Machine-derived insights from time-series data

Abstract

Deriving insights from time series data may include receiving subject matter expert (SME) input that characterizes one or more aspects of a time series. A model template that specifies one or more components of the time series may be generated by translating the SME input using a rule-based translator. A machine learning model based on the model template may be a multi-layer neural network with one or more component definition layers, each configured to extract one of the one or more components from a time series data input corresponding to an instantiation of the time series.With respect to a decision generated by the machine learning model based on the time series data input, a component-wise contribution of each of the one or more components to the decision may be determined. An output may be generated, the output including the component-wise contribution of at least one of the one or more components.

Description

TECHNICAL FIELDThis disclosure relates to computer processing of time-series data, and more particularly to automatically generating machine learning models capable of processing and deriving findings from time-series data.BACKGROUNDA time series is a sequence of quantitative values or observations that are acquired at different times (e.g., at fixed intervals) or continuously over a period of time. As one type of stochastic process, a time series is often modeled as a sequence of random variables. Examples include signals generated by a thermal sensor during a chemical process, daily stock prices sold at the New York Exchange, and seasonal deposits in the Londoner city area. A time series may explicitly encode the concept of causality of a process. The temporal resolution of time-series data may enable efficient data mining of the causal structure of a process. Tasks involving time-series data include, for example, prediction, abnormality detection, and classification.The analysis of a time series can be carried out on the basis of the statistical relationship of the time series data. The statistical relationship may be determined using autoregressive or correlated metrics to measure whether a time-series value at a particular time is likely to be related to a value at another time, and if so, to what extent it is, or whether the data is fully random. An alternative time-series analysis approach determines the attributes of a time series by decomposing the time series into the frequency spectrum. Time-series analysis has many applications and is used in one form or another in virtually all fields of applied science, technology and economic technology.SUMMARYIn one or more embodiments, a method for deriving findings from time-series data may include receiving a Subject Matter Expert (SME) input, the SME input characterizing one or more aspects of a time-series. The method may include generating a model template by translating the SME input using a rule-based translator, the model template defining one or more components of the time series. The method may include generating a machine learning model configured based on the model template. The machine learning model may be configured as a multi-layer neural network having one or more component definition layers. Each component definition layer may be configured to extract one of the one or more components from a time-series data input corresponding to an instantiation of the time-series. With respect to a decision generated by the machine learning model based on a time-series data input, the method may include determining a component-related contribution of each of the one or more components to the decision. The method may include outputting the component-related contribution from at least one of the one or more components.In one or more embodiments, a system for deriving findings from time-series data includes one or more processors configured to initiate operations. The operations may include receiving a respective SME input, the SME input characterizing one or more aspects of a time series. The operations may include generating a model template by translating the SME input using a rule-based translator, the model template defining one or more components of the time series. The operations may include generating a machine learning model configured based on the model template. The machine learning model may be configured as a multi-layer neural network having one or more component definition layers. Each component definition layer may be configured to extract one of the one or more components from a time-series data input corresponding to an instantiation of the time-series. With respect to a decision generated by the machine learning model based on a time-series data input, the operations may include determining a component-related contribution of each of the one or more components to the decision. The operations may include outputting the component-related contribution from at least one of the one or more components.In one or more embodiments, a computer program product includes one or more computer readable storage media having instructions stored thereon. The instructions are executable by a processor to initiate operations. The operations may include receiving a respective SME input, the SME input characterizing one or more aspects of a time series. The operations may include generating a model template by translating the SME input using a rule-based translator, the model template defining one or more components of the time series. The operations may include generating a machine learning model configured based on the model template. The machine learning model may be configured as a multi-layer neural network having one or more component definition layers. Each component definition layer may be configured to extract one of the one or more components from a time-series data input corresponding to an instantiation of the time-series. With respect to a decision generated by the machine learning model based on a time-series data input, the operations may include determining a component-related contribution of each of the one or more components to the decision. The operations may include outputting the component-related contribution from at least one of the one or more components.This summary is provided merely to introduce certain concepts, and not to identify key or essential features of the claimed subject matter. Other features of the inventive arrangements will be apparent from the accompanying drawings and from the following detailed description.BRIEF DESCRIPTION OF THE DRAWINGSThe inventive arrangements are illustrated by way of example in the accompanying drawings. However, the drawings should not be construed as limiting the inventive arrangements solely to the implementations shown. Various aspects and advantages will become apparent from review of the following detailed description and with reference to the drawings. FIG. 1 illustrates an example system for deriving findings from time-series data. FIG. 2 illustrates an example method of using the system of FIG. 1. FIG. 3 illustrates certain operational aspects of the system of FIG.. 1 FIGS. 4A and 4B illustrate certain other operative aspects of the system of FIG. 1. FIG. 5 illustrates an example prototype of a model template generated with the system of FIG. 1.FIGS. 6A-6C illustrate example model strategies implemented by the system of FIG. 1. FIG. 7 illustrates an example computing environment for implementing aspects of the system of FIG. 1.DETAILED DESCRIPTIONAlthough the disclosure ends up with claims defining novel features, the various features described in this disclosure will be better understood when the description is taken in conjunction with the drawings. The process(s), machine(s), article of manufacture / articles of manufacture, as well as any modifications thereof described in this disclosure, are provided for purposes of illustration. Specific structural and functional details described in this disclosure are not to be considered as limiting, but merely as a basis for the claims and as a representative basis for teaching one skilled in the art how to utilize the described features in virtually any structure that is suitably embodied in detail. Furthermore, the terms and phrases used in this disclosure are not intended to be limiting, but are intended to provide an comprehensible description of the described features.This disclosure relates to computer processing of time-series data, and more particularly to using computer-implemented machine learning to derive directed findings from time-series data. A time series may consist of different components. If the time series has a constant variance, the time series may be modeled as a sum of the components. As the variance of the time series changes over time, the time series may be modeled as a product of the components. A time series component may include the entire time series or certain portions thereof. Many time-series models are structured using four components, for example - trend, cyclicity, seasonality and randomness.A time series comprising decomposable components may be modeled as a data generating process. For example, the Autoregressive Moving Average (ARMA) process has a structure that is a combination of an Autoregressive (AR) component and a Sliding (MA) component. The AR component is obtained by regression of a variable to its own delayed (past) values. The MA component models the difference between the values predicted by the model and the actual values as a linear combination of current and past errors. The Autoregressive Integrated Moving Average (ARIMA) process supplements the ARMA process