Intelligently optimized machine learning models
An approach for intelligent optimization of machine learning models for a target environment may be provided herein. The approach may include extracting metadata from a training pipeline for a first machine learning model that has been configured to operate within a first computing environment. The approach may also include mapping the extracted metadata to one or more constraints associated with a second machine learning model that has been configured to operate within a second computing environment. The approach may also include training the second machine learning model configured to the second computing environment, with the dataset that was used to train the first machine learning model, based on the mapped constraints. The approach may also include comparing performance metrics of the first machine learning model to corresponding metrics of the now trained second machine learning model.
1. A computer-implemented method for intelligent optimization of machine learning models, the method comprising:
extracting, by a processor, metadata from a training pipeline for a first machine learning model, wherein the first machine learning model is configured to a first computing environment, and wherein the first machine learning model comprises a convolutional neural network;
mapping, by the processor, the extracted metadata to one or more constraints associated with a second machine learning model configured to a second computing environment, and wherein the second machine learning model comprises a convolutional neural network;
training, by the processor, the second machine learning model configured to the second computing environment based, at least in part, on the mapped one or more constraints, with a dataset used to train the first machine learning model, wherein training comprises preprocessing a dataset by reducing dimension of images in a dataset and adjusting a number of pixels per filter in the convolution; and
comparing, by the processor, one or more performance metrics of the first machine learning model to one or more performance metrics of the trained second machine learning model configured to the second computing environment.
2. The computer-implemented method of claim 1 , further comprising:
determining, by the processor, if the one or more performance metrics of the trained second machine learning model configured to the second computing environment exceed a performance threshold; and
responsive to determining the one or more performance metrics of the trained second machine learning model configured to the second computing environment exceed the performance threshold, deploying, by the processor, the second machine learning model configured to the second computing environment to the second computing environment.
3. The computer-implemented method of claim 1 , further comprising:
extracting, by the processor, a machine learning algorithm from the first machine learning model based, at least in part, on the training pipeline metadata; and
identifying, by the processor, the second machine learning model configured to the second computing environment based, at least in part on, the extracted machine learning algorithm.
4. The computer-implemented method of claim 3 , wherein mapping the metadata extracted from the training pipeline for the first machine learning model to one or more constraints further comprises:
matching, by the processor, at least one stage of the first machine learning model's training pipeline to at least one stage of the second machine learning model configured to the second computing environment; and
configuring, by the processor, a training pipeline for the second machine learning model based, at least in part, on the matched at least one stage of the first machine learning model training pipeline.
5. The computer-implemented method of claim 3 , further comprising:
identifying, by the processor, a third machine learning model configured for the second environment based, at least in part, on the machine learning algorithm extracted from the first machine learning model;
mapping, by the processor, the extracted metadata to one or more constraints associated with the third machine learning model configured to the second computing environment; and
training, by the processor, the third machine learning model configured to the second computing environment based, at least in part, on the mapped one or more constraints, with the dataset used to train the first machine learning model.
6. The computer-implemented method of claim 5 , further comprising:
scoring, by the processor, a performance of each of the second trained machine learning model and the third trained machine learning model based, at least in part, on respective performance metrics for the second and third trained machine learning models; and
deploying, by the processor, to the second computing environment, a machine learning model with a highest score.
7. The computer-implemented method of claim 1 , wherein extracting metadata from the first machine learning model further comprises:
parsing, by the processor, a plurality of embedded tags in the training pipeline, wherein the embedded tags are within a programming code of the first machine learning model.
8. A computer system for intelligent optimization of machine learning models, the system comprising:
one or more computer processors;
one or more non-transitory computer readable storage devices; and
computer program instructions stored on the one or more computer readable storage devices, comprising instructions to:
extract metadata from a training pipeline for a first machine learning model, wherein the first machine learning model is configured to a first computing environment, and wherein the first machine learning model comprises a convolutional neural network;
map the extracted metadata to one or more constraints associated with a second machine learning model configured to a second computing environment, wherein the second machine learning model comprises a convolutional neural network;
train the second machine learning model configured to the second computing environment based, at least in part, on the mapped one or more constraints, with a dataset used to train the first machine learning model, wherein training comprises preprocessing a dataset by reducing dimensions of images in a dataset and adjusting a number of pixels per filter in the convolution; and
compare one or more performance metrics of the first machine learning model to one or more performance metrics of the trained second machine learning model configured to the second computing environment.
