Abstract: Provided are systems and methods for auto-completing debriefing processing for a machine learning model pipeline based on a type of predictive algorithm. In one example, the method may include one or more of building a machine learning model pipeline via a user interface, detecting, via the user interface, a selection associated with a predictive algorithm included within the machine learning model pipeline, in response to the selection, identifying debriefing components for the predictive algorithm based on a type of the predictive algorithm from among a plurality of types of predictive algorithms, and automatically incorporating processing for the debriefing components within the machine learning model pipeline such that values of the debriefing components are generated during training of the predictive algorithm within the machine learning model pipeline.

Abstract: Systems and methods include reception of a plurality of records, each of the plurality of records associating each of a plurality of features with a respective value, a second feature with a value, and a target feature with a value, a first machine learning model trained based on the plurality of records to output a value of the target feature based on values of each of the plurality of features, a second machine learning model trained based on the plurality of records to output a value of the second feature based on the values of each of the plurality of features, determination, based on the trained second machine learning model, of a first one or more of the plurality of features which are correlated to the second feature, determination of an influence of each of the first one or more features on the trained first machine learning model, and determination of a first value associated with the second feature based on the determined influences and on the trained second machine learning model.

Abstract: Provided is a system and method which build a composite time-series machine learning model including a core model and a debrief model that includes a combination of the core model and a surrogate model. In one example, the method may include executing the plurality of models on test data and determining accuracy values and interpretability toughness values for the plurality models, selecting a most accurate model as a core model based on the accuracy values and select a most interpretable model as a surrogate model from among other models remaining in the plurality of models based on the interpretability toughness values, building a composite model comprising the core model, the surrogate model, and instructions for generating a debrief model for debriefing the core model based on a combination of the core model and the surrogate model, and storing the composite model within the memory.

Abstract: Systems and methods include identification first members of a child level of a dimension hierarchy which are associated with boundaries between second members of a parent level of the dimension hierarchy, training of a decision tree model based on data associated with the child level, extraction of predicates on the child level from the trained decision tree model, determination of a value based on the identified first members of the child level and on the extracted predicates on the child level, and determination, based on the value, whether to include the parent level and the child level within training data or to include the parent level and not include the child level within the training data.

Abstract: Provided is a system and method which generates a composite machine learning model that can filter downtime data from a time-series data signal and perform a prediction on remaining time-series data. In one example, the method may include detecting a pattern of downtime data within a time-series data signal, removing a subset of data from the time-series data based on the detected pattern of downtime and building a machine learning model to make predictions based on remaining data in the time-series data, generating segregation instructions configured to remove downtime data from a time-series data signal of a same type and to predict zero on future dates matching the downtime segregation codes, and building a composite machine learning model that includes the trained machine learning model and the segregation instructions for filtering data that is input to the trained machine learning models.

Abstract: Gradient Boosting Decision Tree (GBDT) successively stacks many decision trees which at each step try to fix the residual errors from the previous steps. The final score produced by the GBDT is the sum of the individual scores obtained by the decision trees for an input vector. Overfitting in GBDT can be reduced by removing the input values that have the least impact on the output from the training data. One way to determine which input variable has the lowest predictive value is to determine the input variable that is used for the first time in the latest decision tree in the GBDT. This method of identifying the low-predictive features to be removed does not require that earlier trees be regenerated to generate the new GBDT. Since the removed feature was already not used in the earlier trees, those trees already ignore the removed feature.

Abstract: Gradient Boosting Decision Tree (GBDT) successively stacks many decision trees which at each step try to fix the residual errors from the previous steps. The final score produced by the GBDT is the sum of the individual scores obtained by the decision trees for an input vector. Overfitting in GBDT can be reduced by removing the input values that have the least impact on the output from the training data. One way to determine which input variable has the lowest predictive value is to determine the input variable that is used for the first time in the latest decision tree in the GBDT. This method of identifying the low-predictive features to be removed does not require that earlier trees be regenerated to generate the new GBDT. Since the removed feature was already not used in the earlier trees, those trees already ignore the removed feature.

Abstract: Features are used to train one or more ML models in a modelling layer. In a feature selection layer, each generated ML model is analyzed to determine, for each input feature, a degree of importance of the feature on the results generated by the ML model. Features with low importance are identified and the information is propagated backward to the data source and feature engineering layers. In response, the data source and feature engineering layers refrain from gathering or generating the unimportant features. Based on a confidence measure of the determination that each feature is important or unimportant, a number of periods between reevaluation of the feature importance is determined. After the number of periods has elapsed, a removed feature is restored to the pipeline.

Abstract: Disclosed herein are system, method, and computer program product embodiments for generating a bridge between analytical models. In an embodiment, a server can extract a first variable dependency schema from a first model (e.g., predictive model or business intelligence report) and a second variable schema from a second model (e.g., predictive model or business intelligence report). The first variable dependency schema includes a first definition of a relationship between a first variable and a second variable. The server can compare the first variable dependency schema and the second variable dependency schema. Furthermore, the server can generate a modification to be made in the second variable dependency schema based on the first definition of the relationship between the first and second variable and outputs the modification to be made to the second variable dependency schema.

Abstract: A predictive model pipeline data store may contain electronic records defining a predictive model pipeline composed of operation nodes. Based on the information in the data store, an execution framework platform may calculate a hash value for each operation node by including all recursive dependencies using ancestor node hash values and current node parameters. The platform may then compare each computed hash value with a previously computed hash value associated with a prior execution of a prior version of the pipeline. Operation nodes that have an unchanged hash value may be tagged “idle.” Operation nodes that have a changed hash value may be tagged “train and apply” or “apply” based on current node parameters (and an “apply” tag may propagate backwards through the pipeline to ancestor nodes). The platform may then ignore the operation nodes tagged “idle” when creating a physical execution plan to be provided to a target platform.