by difference formation. Based on the difference formation, a stationarity can be introduced into an otherwise non-stationary process. The design of models such as generalized autoregressive conditional heteroskedasticity (GARCH) is specifically aimed at estimating changes in variance of time series data over time (heteroskedasticity). Exponential smoothing of the moving averages may be used to weight the time-series data with exponentially decreasing weights. Various other structures may be used to model processes by which time-series data is generated.A common concept of statistical analysis is causal inference. Causal inference on time-series data relates in particular to the question as to whether a particular event or phenomenon causes an observable change in values of the time-series data. For example, aggregate daily stock prices cause fluctuations in trade volume? For example, Surface Temperatures of Pacific Ocean influence Sardine Capture Amount? Various concepts of time-series causality have been proposed, wherein the Granger causality was one of the first. Causality applies to autoregressive and spectral representations of time series. Causality assumes that current data is not affected by future data still to be generated, and may be applied as a structural constraint to a time series.Knowledge from which a suitable structure for modelling a time series could result is often known only to an expert (SME). For example, a cardiologist may know that a time-series representation of a patient's ECG will typically include a P-wave, a QRS complex, and a T-wave occurring in a predetermined sequence. For example, a business scientist may know that a particular exogenous shock (e.g., an interruption in oil supply) is likely to be followed over time by a cyclical downslope in a time series of the gross inland product (BIP) of a country. SME knowledge may be structural, for example, knowing that the P wave, QRS complex and T wave occur sequentially and that the QRS wave is steeper than both the P and T waves. SME knowledge may also be behavioral, for example knowing that a discount in the price of a product is likely to precede an increase in sales figures of the product or that a change in the Hurst exponent (which measures the long term "memory" of a time series) precedes a change in market sendite of an asset.Despite the findings provided by SME knowledge, it is often difficult to integrate this knowledge into existing time-series models. One difficulty encountered thereby concerns the problem of model selection. Although SME knowledge is informative, it does not usually allow a particular model to be selected and does not provide any information as to which components in which arrangement should be used to build the model. In addition, a model may not sufficiently encode all model assumptions resulting from SME knowledge.In addition, SME knowledge is often qualitative in nature. SME knowledge can accordingly provide little or no insights as to which of several components make the greatest contribution to a decision of the time series model (e.g. regression, classification, anomaly detection).The systems, methods, and computer program products disclosed herein are capable of generating an automated framework for deriving findings from time-series data using domain knowledge. As defined herein, "insight" means identifying one or more components of a time-series model and determining the contribution of each identified component to the decision or result of the time-series model. The decision may be a prognosis, classification, anomaly detection, or other model decision generated based on the specific values of a particular time series. In this context, "component" as defined herein means a factor that affects the observed values of a time series. A component may thus correspond to an event or a physical phenomenon. A time series is characterized by the components of the time series. An ECG output is, for example, a time series characterized by three components - P-wave, QRS complex and T-wave - corresponding to physical phenomena associated with a patient's cardiac activity. For example, a time series of a country's BIP is typically characterized by a longer term trend component associated with economic growth and cyclic components associated with business cycles. A company's sales may also have a trend component as well as components associated with sales highs and lows that occur seasonal or in response to specific events such as flash sales.One aspect of using domain knowledge in accordance with the inventive arrangements disclosed herein is the automated generation of a knowledge-based model template generated based on an SME input. In certain arrangements, the model template is generated by computer-implemented translation of an SME input. The model template may specify a modeling strategy for modeling a particular time series. The model template may specify the structural components (e.g., trend or cycle) of the time series based on an SME input. The model template may define a causal structure, e.g., a predetermined event that causes a spike or a decrease in time-triggered values of the time series.Accordingly, another aspect of the inventive arrangements is the use of the model template to generate a time series according to the template defined components and the component structure. In certain arrangements, the model template defines a relative position of each of the components. Thus, according to the inventive arrangements, a time-series model can be automatically realized with the components structured according to the template generated in the translation of the SME input. The structure may include, for example, the relative positioning of the components in a sequence. For example, in conjunction with anomaly detection with respect to time series generated by an ECG, the model template may specify that a normal component structure is an ordered sequence in which a P-wave is followed by a QRS complex, which in turn is followed by a T-wave. The structure may include a causality constraint. Based on the SME input, the template may specify, for example, with respect to a business's translation time series, that an advertisement event precedes a translation jump.Another aspect of the inventive arrangements is the generation of a time-series machine learning model, the generation based on the model template. The machine learning model may be a deep learning model in certain arrangements. The machine learning model may be, for example, a recurrent neural network, a convolutional neural network, a generalized additive model (GAM), or another deep learning model.Another aspect of the inventive arrangements is a computer generated estimate of the contribution of each component to a decision (e.g., prediction, classification, anomaly detection) of the time-series model. In connection with ECG anomaly detection, a system implemented according to an inventive arrangement may be configured, for example, to determine which signal component of the time series-P-wave, QRS complex or T-wave-most likely contributes to the detection of an anomaly. The arrangement can thus provide an interpretable classifier that can identify which component is likely to result in a correct diagnosis.For example, in the context of predicting, the system may be configured to generate a prediction of economic activity (e.g., BIP) using a time-series model whose components are decomposed into structural events that include cyclic spikes, seasonal swings, and recent production jumps in response to open-loop incentives. Accordingly, the arrangement may provide an interpretable prediction device. The interpretable predictionr may identify spectral components (e.g., trend, cyclicity) or regression components that produce a time-specific prediction.In still other arrangements, different models may be generated based on different machine generated templates based on SME inputs. According to certain novel arrangements, a system may generate a contrasting score of the respective models. The contrasting score may indicate the respective prediction accuracy of each of the different models, as well as which components of each model were most important in generating a decision of each model with respect to a given set of time-series data.Further aspects of the inventive aspects described within this disclosure will be described in more detail with reference to the following figures. For simplicity and clarity of illustration, the elements shown in the figures have not necessarily been drawn to scale. For example, for clarity, the dimensions of some elements may be exaggerated relative to other elements. Further, where considered appropriate, reference numerals may be repeated in the figures to indicate analogous or corresponding features.Referring initially to FIGS. 1 and 2, an example system 100 for deriving findings from time-series data (system) and methodology 200 for implementing certain operational aspects of the disclosure are illustrated. The system 100 is operatively capable of generating a template based on a particular received knowledge-based SME input. The SME input may specify components that characterize a time series. Using the template, the system 100 is capable of generating one or more machine learning models that are created based on the components specified by the SME input. Each machine learning model generated by the system 100 thus provides a model representation of a time series that includes the specified components. In certain arrangements, the template may also include a user-specified modeling strategy. The SME input may thus specify a strategy for modeling