Hyperparameter optimization

1 . A computer-implemented method for supplying data domain datasets in machine training trials, the computer-implemented method comprising:

decoding a parameter code in a computer request for a training trial of the machine training trials, to identify a next dataset in a set of sample data stored in a first computer memory, wherein the parameter code corresponds to one or more predetermined data domain hyperparameter values of the next dataset;

responding to the computer request for the training trial by determining whether the next dataset is available in a second computer memory;

in a case where the next dataset is available in the second computer memory, supplying the next dataset that is available in the second computer memory; and

in a case where the next dataset is not available in the second computer memory, supplying the next dataset from the set of sample data.

2 . The computer-implemented method of claim 1 , wherein the one or more predetermined data domain hyperparameter values comprise at least one data domain hyperparameter value comprising a window size, a patch length, or a stride.

3 . The computer-implemented method of claim 1 , further comprising:

encoding the next dataset, sampled from the set of sample data, with the parameter code; and

storing the encoded next dataset to the second computer memory.

4 . The computer-implemented method of claim 1 , wherein the supplying of the next dataset from the set of sample data comprises:

extracting the next dataset from the set of sample data stored in the first computer memory;

transforming the next dataset into a training sample for performing the training trial; and

loading the next dataset with scheduled datasets, of the data domain datasets, into a predetermined order of the machine training trials.

5 . The computer-implemented method of claim 4 , wherein the transforming comprises at least one computer process comprising resizing images, decoding videos, or data augmentation.

6 . The computer-implemented method of claim 4 , wherein the transforming comprises batching together domain features and labels of the next dataset.

7 . The computer-implemented method of claim 4 , wherein the second computer memory is shared between a central processing node and a multiprocessing node to perform the machine training trials in parallel in the predetermined order.

8 . The computer-implemented method of claim 7 , wherein the predetermined order runs the machine training trials in parallel in the multiprocessing node with datasets, of the data domain datasets, corresponding to a same parameter code to optimize a similarity of the datasets in the multiprocessing node.

9 . The computer-implemented method of claim 7 , wherein the predetermined order runs the machine training trials in parallel in the multiprocessing node with datasets, of the data domain datasets, having different parameter codes to optimize a utilization of the multiprocessing node.

10 . The computer-implemented method of claim 7 , wherein:

the second computer memory is shared between the central processing node and a plurality of multiprocessing nodes;

the predetermined order runs the machine training trials in parallel with datasets, of the data domain datasets, having a same parameter code in at least one multiprocessing node of the plurality of multiprocessing nodes; and

the predetermined order runs the machine training trials in parallel with datasets, of the data domain datasets, having different parameter codes in the at least one multiprocessing node of the plurality of multiprocessing nodes.

11 . The computer-implemented method of claim 1 , further comprising selectively deleting the next dataset from the second computer memory.

12 . A computer program product for supplying data domain datasets in machine training trials, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, wherein the program instructions, when executed by a computer processor, cause the computer processor to:

decode a parameter code in a computer request for a training trial of the machine training trials, to identify a next dataset in a set of sample data stored in a first computer memory, wherein the parameter code corresponds to one or more predetermined data domain hyperparameter values of the next dataset;

respond to the computer request for the training trial by determining whether the next dataset is available in a second computer memory;

in a case where the next dataset is available in the second computer memory, supplying the next dataset that is available in the second computer memory; and

in a case where the next dataset is not available in the second computer memory, supplying the next dataset from the set of sample data.

13 . The computer program product of claim 12 , wherein the program instructions further cause the computer processor to:

encode the next dataset, sampled from the set of sample data, with the parameter code; and

store the encoded next dataset to the second computer memory.

14 . The computer program product of claim 13 , wherein the program instructions further cause the computer processor to share the second computer memory between a central processing node and a multiprocessing node to perform the machine training trials in parallel in a predetermined order.

15 . The computer program product of claim 14 , wherein the program instructions further cause the computer processor to run the machine training trials in parallel in the multiprocessing node with datasets, of the data domain datasets, having a same parameter code according to the predetermined order to optimize a similarity of the datasets in the multiprocessing node.

16 . The computer program product of claim 14 , wherein the program instructions further cause the computer processor to run the machine training trials in parallel in the multiprocessing node with datasets, of the data domain datasets, having different parameter codes according to the predetermined order to optimize a utilization of the multiprocessing node.

17 . The computer program product of claim 14 , wherein:

the program instructions further cause the computer processor to share the second computer memory between the central processing node and a plurality of multiprocessing nodes;

the predetermined order runs the machine training trials in parallel with datasets, of the data domain datasets, having a same parameter code to optimize a similarity of the datasets in at least one multiprocessing node of the plurality of multiprocessing nodes; and

the predetermined order runs the machine training trials in parallel with datasets, of the data domain datasets, having different parameter codes to optimize a utilization of the at least one multiprocessing node of the plurality of multiprocessing nodes.

18 . The computer program product of claim 14 , wherein the program instructions further cause the computer processor to selectively delete the next dataset from the second computer memory.

19 . A computer system for supplying data domain datasets in machine training trials, the computer system having a processor, a computer-readable memory, a computer-readable tangible storage device, and program instructions stored on the computer-readable tangible storage device for execution by the processor via the computer-readable memory, wherein the computer system is configured to perform:

decoding a parameter code in a computer request for a training trial of the machine training trials, to identify a next dataset in a set of sample data stored in a first computer memory, wherein the parameter code corresponds to one or more predetermined data domain hyperparameter values of the next dataset;

responding to the computer request for the training trial by determining whether the next dataset is available in a second computer memory;

in a case where the next dataset is available in the second computer memory, supplying the next dataset that is available in the second computer memory; and

in a case where the next dataset is not available in the second computer memory, supplying the next dataset from the set of sample data.