Abstract: Systems and methods include determination of a first plurality of sets of data, each including values associated with respective ones of a first plurality of features, partial training of a first machine-learning model based on the first plurality of sets of data, determination of one or more of the first plurality of features to remove based on the partially-trained first machine-learning model, removal of the one or more of the first plurality of features to generate a second plurality of sets of data, partial training of a second machine-learning model based on the second plurality of sets of data, determination that a performance of the partially-trained second machine-learning model is less than a threshold, addition, in response to the determination, of the one or more of the first plurality of features to the second plurality of sets of data, and training of the partially-trained first machine-learning model based on the first plurality of sets of data.

Abstract: Features are used to train one or more ML models in a modelling layer. In a feature selection layer, each generated ML model is analyzed to determine, for each input feature, a degree of importance of the feature on the results generated by the ML model. Features with low importance are identified and the information is propagated backward to the data source and feature engineering layers. In response, the data source and feature engineering layers refrain from gathering or generating the unimportant features. Based on a confidence measure of the determination that each feature is important or unimportant, a number of periods between reevaluation of the feature importance is determined. After the number of periods has elapsed, a removed feature is restored to the pipeline.

Abstract: Provided is a system and method which can identify a causal relationship for anomalies in a time-series signal based on co-occurring and preceding anomalies in another time-series signal. In one example, the method may include identifying a recurring anomaly within a time-series signal of a first data value, determining a time-series signal of a second data value that is a cause of the recurring anomaly in the time-series signal of the first data value based on a preceding and co-occurring anomaly in the time-series signal of the second data value, and storing a correlation between the preceding and co-occurring anomaly in the time-series signal of the second data value and the recurring anomaly in the time-series signal of the first data value.

Abstract: Provided is a system and method which decomposes a predicted output signal of a time-series forecasting model into a plurality of sub signals that correspond to a plurality of components, and determines and displays a global contribution of each component. In one example, the method may include iteratively predicting an output signal of a time-series data value via execution of a time-series model, decomposing the predicted output signal into a plurality of component signals corresponding to a plurality of components of the time-series machine learning algorithm, respectively, and displaying the plurality of global values via a user interface.

Abstract: Provided is a system and method which trains a model based on a horizon-wise cost function which accounts for error across a horizon rather than just a next point in time thereby improving the accuracy of the trained model in the long term. In one example, the method may include storing time-series data, executing a training iteration for a machine learning model based on one or more parameter values, determining error values between the predicted values output by the machine learning model and actual values of the time-series data for a plurality of intervals included in a horizon of the time-series data, generating a total error value for the horizon based on the determined error values for the intervals, and storing the generated total error value for the horizon. The method also enables a user to dynamically adjust a weight for each interval of the horizon.

Abstract: Disclosed herein are system, method, and computer program product embodiments for generating a bridge between analytical models. In an embodiment, a server can extract a first variable dependency schema from a first model (e.g., predictive model or business intelligence report) and a second variable schema from a second model (e.g., predictive model or business intelligence report). The first variable dependency schema includes a first definition of a relationship between a first variable and a second variable. The server can compare the first variable dependency schema and the second variable dependency schema. Furthermore, the server can generate a modification to be made in the second variable dependency schema based on the first definition of the relationship between the first and second variable and outputs the modification to be made to the second variable dependency schema.

Abstract: Features are used to train one or more ML models in a modelling layer. In a feature selection layer, each generated ML model is analyzed to determine, for each input feature, a degree of importance of the feature on the results generated by the ML model. Features with low importance are identified and the information is propagated backward to the data source and feature engineering layers. In response, the data source and feature engineering layers refrain from gathering or generating the unimportant features. Based on a confidence measure of the determination that each feature is important or unimportant, a number of periods between reevaluation of the feature importance is determined. After the number of periods has elapsed, a removed feature is restored to the pipeline.

Abstract: Gradient Boosting Decision Tree (GBDT) successively stacks many decision trees which at each step try to fix the residual errors from the previous steps. The final score produced by the GBDT is the sum of the individual scores obtained by the decision trees for an input vector. Overfitting in GBDT can be reduced by removing the input values that have the least impact on the output from the training data. One way to determine which input variable has the lowest predictive value is to determine the input variable that is used for the first time in the latest decision tree in the GBDT. This method of identifying the low-predictive features to be removed does not require that earlier trees be regenerated to generate the new GBDT. Since the removed feature was already not used in the earlier trees, those trees already ignore the removed feature.

Abstract: Techniques are described for identifying an input training dataset stored within an underlying data platform; and transmitting instructions to the data platform, the instructions being executable by the data platform to train a predictive model based on the input training dataset by delegating one or more data processing operations to a plurality of nodes across the data platform.

Abstract: A predictive model pipeline data store may contain electronic records defining a predictive model pipeline composed of operation nodes. Based on the information in the data store, an execution framework platform may calculate a hash value for each operation node by including all recursive dependencies using ancestor node hash values and current node parameters. The platform may then compare each computed hash value with a previously computed hash value associated with a prior execution of a prior version of the pipeline. Operation nodes that have an unchanged hash value may be tagged “idle.” Operation nodes that have a changed hash value may be tagged “train and apply” or “apply” based on current node parameters (and an “apply” tag may propagate backwards through the pipeline to ancestor nodes). The platform may then ignore the operation nodes tagged “idle” when creating a physical execution plan to be provided to a target platform.