the time series as well as the components of the time series. The components define the structure of the model. The modeling strategy determines how the model is presented, i.e., the architecture of the machine learning model. The modeling strategy specified by the template may specify a known architecture (e.g., a recurrent neural network, a one-dimensional convolutional neural network) or a user-adjusted architecture. Accordingly, in some arrangements, one or more keywords indicative of a modeling strategy may be associated with a given machine learning architecture. In other arrangements, one or more keywords may be associated with a specification that specifies particular attributes of a user-customized machine learning architecture (e.g., a number of hidden layers, activation functions, a gradient descent criterion function, a stop criterion function).The system 100 is not only capable of generating the machine learning model(s) based on the components specific to the SME input, but is also capable of deriving findings on the time series based on the time series data input to the machine learning model(s) after the generation. The time-series data input is data corresponding to instantiation of a time series. For example, the output of a patient's ECG corresponds to an instantiation of a time series that includes a P-wave component, a QRS complex component, and a T-wave component. By executing the machine learning model / models for a particular set of time-series data, the system 100 may determine the contribution each component makes to the decision (e.g., prognosis, anomaly detection, classification) of the machine learning model. In certain arrangements, the system 100 may determine on a predetermined quantitative scale (e.g., between 0 and 1) which components contribute a contribution that is above a predetermined threshold (e.g., 0.5) and which components do not. In conjunction with ECG output, knowing which component is most likely to indicate heart abnormality, for example, may help a medical technician make a diagnosis faster. In some cases, for example, reconfiguring a machine learning model by retraining components that have been recognized as meaningful and discarding components that are not the case may improve the prediction accuracy of the model and presumably reduces the processing load for the resources of a computer system.With continued reference to FIGS. 1 and 2, in certain arrangements, the system 100 may include a translator 102, a machine learning model generator 104, and an determiner 106. The translator 102, the machine learning model generator 104, and the determiner 106 may be implemented in hardware (e.g., dedicated hard-wired circuitry), software (e.g., program code executed by one or more processors), or a combination thereof. For example, in certain embodiments, the system 100 may be implemented in computer readable program instructions loaded into a computer, e.g., a computer 701 of a computing environment 700 (FIG. 7 ).In a block 202, the system 100 may receive an SME input 108 from a user. The SME input 108 may be received as a natural language input to a computer via a user interface, as illustrated by various elements of a user interface unit set 723 (FIG. 7 ), in some arrangements. The natural language input may be provided by text or spoken language, which may optionally be converted to text. The system 100 may be configured to perform natural language processing for the natural language input. The SME input 108 may be extracted from the natural language input by the system 100. The SME input 108 may characterize aspects of a predetermined time series. For example, the SME input 108 may characterize components (e.g., P-wave, QRS complex, T-wave) of a time series of electrode-generated signals recorded by an ECG or a time series of interest sets represented by computer encoded data retrieved online from a central bank (e.g., the U.S. Federal Reserve Note bank). In some arrangements, when performing natural language processing, the system 100 parses a natural language input (e.g., using a parser such as the Berkeley neural parser) to identify word types of a natural language input and extract therefrom the SME input 108 having particular keywords. As defined herein, "extract" means creating data structures that specify the keywords such that the keywords exist independently and separately from natural language input. The keywords may be encoded as a vector, a matrix, or a higher-dimensional tensor, for example.In a block 204, the translator 102 of the system 100 may generate a model template 110 by translating the SME input 108. In some arrangements, the translator 102 is configured to translate the SME input using a rule-based translation. With the rule-based translation, the translator 102 identifies one or more predefined keywords (e.g., domain-specific keywords) or one or more phrases included in the SME input 108. The rule set 112 associates each keyword and phrase from a complete set of predefined keywords and phrases with specific components of a time series. For example, the translator 102 may associate domain-specific keywords with model-specific keywords (e.g., trend, wave, cycle) corresponding to components characterizing the time series. For example, a component may characterize the structure of the time series (e.g., trend, cyclicity, seasonality). For example, a component may characterize a causality (e.g., a causal structure such as flash sales that cause a spike in the sold lots). Based on specific keywords and phrases identified in the SME input 108, the translator 102 generates the model template 110. The model template 110 may thus include the specific components of a time series that the rule set 112 associates with the keywords and / or phrases of the SME input 108.Thus, the translation of the time series characterizing SME input 108 in the block 204 may provide structural information to create a model representation of the time series. Structural information may indicate, for example, whether the model should be created as a stationary time series, whether the model variance is constant, whether the model has a seasonality, and, if so, to what extent and with what periodicity. For example, the SME input 108 may indicate a causal structure. That is, whether certain phenomena or events detected by the time series precede or follow others. By translating the SME input 108, the translator 102 generates the model template 110 to include the structural information and, optionally, a causal structure. For example, in creating a machine learning model from template 110 as described above, the structural information may set a relative order of temporal corrections (e.g., time window-related predefined units for automatic detection mode, thresholds). Accordingly, a causal structure may be used to identify functional dependencies of a machine learning model (e.g., an explicit causal structure may provide a slim model). These and other types of structure and causality characterizing a time series are collectively referred to as components of the time series.In a block 206, the machine learning model generator 104 may generate the machine learning model 114 based on the components extracted from the SME input 108. The machine learning model 114 is generated according to the template 110 that includes the components extracted from the SME input 108. The machine learning model 114 is a model representation of a time series generated according to the model template 110 and thus created based on the SME input 108. The machine learning model 114 models the time series characterized by the components extracted from the SME input 108 and translated into the model template 110. The machine learning model 114 generated by the machine learning model generator 104 in the block 206 is thus created based on components extracted from the SME input 108 by the translator 102 in the block 204.In certain arrangements, the machine learning model 114 is configured as a neural network having one or more component definition layers based on the model template 110. Each of the one or more component definition layers may be uniquely configured to extract one of the one or more components from the time-series data input. In some arrangements, the component definition layer(s) may be stored in a library and retrieved therefrom based on the translator 102 matching a keyword to a stored component definition layer.In certain arrangements, the machine learning model 114 is a deep learning neural network that includes an input layer, an output layer, and one or more hidden layers, as illustrated in the examples of FIGS. 6A-6C. In some arrangements, the machine learning model 114 is generated by the machine learning model generator 104 as a recurrent neural network. As a recurrent neural network, the machine learning model 114 is specialized for processing a sequence of values (e.g., timed values of a time series) and includes one or more layers of recurrent neurons that receive a current input (e.g., a vector) along with an output of a previous layer in each step. In other arrangements, the machine learning model 114 is generated by the machine learning model generator 104 as a convolutional neural network (e.g., a 1-D CNN) characterized by sharing parameters.In still other arrangements, the machine learning model 114 may be generated by the machine learning model generator 104 to have other different or specialized characteristics in the block 206. As described further below, according to certain arrangements, the model template 110 includes a modeling strategy field that determines the specific characteristics. For example, the