International Business Machines Corporation

Yi Shao

Lei Tian

Jing James Xu

Peng Xue

Liang Wang

Xue Ying Zhang

Ning Zhang

Chun Lei Xu

Peng Hui Jiang

Jing Min Xu

Wei Wu

Si Er Han

Jia Wang

Na Zhao

Di Ling Chen

Jing Xu

Han Zhang

Zhe Zl Liu

Xiao Li Zhang

Shi Bin Liu

Hao Zheng

Fu Ju An

Ying Xu

Yi Li

A computer-implemented technique for decomposing semi-structured data is provided. In this technique, metadata for a predetermined number of records can be collected from semi-structured data that includes several records. A structured format is generated based on the metadata and the plurality of records is decomposed with the structured format.

Semi-structured data decomposition

A method, system, and computer for balance weighted voting. The method may comprise receiving, by a network interface, a scoring request. The method may further comprise, by a processing unit in response to the scoring request, generating a plurality of scores using a plurality of models. normalizing the plurality of scores, calculating an evaluation-based weighting factor from a first subset of the normalized scores, calculating a prediction-based based weighting factor from a second subset of the normalized scores, and calculating a balanced weighting predictor from the evaluation-based weighting factor and the prediction-based weighting factor. The method may further comprise returning, by the network interface, the balanced weighting predictor as an ensemble score for the scoring request.

Balance weighted voting

Disclosed are a computer-implemented method, a system and a computer program product for feature processing. In the computer-implemented method for feature processing, two input features selected from multiple features of each sample in a sample set are projected to one resulting feature by one or more processing units based on a specified curve. The sample set is updated by replacing the two input features with the one resulting feature for each sample in the sample set by one or more processing units. The projecting and the updating for the sample set are repeated by one or more processing units until the number of features of each sample in the sample set reaches a predetermined criterion.

Feature processing for machine learning

An approach is provided in which the approach trains a first machine learning model using a set of features corresponding to a set of build blocks. The set of build blocks include at least one dependency build block and at least one artifact package build block. The approach predicts a set of risk values of the set of build blocks using the trained first machine learning model, and marks at least one of the build blocks as a bottleneck in response to comparing the set of risk values against a risk threshold.

Deploy build blocks and analyze build bottleneck

According to embodiments of the present disclosure, a method, a device and a computer program product for code deployment are proposed. In the method, a deployment strategy for deploying code into a plurality of computing environments and respective amounts of resources provided by the plurality of computing environments are obtained. At least one code segment of the code to be deployed in a corresponding computing environment comprised in the plurality of computing environments is determined based on the deployment strategy and the respective amounts of resources. An amount of resources provided by the corresponding computing environment is sufficient to run the at least one code segment. The at least one code segment is deployed into the corresponding computing environment.

Code deployment

Methods, computer program products, and systems are presented. The methods include, for instance: obtaining a request for a predicted ensemble score in real-time. A subset of base model instances is formed by use of a preconfigured priority policy. A fitness score of the formed subset, quantifying the accuracy of the subset, is calculated as a sum of weights respective to the base model instances in the subset. A number of base models represented in the subset is less than or equal to a number of all based models.

Real time ensemble scoring optimization

According to an embodiment, a method, computer system, and computer program product for managing data is provided. The present invention may include accumulating a plurality of predicted outputs according to a data accumulation rule. The plurality of predicted outputs is generated by a predictive model executed by a first system. The present invention may include evaluating, by a second system, an accuracy of the predictive model. Evaluating the accuracy of the predictive model may include determining a degree of difference between the plurality of predicted outputs and information generated during a development stage of the predictive model. The present invention may include determining whether the accuracy of the predictive model has declined by an amount which exceeds a pre-determined threshold. The present invention may include updating the predictive model.

Automatic model refreshment based on degree of model degradation

An approach to optimizing predictive model analysis, comprising creating one or more model templates, decomposing a predictive model, wherein model information is extracted from the predictive model, storing the model information in the one or more model templates, creating a plurality of sub-models, associated with the predictive model, using the stored model information, sending the plurality of sub-models to a scoring engine, receiving results based on the plurality of sub-models from the scoring engine and generating predictions based on combining the results received from the scoring engine. The generated predictions can be sent to one or more analytic applications for further processing.

Predictive analysis with large predictive models

Growing community of inventors

Xi'an, China

Yi Shao