Systems and methods implementing an intelligent optimization platform
A system and method includes receiving a tuning work request for tuning an external machine learning model; implementing a plurality of distinct queue worker machines that perform various tuning operations based on the tuning work data of the tuning work request; implementing a plurality of distinct tuning sources that generate values for each of the one or more hyperparameters of the tuning work request; selecting, by one or more queue worker machines of the plurality of distinct queue worker machines, one or more tuning sources of the plurality of distinct tuning sources for tuning the one or more hyperparameters; and using the selected one or more tuning sources to generate one or more suggestions for the one or more hyperparameters, the one or more suggestions comprising values for the one or more hyperparameters of the tuning work request.
1. A machine learning-based system comprising:
a number of tuning worker machines to operate in parallel to set a tuning operation for tuning hyperparameters of an external model based on one or more tuning parameters of a tuning work request, the number of tuning worker machines including:
a first tuning worker machine to set a first tuning operation to reconfigure second hyperparameter values of a tuning source, the tuning source to tune hyperparameters of the external model; and
a second tuning worker machine to set a second tuning operation to generate suggestions of first hyperparameter values, the first hyperparameter values to replace the hyperparameters of the external model, the first tuning worker machine and the second tuning worker machine to operate in parallel; and
tuning sources to:
generate a plurality of first the first hyperparameter values for the hyperparameters of the external model based on the second tuning operation;
generate, in response to a status request corresponding to a partially complete tuning work request, a surrogate model to test an efficacy of third hyperparameter values generated before the status request; and
cause the second tuning worker machines to return the second hyperparameter values for the hyperparameters of the external model as a response to the tuning work request.
2. The machine learning-based system of claim 1 , wherein the number of tuning worker machines are to:
test an efficacy of the first hyperparameter values for the hyperparameters of the external model; and
identify a subset of the first hyperparameter values based on the testing to return as the response to the tuning work request.
3. The machine learning-based system of claim 1 , wherein the tuning sources include one of a machine learning-based optimization source and a Bayesian-based optimization source.
4. The machine learning-based system of claim 1 , wherein ones of the number of tuning worker machines are implemented by computers or computing servers that enable the number of tuning worker machines to operate at least one of independently or in parallel.
5. The machine learning-based system of claim 1 , wherein the number of tuning worker machines are to test the efficacy of the third hyperparameter values of the surrogate model against at least one objective function of the external model.
6. The machine learning-based system of claim 1 , wherein the number of tuning worker machines are to:
estimate a structure of the external model based on at least one or more objective functions of the external model; and
generate the surrogate model utilizing at least one of the tuning sources.
7. The machine learning-based system of claim 1 , further including an application programming interface to operably interface with the number of tuning worker machines to configure the tuning work request.
8. The machine learning-based system of claim 1 , further including an asynchronous work queue to store a plurality of tuning work requests accessible by the number of tuning worker machines for asynchronous processing.
9. The machine learning-based system of claim 1 , wherein:
the external model includes a machine learning model; and
the tuning work request indicates hyperparameters to be tuned for the machine learning model and one or more tuning constraints defining one or more tuning parameters that constrain one or more tuning operations when tuning the hyperparameters.
10. The machine learning-based system of claim 9 , wherein the one or more tuning constraints include:
(i) a tuning budget that limits a number of tuning operations of the number of tuning worker machines against the hyperparameters of the machine learning model; and
(ii) bounded data that identifies a maximum value and minimum value for ones of the hyperparameters, the bounded data to limit discoverable values for the ones of the hyperparameters during the one or more tuning operations.
11. The machine learning-based system of claim 1 , wherein the tuning sources are to:
execute the tuning work request to tune the hyperparameters of the external model based on an estimated time of completion;
obtain an intervening tuning work request corresponding to a second tuning work request that is received at a time during the execution of the tuning work request before the estimated time of completion of the tuning work request; and
in response to obtaining the intervening tuning work request, automatically switch between the tuning sources to expedite an evaluation of fourth hyperparameter values generated before the estimated time of completion.
12. The machine learning-based system of claim 1 , wherein in response to returning the first hyperparameter values, the tuning sources are to improve an effectiveness of the external model by tuning the external model using one or more of the first hyperparameter values.
13. The machine learning-based system of claim 1 , wherein the tuning sources are to generate, in response to a tuning worker machine of the number of tuning worker machines failing, the second hyperparameter values for the hyperparameters of the external model based on partial information from a subset of the number of tuning worker machines that remain operable.
14. A method for tuning an external model, the method comprising:
setting a first tuning operation to reconfigure second hyperparameter values of a tuning source, the tuning source to tune hyperparameters of the external model;
setting a second tuning operation to generate suggestions of first hyperparameter values, the first hyperparameter values to replace the hyperparameters of the external model based on one or more tuning parameters included in a tuning work request, the first tuning operation and the second turning operation operating in parallel;
generating, using one or more tuning sources, the first hyperparameter values for the hyperparameters of the external model based on the second tuning operation;
generating, in response to a status request corresponding to a partially complete tuning work request, a surrogate model to test an efficacy of third hyperparameter values generated before the status request; and
returning, via an application programming interface, the first hyperparameter values for the hyperparameters of the external model as a response to the tuning work request.
15. The method of claim 14 , further including:
testing an efficacy of ones of the first hyperparameter values for the hyperparameters of the external model; and
identifying a subset of the first hyperparameter values based on the testing to return as the response to the tuning work request.
16. The method of claim 14 , wherein generating the first hyperparameter values includes utilizing one of a machine learning-based optimization source and a Bayesian-based optimization source.
17. The method of claim 14 , wherein setting the second tuning operation includes setting a plurality of tuning operations independently or parallelly utilizing a number of different computers or a number of different computing servers.
18. The method of claim 14 , further including testing the efficacy of the second third hyperparameter values of the surrogate model based at least on an objective function of the external model.
19. The method of claim 14 , further including estimating a structure of the external model based on at least one or more objective functions of the external model.
20. The method of claim 14 , further including:
selecting a first tuning source of a plurality of tuning sources to generate the first hyperparameter values;
providing hyperparameter data corresponding to the hyperparameters of the external model and tuning constraint data derived from the tuning work request to the selected first tuning source to enable a third tuning operation by the first tuning source; and
generating first tuning data based on the third tuning operation.
21. The method of claim 20 , further including:
selecting a second tuning source of the plurality of tuning sources that is different than the first tuning source;
providing the hyperparameter data corresponding to the hyperparameters of the external model and the tuning constraint data derived from the tuning work request to the selected second tuning source to enable a fourth tuning operation by the second tuning source; and
generating second tuning data based on the fourth tuning operation.
22. The method of claim 14 , further including:
determining that a tuning worker machine of a number of tuning worker machines fails during the first tuning operation; and
generating, in response to the tuning worker machine failing, the first hyperparameter values for the hyperparameters of the external model based on partial information from a subset of the number of tuning worker machines that remain operable.