machine learning model 114 generated by the machine learning model generator 104 may include a wavelet converter. The wavelet converter may identify frequency components of a time series. For example, if the machine learning model 114 is configured as a predictive model, the frequency components may be used to generate predictions.After being generated at block 206, a machine learning model 114 is capable of making a decision in response to entering time-series data 116. The decision may be a prediction of a value, a classification of the time series from which the time series data 116 is derived (e.g., classifying the time series as an anomaly or not), or another decision, depending on the specific task for which the machine learning model 114 is configured to be performed by the model template 110.In a block 208, the determiner 106 derives one or more time-series experiences 118 from the output of the machine learning model 114 based on the time-series data 116. The time series awareness(s) 118 may include a component-related contribution that each of the one or more components of the time series make to the decision of the model. Accordingly, the system 100 not only creates the machine learning model 114 based on components extracted from the SME input, but can also determine how much each of the components contributes to the decision of the machine learning model 114. In a block 210, the system 100 outputs the time-series awareness(s) 118 to the user.In certain arrangements, the determiner 106 implements a backpropagation algorithm to determine the effect of each component on the decision of the machine learning model 114. Backward propagation is an efficient mechanism for calculating derivatives of a single target quantity (e.g., an error in pattern classification) with respect to a large set of input quantities (e.g., parameters) in any large system consisting of elementary subsystems or calculations represented by known differentiable functions. In implementing a backward propagation algorithm, determiner 106 calculates a gradient with respect to the parameters of machine learning model 114 by applying the rule-of-chain theory to an ordered set of partial derivatives. The values of the partial derivatives indicate the magnitudes of a unit change in each functional unit corresponding to a component of the machine learning model 114. The determiner 106 uses the values to determine the effect of each component on the decision (e.g., prognosis, regression, classification, anomaly detection) of the machine learning model 114.In certain arrangements, the determiner 106 is configured to implement the gradient-weighted class activation mapping (CAM) approach. Grade CAM is applicable to a wide variety of CNN models. Grade CAM may use gradients of a target concept that are incorporated into the last convolutional layer of a CNN to generate a coarse localization map, where the map highlights, for example, important regions of an image to predict the concept.The effect of a component on the decision of the machine learning model 114 is itself a time-series knowledge derived by the determiner 106 from the time-series data 116, and is among the ways in which the system 100 can evaluate the decision(s) made by the machine learning model 114. In various arrangements, the time-series awareness(s) 118 may include other, additional awareness. The time-series knowledge(s) 118 may / may provide, for example, an instance-based explanation of which data point or group of data points contributes to the decision of the machine learning model 114. The time-series awareness(s) 118 may / may vary depending on the particular decision-related task of the machine learning model 114. For example, when the machine learning model 114 is configured for anomaly detection, the time-series awareness(s) 118 derived by the determiner 106 may indicate which component or components substantially contribute / contribute to the anomaly. Knowing the component likely to contribute to the anomaly may assist a user in detecting the cause of the anomaly. In connection with classifying a time series generated by an ECG as abnormal, knowing whether the P wave, the QRS complex, or the T wave is the most likely component contributing to the abnormality may aid in individual diagnoses and assist researchers in discovering likely causes of the abnormality. Thus, as mentioned above, the system 100 not only generates the machine learning model 114 based on components specified by the SME input 108, but can also evaluate the model by identifying which components most contribute to the decision (e.g., prognosis, classification, anomaly detection) of the machine learning model 114.Another example of findings derived by the system 100 relate to exogenous variables as well as contributions of components. Because components of a time series may correspond to a specific event, when the machine learning model 114 is configured to generate a prognosis or prediction, for example, the time series awareness(s) 118 derived by the determiner 106 may identify an exogenous variable that substantially contributes to the prognosis or prediction. For example, when predicting or predicting the future BIP using the machine learning model 114, the determiner 106 may determine the likely impact of an exogenous variable, e.g., an "oil shock" (supply cut), based on the time-series data. Accordingly, when the machine learning model 114 is configured to predict a sales of a business, the determiner may identify a possible seasonal component that will likely affect the sales, which in turn may suggest a suitable marketing strategy.With additional reference to FIG. 3, certain operational aspects of the system 100 are schematically illustrated. Illustratively, the model template 110 obtained by translating the SME input 108 extracted from the natural language input defines component configurations 300 a, 300 band 300 n. The component configurations 300 a, 300 b, and 300 neach correspond to a particular component of a time series 302, as well as any functional relationships that a component has with respect to other components of the time series 302. The time series 302 is generated, for example, by one or more sensors (e.g., ECG electrodes) or acquired from one or more sources (e.g., tin rates merged by the federal reserve). The time series data 116 is extracted from the time series 302 and input to the machine learning model 114. The machine learning model 114 includes component definition layers that extract the component configurations 300 a, 300 b, and 300 n. Processing 304 ais performed by the machine learning model 114 with respect to the component configuration 300 a. Respective processings 304 band 304 nare performed with respect to the component configurations 300 band 300 n, respectively, by the machine learning model 114. The system 100 outputs the time-series knowledge(s) 118, which illustratively include model knowledge 306 a, 306 b, and 306 n. The model experiences 306 a, 306 b, and 306 nare generated by the processing 304 a, 304 b, and 304 n, respectively, performed by the machine learning model 114 with respect to the component configurations 300 a, 300 b, and 300 nas defined by the model template 110.Optionally, the system 100 may summarize the model experiences 306 a, 306 b, and 306 nand provide the user as a consolidated report 308. In other arrangements, the system 100 optionally generates multiple machine learning models, each configured based on a different model template. The system 100 may generate a contrasting explanation included in the consolidated report 308 that demonstrates differences between the machine learning models. Corresponding time-series knowledge provided as part of the contrasting explanation may indicate which aspects of a time-series emphasize each machine learning model, which model provides greater prediction accuracy, and / or other such time-series knowledge, for example.With additional reference to FIGS. 4A and 4B, certain operational features of the translation performed in block 204 by translator 102 are illustrated in the specific context of an ECG time series. Illustratively, in FIG. 4A, translator 102 identifies domain-specific keywords (e.g., "spikes") and phrases (e.g., "occurs... ", "steeper than", "fixed signal basis", and "regular distance") shown within a bold SME input 400. Certain domain-specific terms (e.g., "P-wave", "QRS complex", "T-wave", and "at least 2,500 observations") also occur within the SME input 400. Translator 102 translates SME input 400 into certain illustrative portions of template 110 that are expressed as processor executable code that can be used to generate a time-series model (described below) whose components and structural arrangement correspond to SME input 400. The template 110, expressed as processor executable code, is used by the system 100 to create the machine learning model 114. Because it is created according to the components specified by the template 110, the machine learning model 114 models a time series including a P-wave, a QRS complex, and a T-wave.In FIGS. 4A and 4B, machine learning model 114 is illustratively configured for the task of abnormality detection based on classifying a time series generated by an ECG monitor used to detect a patient's cardiac activity. Entering a specific set of time-series data generated by the ECG into the machine learning model 114 illustratively generates a result 402 shown in FIG. 4B. Based on the data input, the machine learning model 114 assigns the time series to the class "MI" indicating the anomaly "myocardial infarction.". The output "score" indicates a confidence value of the classification