the estimated performance of the set of germplasm as a percentage overlap between the set of germplasm and a reference set of germplasm, and to compare the estimated, calculated performance to the performance threshold.
6. The system of claim 5 , wherein the computing device is further configured, for each scale parameter value in the scale parameter schedule, and for each of the N iterations, to identify the modified set of weights as an output for the N iterations in response to a stop condition being satisfied.
7. The system of claim 6 , wherein the stop condition includes at least one of: the estimated performance of the modified set of germplasm exceeding a threshold and/or a degree of performance increase over a number of iterations.
8. The system of claim 1 , wherein the computing device, in connection with advancing the modified set of weights to a next iteration as the start set of weights, is configured to set the estimated performance of the identified set of germplasm as the performance threshold.
9. The system of claim 1 , wherein the computing device is further configured, in response to the estimated performance of the identified set of germplasm failing to satisfy the performance threshold and the iteration being less than N, to:
calculate the further threshold as an acceptance probability threshold for the estimated performance of the identified set of germplasm based on:
p
=
exp
(
-
E
′
-
E
S
)
,
wherein E′ is the estimated performance of the identified set of germplasm, E is the performance threshold, and S is the scale parameter value;
compare the acceptance probability to a randomly generated number for the scale parameter value and the N iteration; and
determine the further threshold is satisfied in response to the randomly generated number satisfying the acceptance probability.
10. The system of claim 1 , wherein the computing device is further configured, for each scale parameter value in the scale parameter schedule, for each of the N iterations, to:
discard the modified set of weights; and
advance the start set of weights to a next iteration as the start set of weights, in response to i) the estimated performance of the identified set of germplasm failing to satisfy the performance threshold, ii) the estimated performance of the identified set of germplasm failing to satisfy the further threshold, and iii) the iteration being less than N.
11. The system of claim 1 , wherein Nis less than 500.
12. The system of claim 1 , wherein the computing device is further configured to identify at least one germplasm based on the final set of weights and the selection algorithm.
13. The system of claim 12 , further comprising a plant disposed in a growing space of a breeding pipeline of the breeding program, the plant derived from the at least one germplasm.
14. The system of claim 1 , wherein the computing device is further configured to generate a report of the final set of weights and the estimated performance associated with the final set of weights.
15. A method for use in identifying weights to be employed in a selection algorithm associated with plant advancement, the method comprising:
identifying i) a start set of weights for a selection algorithm associated with a breeding program and ii) a start set of germplasm based on historical test data, the start set of germplasm associated with a performance threshold;
for each scale parameter value in a scale parameter schedule:
for each of N iterations, where Nis an integer:
modifying the start set of weights based on the scale parameter value;
identifying a set of germplasm based on the selection algorithm, the modified set of weights, and the scale parameter value;
advancing the modified set of weights to a next iteration as the start set of weights, in response to i) an estimated performance of the identified set of germplasm satisfying the performance threshold and ii) the iteration being less than N; and
identifying the modified set of weights as an output for the N iterations in response to i) the estimated performance of the identified set of germplasm satisfying the performance threshold and ii) the iteration being equal to N; and
returning one of the outputs, for one of the scale parameter values in the scale parameter schedule, as a final set of weights for the selection algorithm associated with the breeding program.
16. The method of claim 15 , wherein modifying the start set of weights includes randomly modifying the start set of weights consistent with the scale parameter value; and
wherein the scale parameter schedule includes multiple scale parameter values, and wherein each of the scale parameter values is indicative of variability available for modifying the set of weights.
17. The method of claim 15 , further comprising, for each scale parameter value in the scale parameter schedule, and for each of the N iterations:
calculating the estimated performance of the set of germplasm as a percentage overlap between the set of germplasm and a reference set of germplasm; and
comparing the estimated, calculated performance to the performance threshold.
18. The method of claim 17 , further comprising, in response to the estimated performance of the identified set of germplasm failing to satisfy the performance threshold and the iteration being less than N:
calculating the further threshold as an acceptance probability threshold for the estimated performance of the identified set of germplasm based on:
p
=
exp
(
-
E
′
-
E
S
)
,
wherein E′ is the estimated performance of the identified set of germplasm, E is the performance threshold, and Sis the scale parameter value;
comparing the acceptance probability to a randomly generated number for the scale parameter value and the N iteration; and
determining the further threshold is satisfied in response to the randomly generated number satisfying the acceptance probability.
19. The method of claim 15 , wherein Nis less than 500.
20. The method of claim 15 , further comprising planting a plant in a growing space of a breeding pipeline of the breeding program, the plant derived from the at least one germplasm.