Method and apparatus for generating information

1. A method for generating information, comprising:

receiving a modeling request, the modeling request comprising training data, a model type, a target number, and an evaluation indicator;

determining a target number of initial machine learning pipelines according to a type of the training data and the model type, the initial machine learning pipelines being used for model training; and

executing following model generation steps using the target number of initial machine learning pipelines: generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines; performing model training based on the training data, the target number of initial machine learning pipelines, and the target number of new machine learning pipelines, to generate trained models; evaluating the obtained trained models respectively according to the evaluation indicator; determining whether a preset training termination condition is reached; and determining, in response to determining that the preset training termination condition is reached, a target trained model from the obtained trained models according to evaluation results.

2. The method according to claim 1 , wherein the method further comprises:

selecting, in response to determining that the preset training termination condition is not reached, the target number of machine learning pipelines from the target number of initial machine learning pipelines and the target number of new machine learning pipelines as new initial machine learning pipelines according to the evaluation results, and continuing to execute the model generation steps.

3. The method according to claim 1 , wherein the method further comprises:

generating a model file for the target trained model, and evaluating the target trained model; and

pushing the model file and the evaluation result of the target trained model.

4. The method according to claim 1 , wherein the modeling request further comprises a maximum number of iterations, and the training termination condition comprises:

a number of executions of the model generation steps reaches the maximum number of iterations; or

in response to determining that the number of the executions of the model generation steps does not reach the maximum number of iterations, and the evaluation result of an optimal trained model obtained by a preset number of continuous executions of the model generation steps is unchanged.

5. The method according to claim 1 , wherein the generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines comprises:

selecting, according to a preset exchange ratio, initial machine learning pipelines from the target number of initial machine learning pipelines to form an initial machine learning pipeline subset, and executing following data exchange steps based on every two initial machine learning pipelines in the initial machine learning pipeline subset: determining whether two models corresponding to the two selected initial machine learning pipelines are the same; exchanging, in response to determining the same, model parameters of the two models to obtain two new machine learning pipelines; exchanging, in response to determining different, the models corresponding to the two selected initial machine learning pipelines to obtain two new machine learning pipelines; and

generating, for an initial machine learning pipeline in the target number of initial machine learning pipelines except the initial machine learning pipelines comprised in the initial machine learning pipeline subset, a new machine learning pipeline based on the excluded initial machine learning pipeline.

6. The method according to claim 1 , wherein the generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines comprises:

selecting two initial machine learning pipelines from the target number of initial machine learning pipelines, and executing following data exchange steps: determining whether two models corresponding to the two selected initial machine learning pipelines are the same; exchanging, in response to the same, model parameters of the two models to obtain two new machine learning pipelines; exchanging, in response to determining different, the models corresponding to the two selected initial machine learning pipelines to obtain two new machine learning pipelines; and

selecting two initial machine learning pipelines that do not undergo the data exchange steps from the target number of initial machine learning pipelines, and continuing to execute the data exchange steps.

7. The method according to claim 1 , wherein the initial machine learning pipeline comprises at least one data processing phase and one model training phase; and

the generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines comprises:

selecting, for a machine learning pipeline in the target number of initial machine learning pipelines, a preset number of parts of models corresponding to the at least one data processing phase and/or one model training phase of the initial machine learning pipeline for changing to generate a new machine learning pipeline.

8. An apparatus for generating information, comprising:

at least one processor; and

a memory storing instructions, the instructions when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:

receiving a modeling request, the modeling request comprising training data, a model type, a target number, and an evaluation indicator;

determining a target number of initial machine learning pipelines according a type of the training data and the model type, the initial machine learning pipelines being used for model training; and

executing following model generation steps using the target number of initial machine learning pipelines: generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines; performing model training based on the training data, the target number of initial machine learning pipelines, and the target number of new machine learning pipelines, to generate trained models; evaluating the obtained trained models respectively according to the evaluation indicator; determining whether a preset training termination condition is reached; and determining, in response to determining that the preset training termination condition is reached, a target trained model from the obtained trained models according to evaluation results.

9. The apparatus according to claim 8 , wherein the operations further comprise:

selecting, in response to determining that the preset training termination condition is not reached, the target number of machine learning pipelines from the target number of initial machine learning pipelines and the target number of new machine learning pipelines as new initial machine learning pipelines according to the evaluation results, and continuing to execute the model generation steps.

10. The apparatus according to claim 8 , wherein the operations further comprise:

generating a model file for the target trained model, and evaluate the target trained model; and

pushing the model file and the evaluation result of the target trained model.

11. The apparatus according to claim 8 , wherein the modeling request further comprises a maximum number of iterations, and the training termination condition comprises:

a number of executions of the model generation steps reaches the maximum number of iterations; or

in response to determining that the number of the executions of the model generation steps does not reach the maximum number of iterations, and the evaluation result of an optimal trained model obtained by a preset number of continuous executions of the model generation steps is unchanged.

12. The apparatus according to claim 8 , wherein the generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines comprises:

selecting, according to a preset exchange ratio, initial machine learning pipelines from the target number of initial machine learning pipelines to form an initial machine learning pipeline subset, and executing following data exchange steps based on every two initial machine learning pipelines in the initial machine learning pipeline subset: determining whether two models corresponding to the two selected initial machine learning pipelines are the same; exchanging, in response to determining the same, model parameters of the two models to obtain two new machine learning pipelines; exchanging, in response to determining different, the models corresponding to the two selected initial machine learning pipelines to obtain two new machine learning pipelines; and

generating, for an initial machine learning pipeline in the target number of initial machine learning pipelines except the initial machine learning pipelines comprised in the initial machine learning pipeline subset, a new machine learning pipeline based on the excluded initial machine learning pipeline.

13. The apparatus according to claim 8 , wherein the generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines comprises:

selecting two initial machine learning pipelines from the target number of initial machine learning pipelines, and execute following data exchange steps: determining whether two models corresponding to the two selected initial machine learning pipelines are the same; exchanging, in response to the same, model parameters of the two models to obtain two new machine learning pipelines; exchanging, in response to determining different, the models corresponding to the two selected initial machine learning pipelines to obtain two new machine learning pipelines; and

selecting two initial machine learning pipelines that do not undergo the data exchange steps from the target number of initial machine learning pipelines, and continuing to execute the data exchange steps.

14. The apparatus according to claim 8 , wherein the initial machine learning pipeline comprises at least one data processing phase and one model training phase; and

the generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines comprises:

selecting, for a machine learning pipeline in the target number of initial machine learning pipelines, a preset number of parts of models corresponding to the at least one data processing phase and/or one model training phase of the initial machine learning pipeline for changing to generate a new machine learning pipeline.

15. A non-transitory computer readable medium storing a computer program, wherein the computer program, when executed by a processor, causes the processor to perform operations, the operations comprising:

receiving a modeling request, the modeling request comprising training data, a model type, a target number, and an evaluation indicator;

determining a target number of initial machine learning pipelines according a type of the training data and the model type, the initial machine learning pipelines being used for model training; and

executing following model generation steps using the target number of initial machine learning pipelines: generating the target number of new machine learning pipelines based on the target number of initial machine learning pipelines; performing model training based on the training data, the target number of initial machine learning pipelines, and the target number of new machine learning pipelines, to generate trained models; evaluating the obtained trained models respectively according to the evaluation indicator; determining whether a preset training termination condition is reached; and determining, in response to determining that the preset training termination condition is reached, a target trained model from the obtained trained models according to evaluation results.