by the machine learning model 114. The output "component" indicates that the QRS complex makes a significantly greater contribution (0.75) to the decision (abnormality detected) of the machine learning model 110 than either the P wave or the T wave (0.05 or 0.12, respectively). Thus, FIGS. 4A and 4B illustrate automated creation of the machine learning model 114 by the system 100 based on the components (P-wave, QRS complex, and T-wave) of the SME input 108 specified by the template 110, as well as derivation of findings from time-series data by executing the machine learning model 114 with the data. FIGS. 4A and 4B further illustrate evaluating the importance of the components for the decision of the machine learning model 114 by the system 100. Accordingly, as mentioned above, another example of deriving findings from time-series data by the system 100 determines which of multiple components are most likely to determine classification by the machine learning model 114.According to various arrangements, template 110 is capable of defining multiple different aspects of a model representation of a particular time series. The template 110 may set the model components of the time-series model. The components of the time series model are different patterns. As mentioned above, components may include a component defined for an entire time domain, e.g., a trend and a seasonality spanning an entire time-based window of the row. Other components relate to specific events that occur during a particular time period or a limited portion of a complete time series. An example associated with a time series that detects the sale of a product is a flash sale that lasts for a limited time (time period) and is associated with an increase in the sales figures of the product. Template specific components may also correspond to predefined metrics that set constraints for a time series, e.g., autocorrelation, constant variance, linear trend, stationarity, and the like. As described below, each component (e.g., trend, temporal segment) may be stored in a separate presentation channel of the time-series model.As mentioned above, based on the SME input 108, the template 110 may specify a modeling strategy to generate a model representation of the particular time series. In certain arrangements, the modeling strategy for template 110 must be obligatory set. The system 100 may be configured to prompt a user for input if the SME input 108 does not include a modeling strategy. The system 100 uses the modeling strategy to structure a base model. Depending on the other SME input, different versions of the time-series model can be created by modifying the basic structure. For example, the modeling strategy may be used by the system 100 to subdivide the components of the time-series model into temporal segments. Each part can be modeled separately by modifying the structure of the base model. If the modeling structure is obligatory, the system 100 may automatically set a strategy if the user does not. In some arrangements, ablation is a standard modeling strategy.FIG. 5 illustrates an example prototype 500 of various templates that may be generated by the translator 102. Illustratively, the template may include three elements that may determine the configuration of a model representation of a time series. Illustratively, the three elements are structural knowledge, relational knowledge, and model selection as illustrated in the first column of prototype 500. The structural knowledge corresponds to the components that make up the model. By implementing the template as a machine readable data structure, the data structure may include separate fields that optionally specify the type of component, component name, component filter, and component order, as illustrated in the second column of prototype 500. Each of the fields may set one or more parameters as illustrated in column 3 and described in column 4. Relational knowledge illustratively includes a field that defines a component dependency between two or more of the model components. The component dependency indicates that there is a statistically significant probability that a component is likely to precede or follow another component. Each component that has a structural relationship to another component is characterized by a parameter that indicates whether the component is superordinate (probably preceding another component) or subordinate (probably following another component). The model selection illustratively also corresponds to a single field identified by a strategy field illustrated in the second column of prototype 500. In the third column of prototype 500, three different modeling strategies (labeled strategy parameters) are illustrated. The modeling strategies-time precession, independent component and ablation-are only for illustrative purposes. Various other strategies may be set up in a template generated by the translator 102.FIGS. 6A-6C illustrate three example modeling strategies. FIG. 6A illustrates an example modeling strategy 600. A machine learning model created according to modeling strategy 600 is capable of detecting a temporal precession by encoding the model with exponential weights. In conjunction with time series, data points that are closer together in time often affect prediction more strongly than data points that are farther apart in time. Thus, if η(y t) is a numerical measure of the prediction contribution, then η(y t) > η(y t-1) holds for all t > t - 1. The machine learning model 114 implemented according to the modeling strategy 600 includes one or more component definition layers 602, each capable of recognizing a component specified by the model template 110. The components are extracted from time-series data 604 and processed using one or more hidden layers 606. Illustratively, individual components 608a, 608b, 608c, and 608d are respectively communicated to functions 610a, 610b, 610c, and 610d and weighted by exponentially decreasing weighting functions h 1, h 2, h 3 and h 4.FIG. 6B illustrates an example modeling strategy 612. The modeling strategy 612 creates a machine learning model with a regression model architecture that extracts approximate representations of an independent basis function. The model is capable of automatically recognizing independent components, i.e. components which are (at least) non-overlapping (minimum) segments of the time series. The structure is analogous to a MODWT (Maximum Overlap Discrete Wavelet Transform). The machine learning model 114 implemented according to the modeling strategy 612 includes one or more component definition layers 614 each capable of recognizing a component specified by the model template 110. The components are extracted from time-series data 616 and processed by one or more hidden layers 618, generating orthogonal component extraction 620 that separates non-overlapping components (time-based segments). The individual components are processed by the functions g 1, g 2, g 3 and g 4 and by the independent basic functions h 1, h 2, h 3 and h 4.FIG. 6C illustrates an example modeling strategy 622. The modeling strategy 622 builds a machine learning model with a wavelet prediction architecture. The model is created so that very frequently occurring components of the time series affect the more immediate development (e.g., cyclicity) of the series, while less frequently occurring components (e.g., trend) affect the series over longer periods of time. The machine learning model architecture is capable of generating a prediction using a prediction summation and prediction aggregation unit. The machine learning model 114 implemented according to the modeling strategy 622 includes one or more component definition layers 624, each capable of recognizing a component specified by the model template 110. The components are extracted from time-series data 626 and processed by one or more hidden layers 628. Frequency components 630 a, 630 b, 630 c, and 630 dare processed by predictors 632 a, 632 b, 632 c, and 632 d, respectively, whose outputs are aggregated to generate a prediction 634. In some arrangements, predictors 632 a, 632 b, 632 c, and 632 dare implemented as a GAM. The GAM extends linear regression by allowing a linear model to learn non-linear relationships between variables and generating a prediction by summing individual functions of each variable. In other arrangements, predictors 632 a, 632 b, 632 c, and 632 dexecute time-series prediction using neural base expansion (N-BEATS). In various arrangements, the modeling strategy 622 may use other types of predictors to build the machine learning model with a wavelet prediction architecture. Modeling strategies 600, 612, and 622 are also for illustrative purposes only. The model template 110 may define other modeling strategies in other arrangements.Various aspects of the present disclosure are described by descriptive text, block diagrams of computer systems, and / or block diagrams of machine logic included in computer program product (CPP) embodiments. With respect to any flow charts, depending on the technology involved, the operations may be performed in a different order than shown in a particular flow chart. For example, depending also on the technology involved, two operations shown in successive flowchart blocks may be performed in the reverse order or as a single, integrated step, simultaneously or in a manner that at least partially overlap in time.A computer program product ("CPP" or "CPP") embodiment is a term used in the present disclosure to describe any set of one or more storage media (also referred to as "media") that are commonly included in a set of one or more storage