9. The computer system of claim 8 , further comprising instructions to:
determine if the one or more performance metrics of the trained second machine learning model configured to the second computing environment exceed a performance threshold; and
responsive to determining the one or more performance metrics of the trained second machine learning model configured to the second computing environment exceed the performance threshold, deploy the second machine learning model configured to the second computing environment to the second computing environment.
10. The computer system of claim 8 , further comprising instructions to:
extract a machine learning algorithm from the first machine learning model based, at least in part, on the training pipeline metadata; and
identify the second machine learning model configured to the second computing environment based, at least in part on, the extracted machine learning algorithm.
11. The computer system of claim 10 , wherein mapping the metadata extracted from the training pipeline for the first machine learning model to one or more constraints further comprises:
match at least one stage of the first machine learning model's training pipeline to at least one stage of the second machine learning model configured to the second computing environment; and
configure a training pipeline for the second machine learning model based, at least in part, on the matched at least one stage of the first machine learning model training pipeline.
12. The computer system of claim 10 , further comprising instructions to:
identify a third machine learning model configured for the second environment based, at least in part, on the machine learning algorithm extracted from the first machine learning model;
map the extracted metadata to one or more constraints associated with the third machine learning model configured to the second computing environment; and
train the third machine learning model configured to the second computing environment based, at least in part, on the mapped one or more constraints, with the dataset used to train the first machine learning model.
13. The computer system of claim 12 , further comprising instructions to:
score a performance of each of the second trained machine learning model and the third trained machine learning model based, at least in part, on respective performance metrics for the second and third trained machine learning models; and
deploy to the second computing environment, a machine learning model with a highest score.
14. The computer system of claim 10 , wherein extracting metadata from the first machine learning model further comprises:
parse a plurality of embedded tags in the training pipeline, wherein the embedded tags are within a programming code of the first machine learning model.
15. A computer program product for intelligent optimization of machine learning models, the computer program product comprising:
a non-transitory computer readable storage device having program instructions embodied therewith, the program instructions executable by a processor to cause the processors to perform a function, the program instructions comprising:
program instructions to extract metadata from a training pipeline for a first machine learning model, wherein the first machine learning model is configured to a first computing environment, and wherein the first machine learning model comprises a convolutional neural network;
program instructions to map the extracted metadata to one or more constraints associated with a second machine learning model configured to a second computing environment, and wherein the second machine learning model comprises a convolutional neural network;
program instructions to train the second machine learning model configured to the second computing environment based, at least in part, on the mapped one or more constraints, with a dataset used to train the first machine learning model, wherein training comprises preprocessing a dataset by reducing dimensions of images in a dataset and adjusting a number of pixels per filter in the convolution; and
program instructions to compare one or more performance metrics of the first machine learning model to one or more performance metrics of the trained second machine learning model configured to the second computing environment.
16. The computer program product of claim 15 , further comprising:
program instructions to determine if the one or more performance metrics of the trained second machine learning model configured to the second computing environment exceed a performance threshold; and
responsive to determining the one or more performance metrics of the trained second machine learning model configured to the second computing environment exceed the performance threshold, program instructions to deploy the second machine learning model configured to the second computing environment to the second computing environment.
17. The computer program product of claim 15 , further comprising:
program instructions to extract a machine learning algorithm from the first machine learning model based, at least in part, on the training pipeline metadata; and
program instructions to identify the second machine learning model configured to the second computing environment based, at least in part on, the extracted machine learning algorithm.
18. The computer program product of claim 17 , wherein mapping the metadata extracted from the training pipeline for the first machine learning model to one or more constraints further comprises:
program instructions to match at least one stage of the first machine learning model's training pipeline to at least one stage of the second machine learning model configured to the second computing environment; and
program instructions to configure a training pipeline for the second machine learning model based, at least in part, on the matched at least one stage of the first machine learning model training pipeline.
19. The computer program product of claim 17 , Further comprising:
program instructions to identify a third machine learning model configured for the second environment based, at least in part, on the machine learning algorithm extracted from the first machine learning model;
program instructions to map the extracted metadata to one or more constraints associated with the third machine learning model configured to the second computing environment; and
program instructions to train the third machine learning model configured to the second computing environment based, at least in part, on the mapped one or more constraints, with the dataset used to train the first machine learning model.
20. The computer program product of claim 17 , further comprising:
program instructions to score a performance of each of the second trained machine learning model and the third trained machine learning model based, at least in part, on respective performance metrics for the second and a third trained machine learning models; and
program instructions to deploy to the second computing environment, a machine learning model with a highest score.