devices that commonly include machine readable code corresponding to instructions and / or data for performing computer operations described in a particular CPP claim. A "storage unit" is any tangible unit that can contain and store instructions for use by a computer processor. The computer readable storage medium may include, but is not limited to, an electronic storage medium, a magnetic storage medium, an optical storage medium, an electromagnetic storage medium, a semiconductor storage medium, a mechanical storage medium, or any suitable combination thereof. Some known types of memory devices that contain these media include: a floppy disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a static random access memory (SRAM), a compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, and a flexible disk, a mechanically encoded unit (e.g., punch cards or pits / bumps on a larger surface of a media) or any suitable combination thereof. A computer readable storage medium, as the term is used in the present disclosure, is not intended to be storage in the form of transitory signals per se, e.g., radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide, light pulses passing through an optical fiber cable, electrical signals transmitted through a wire, and / or other transmission media. It will be appreciated by those skilled in the art that data is typically moved during normal operation of a storage unit at certain times, such as during access, defragmentation or upon performance of a clean-up function, but the storage unit thereby becomes non-transitory, as the data is non-transitory while being stored.The computing environment 700 includes an example of an environment for executing at least a portion of the computer code involved in performing the inventive methods, e.g., deriving findings from time-series data illustrated in a block 750. The inventive methods performed with the computer code of block 750 may include implementing procedures for generating a model template from an SME input characterizing a time series, generating one or more machine learning models representing the time series, and outputting findings derived from the time series using the one or more machine learning models described herein in connection with the system 100 and the methodology 200. In addition to the block 750, the computing environment 700 includes, for example, a computer 701, a wide area network (WAN) 702, an end user device (EUD) 703, a remote server 704, a public cloud 705, and a private cloud 706. In this embodiment, the computer 701 includes a processor set 710 (e.g., processing circuitry 720 and cache 721), a communication structure 711, volatile memory 712, persistent storage 713 (e.g., operating system 722 and block 750, as identified above), a peripheral unit set 714 (e.g., user interface unit set 723 (UI), memory 724 and internet of things (IoT) sensors set 725, and a network module 715. Remote server 704 includes remote database 730. The public cloud 705 includes a gateway 706, a cloud orchestration module 741, a host physical machine set 742, a virtual machine set 743, and a container set 744.The COMPUTER 701 may be in the form of a desktop computer, laptop computer, tablet computer, smart phone, smart watch or other wearable computer, mainframe computer, quantum computer, or any other form of computer or mobile device that is currently known or developed in the future and that is capable of executing a program, accessing a network, or retrieving a database such as the remote database 730. As is well known in computer technology, and depending on the technology, the performance of a computer-implemented method may be distributed among multiple computers and / or multiple locations. However, in this discussion of the computing environment 700, the detailed discussion will focus on a single computer, specifically the computer 701, to keep the discussion as simple as possible. The computer 701 may be located in a cloud although not shown in a cloud in FIG. 7. However, the computer 701 does not necessarily have to be located in a cloud, unless this is expressly described in this way.The PROCESSOR SET 710 includes one or more computer processors of any type presently known or developed in the future. The processing circuitry 720 may be distributed among multiple packages, for example, among multiple coordinated IC chips. Processing circuitry 720 may implement multiple processor threads and / or multiple processor cores. Cache 721 is memory residing in processor chip package(s) and is commonly used for data or code that should be available for fast access by the threads or cores executing in processor set 710(s). Cache memories are usually organized in multiple levels depending on the relative proximity to the processing circuitry. Alternatively, the cache for the processor set may be entirely or partially "off-chip.". In some computing environments, processor set 710 may be configured to process qubits and perform quantum computing.The computer readable program instructions are typically read into the computer 701 to cause a sequence of operations to be performed by the processor set 710 of the computer 701, thereby causing a computer-implemented method, such that the instructions performed thereby instantiate the methods performed in flowcharts and / or illustrative descriptions of computer-implemented methods included in this document (collectively referred to as "the inventive methods"). These computer readable program instructions are stored in various types of computer readable storage media, e.g., cache 721 and the other storage media discussed below. The program instructions and associated data are accessed by the processor set 710 to control and direct the inventive methods to be performed. In the computing environment 700, at least some of the instructions for performing the inventive methods may be stored in the persistent storage 713 at block 750.The DATA TRANSMISSION STRUCTURE 711 constitutes the signal line paths that allow the various components of the computer 701 to exchange data with each other. This structure usually consists of switches and electrically conductive paths, e.g. the switches and electrically conductive paths forming buses, bridges, physical input / output ports and the like. Other types of signal transmission paths may also be used, e.g., optical fiber data transmission paths and / or wireless data transmission paths.The VOLATILE MEMORY 712 is any type of volatile memory currently known or developed in the future. Examples thereof include a dynamic type random access memory (RAM) or a static type RAM. The volatile working memory is usually characterized by random access, although this does not necessarily have to be the case unless expressly described in this way. The volatile working memory 712 is located in a single packet in the computer 701 and is arranged inside the computer 701, wherein however, the volatile working memory can alternatively or additionally be distributed over a plurality of packets and / or can be located outside the computer 701.PERSISTENT MEMORY 713 is any type of non-volatile memory for computers that is currently known or will be developed in the future. The non-volatility of this memory means that the stored data is maintained regardless of whether the computer 701 and / or directly the persistent memory 713 are powered. The persistent memory 713 may be read only memory (ROM), but typically at least a portion of the persistent memory enables writing data, erasing data, and rewriting data. Certain common forms of persistent storage include magnetic disks and semiconductor memory devices. The operating system 722 may take several forms, such as various known, proprietary operating systems or open source POSITION (Portable Operating System Interface) type operating systems that use a kernel. The code contained in block 750 typically includes at least a portion of the computer code involved in performing the inventive methods.The PERIPHERAL UNIT SET 714 includes the set of peripheral units of the computer 701. Communication links between the peripherals and the other components of the computer 701 may be implemented in various ways, for example, as Bluetooth links, near-field communication (NFC) links, wired links (e.g., USB (Universal Serial Bus) type links), insertion type links (e.g., Secure Digital Card, SD)), connections over local communication networks, and even connections over wide area networks such as the Internet. In various embodiments, the UI unit set 723 may include components such as a display screen, a speaker, a microphone, wearable units (e.g., glasses and smart watches), a keyboard, a mouse, a printer, a touchpad, game controllers, and haptic units. The memory 724 is an external memory such as an external hard disk drive or a deployable memory such as an SD card. The memory 724 may be persistent and / or volatile. In some embodiments, the memory 724 may be in the form of a quantum computing memory unit for storing data in the form of qubits. In embodiments where the computer 701 requires a large amount of memory (e.g., when the computer 701 locally stores and manages a large database), that memory may then be provided by peripheral storage devices designed to store very large amounts of data, e.g., a storage area network (SAN) shared by numerous geographically distributed computers. The loT sensor set 725 consists of sensors that can be used in Internet of Things applications. For example, one sensor may be a thermometer and another sensor may be a motion detector.The NETWORK MODULE 715 is the collection of computer software, hardware, and firmware that allows the computer 701 to communicate with other computers via the WAN 702. The network module 715 may include hardware, e.g., modems or WLAN signal transceivers, software for packaging and / or unpackaging data for transmission over a communication network, and / or web browser software for transmitting data over the Internet. In some embodiments, network control and network forwarding functions of network module 715 are performed in the same physical hardware device. In other embodiments (e.g., embodiments using software defined networking (SDN)), the control and forwarding functions of network module 715 are performed in physically separate entities, such that the control functions may manage multiple different hardware network entities. Computer readable program instructions for performing the inventive methods may typically be downloaded to the computer 701 from an external computer or storage device via a network adapter card or network interface included in the network module 715.The WAN 702 is a wide area network (e.g., the Internet) capable of transmitting computer data over distances beyond the local frame by any technology for transmitting data currently known or developed in the future. In some embodiments, the WAN may be replaced and / or supplemented with local area networks (LANs) designed to transfer data between entities located in a local area perimeter, e.g., a WLAN network. The WAN and / or LANs may typically include computer hardware such as copper transmission cables, lightwave transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers, and edge servers.The END USER UNIT (EUD) 703 is a computer system used and controlled by an end user (e.g., a customer of a business operating the computer 701) and may take any of the forms discussed above in connection with the computer 701. The EUD 703 typically receives helpful and useful data from the operation of the computer 701. For example, in a hypothetical case where the computer 701 is designed to provide a recommendation to an end user, this recommendation is typically transmitted from the network module 715 of the computer 701 to the EUD 703 via the WAN 702. In this manner, the EUD 703 may display or otherwise present the recommendation to an end user. In some embodiments, the EUD 703 may be a client device, e.g., a thin client, a heavy client, a mainframe computer, a desktop computer, and so forth.The REMOTE SERVER 704 is any computer system that provides at least some data and / or functionality to the computer 701. Remote server 704 may be controlled and used by the same entity that operates computer 701. Remote server 704 stands for the machine(s) that collect / collect and store / store helpful and useful data for use by other computers, e.g., computer 701. In a hypothetical case where the computer 701 is designed and programmed to provide a recommendation based on historical data, this historical data may be provided to the computer 701 from, for example, a remote database 730 of the remote server 704.The PUBLIC CLOUD 705 is any computer system available for use by multiple entities that provides demand-controlled availability of computer system resources, particularly data storage (cloud storage) and computing power, without direct, active management by the user. Cloud computing typically relies on sharing resources to achieve coherence and scale effects. The direct and active management of the computing resources of the public cloud 705 is performed by the computer hardware and / or software of the cloud orchestration module 741. The computing resources provided by the public cloud 705 are typically realized by virtual computing environments executing on various computers that make up the computers of the host physical machine set 742, which is the collection of physical computers present in and / or available from the public cloud 705. The virtual computing environments (VCEs) are typically in the form of virtual machines from the virtual machine set 743 and / or containers from the container set 744. It will be appreciated that these VCEs may be stored as images and transmitted within and between the various host physical machines, either as images or after instantiation of the VCE. The cloud orchestration module 741 manages the transfer and storage of images, provides new instantiations of VCEs, and manages active instantiations of VCE provisions. Gateway 740 is the collection of computer software, hardware, and firmware that enables public cloud 705 to exchange data via WAN 702.Virtualized computing environments (VCEs) are explained in more detail below. VCEs may be stored as "images.". A new active instance of the VCE may be instantiated from the image. Two common types of VCEs are virtual machines and containers. A container is a VCE that uses operating system level virtualization. This relates to an operating system feature in which the kernel enables the presence of multiple isolated user domain instances, referred to as containers. These isolated user area instances typically behave like real computers from the point of view of programs executed therein. A computer program executing on a common operating system may use all resources of the computer in question, e.g. connected entities, files and folders, shared network areas, CPU power and quantifiable hardware capabilities. Programs executed within a container, however, may only use the contents of the container and entities assigned to that container, a feature known as containerization.The PRIVATE CLOUD 706 is similar to the public cloud 705 except that the computing resources are only available for use by a single enterprise. Although the private cloud 706 is described as being in communication with the WAN 702, in other embodiments, a private cloud may be completely disconnected from the Internet and only accessible via a local / private network. A hybrid cloud is a composition of multiple clouds of different types (e.g., of the private, community, or public cloud type) that are often implemented by different providers. Each of the plurality of clouds remains a separate and distinct entity, but the larger hybrid cloud architecture is held together by standardized or proprietary technology that enables orchestration, management, and / or data / application portability between the plurality of clouds involved. In this embodiment, both the public cloud 705 and the private cloud 706 are part of a larger hybrid cloud.The terminology used herein is for describing particular embodiments only and is not intended to be limiting. Nevertheless, some definitions are now set forth which are valid in this document.As defined herein, the singular forms "a / e / r," and "the / s" also include the plural forms unless the context clearly dictates otherwise.As defined herein, "another" means at least a second or more.As used herein, "at least one, one or more" and "and / or" are open ended terms that have both a conjunctival and disjunctive function, unless expressly stated otherwise. For example, each of the terms "at least one of A, B, and C", "at least one of A, B, or C", "one or more of A, B, and C", "one or more of A, B, or C", and "A, B, and / or C" means only A, only B, only C, A, and B, A, and C, B, and C, or A, B, and C.As defined herein, "automatic" means without user intervention.As defined herein, "includes," "including," "has," and / or "having" indicate the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.As defined herein, "when" means "responsive to" or "responsive to", depending on the context. Thus, the phrase "when determined" may be understood to mean "responsive to determining" or "responsive to determining", depending on the context. Accordingly, the phrase "when [a particular condition or event] is detected" may be understood to mean "after detection [of the particular condition or event]", or "in response to detection [of the particular condition or event]", or "responsive to detection [of the particular condition or event]", depending on the context.As used herein, "a single embodiment," "one embodiment," "in one or more embodiments," "in certain embodiments," or similar language means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment described in this disclosure. Thus, uses of the above-mentioned phrases and / or similar formulations used within this disclosure may refer to one and the same embodiment, but need not be so.As defined herein, phrases "responsive to" mean a response to an action or event. Thus, when a second action is performed "in response to" or "responsive to" a first action, there is a causal relationship between an occurrence of the first action and an occurrence of the second action. The phrases "responsive to" and "responsive to" indicate causal relationship.As defined herein, "user" means a human.The terms "first," "second," etc., may be used herein to describe various elements. These elements should not be understood as limited thereto, as these terms are used merely to distinguish one element from another unless otherwise stated or the context clearly dictates otherwise.The inventive arrangements disclosed herein have been presented for purposes of illustration and are not to be construed as exhaustive or limited to the specific arrangements. Those skilled in the art will appreciate that numerous modifications and variations are possible without departing from the scope of the invention as described. The terminology used herein was chosen to best explain the principles of the inventive arrangements, the practical application, or technical improvement over technologies found in the market, or to enable others skilled in the art to understand the inventive arrangements disclosed herein.

Claims

A computer-implemented method, comprising: receiving, via a computer user interface of a computer, a Subject Matter Expert, SME input, wherein the SME input characterizes one or more aspects of a time series; generating a model template by translating the SME input using a rule-based translator realized by the computer, wherein the model template defines one or more components of the time series; generating, by the computer, a machine learning model based on the model template, the machine learning model defining a multi-layer neural network having one or more component definition layers configured to extract the one or more components from a time-series data input corresponding to an instantiation of the time-series; determining, with respect to a decision generated by the machine learning model based on the time-series data input, a component-related contribution of each of the one or more components to the decision; and outputting, via the computer user interface, the component-related contribution of at least one of the one or more components.The computer-implemented method of claim 1, wherein generating the model template by translating the SME input includes associating domain-specific keywords extracted from the SME input with model-specific keywords identifying the one or more components.The computer-implemented method of claim 1, wherein the model template defines a modeling strategy, and wherein generating the machine learning model at least partially bases the modeling strategy.The computer-implemented method of claim 1, wherein the one or more components comprise a plurality of components, and wherein determining a component-related contribution for each of the plurality of components includes decomposing the time-series data input to identify each of the plurality of components.The computer-implemented method of claim 1, wherein the one or more components comprise a plurality of components, and wherein the model template defines a positioning of each of the plurality of components relative to each other within the time series.The computer-implemented method of claim 1, wherein the machine learning model is configured to generate a prognosis based on the time-series data input, and wherein the one or more time-series experiences identify a contribution of an exogenous factor.The computer-implemented method of claim 1, wherein the machine learning model is configured for anomaly detection, and wherein in response to detecting an anomaly based on the time-series data input, the one or more time-series experiences identify a component as a likely cause of the anomaly.The computer-implemented method of claim 1, wherein the model template comprises a plurality of model templates and the machine learning model comprises a plurality of machine learning models each configured based on a corresponding one of the plurality of model templates, and wherein the method further comprises generating a contrasting explanation of differences between the machine learning models.A system, comprising: a processor configured to initiate operations, comprising: receiving an expert, SME, input, the SME input characterizing one or more aspects of a time series; generating a model template by translating the SME input using a rule-based translator, the model template defining one or more components of the time series; generating a machine learning model based on the model template, the machine learning model defining a multilayer neural network having one or more component definition layers configured to extract the one or more components from a time series data input corresponding to instantiating the time series; determining, with respect to a decision generated by the machine learning model based on the time-series data input, a component-related contribution of each of the one or more components to the decision; and outputting the component-related contribution of at least one of the one or more components.The system of claim 9, wherein generating the model template by translating the SME input includes associating domain-specific keywords extracted from the SME input with model-specific keywords identifying the one or more components.The system of claim 9, wherein the model template defines a modeling strategy, and wherein generating the machine learning model at least partially bases the modeling strategy.The system of claim 9, wherein the one or more components comprise a plurality of components, and wherein determining a component-related contribution for each of the plurality of components includes decomposing the time-series data input to identify each of the plurality of components.The system of claim 9, wherein the one or more components comprise a plurality of components, and wherein the model template determines a positioning of each of the plurality of components relative to each other within the time series.A computer program product, the computer program product comprising: one or more computer readable storage media and program instructions stored collectively on the one or more computer readable storage media, the program instructions executable by a processor to cause the processor to initiate operations, including: receiving an expert, SME, input, the SME input characterizing one or more aspects of a time series; generating a model template by translating the SME input using a rule-based translator, the model template defining one or more components of the time series; generating a machine learning model based on the model template, the machine learning model defining a multi-layer neural network having one or more component definition layers configured to extract the one or more components from a time-series data input corresponding to an instantiation of the time series; determining, with respect to a decision generated by the machine learning model based on the time-series data input, a component-related contribution of each of the one or more components to the decision; and outputting the component-related contribution of at least one of the one or more components.The computer program product of claim 14, wherein generating the model template by translating the SME input includes associating domain-specific keywords extracted from the SME input with model-specific keywords identifying the one or more components.The computer program product of claim 14, wherein the model template defines a modeling strategy, and wherein generating the machine learning model at least partially bases the modeling strategy.The computer program product of claim 14, wherein the one or more components comprise a plurality of components, and wherein determining a component-related contribution for each of the plurality of components includes decomposing the time-series data input to identify each of the plurality of components.The computer program product of claim 14, wherein the one or more components comprise a plurality of components, and wherein the model template defines a positioning of each of the plurality of components relative to each other within the time series.The computer program product of claim 14, wherein the machine learning model is configured to generate a prognosis based on the time-series data input, and wherein the one or more time-series experiences identify a contribution of an exogenous factor.The computer program product of claim 14, wherein the machine learning model is configured for anomaly detection, and wherein in response to detecting an anomaly based on the time-series data input, the one or more time-series experiences identify a component as a likely cause of the anomaly.

Images