Tuning context-aware rule engine for anomaly detection
The technology disclosed relates to building ensemble analytic rules for reusable operators and tuning an operations monitoring system. In particular, it relates to analyzing a metric stream by applying an ensemble analytical rule. After analysis of the metric stream by applying the ensemble analytical rule, quantized results are fed back for expert analysis. Then, one or more type I or type II errors are identified in the quantized results, and one or more of the parameters of the operators are automatically adjusted to correct the identified errors. The metric stream is further analyzed by applying the ensemble analytical rule with the automatically adjusted parameters.
1. A system for building an ensemble analytical rule using reusable operators, including:
a storage that stores rules;
one or more processors coupled to the storage;
a rule instantiator including instructions that when executed by the one or more processors connects an ensemble analytical rule to a metric stream of performance information of a network selected for metrics of the metric stream based upon a metric class received from a metric classifier UI;
a rule engine including instructions that when executed by the one or more processors analyzes the metric stream by applying the ensemble analytical rule to process a time series of metrics through multiple stages of reusable operators, quantizes one or more results from the ensemble analytical rule into comparison values including at least normal and non-normal values; and
an alert scorer and/or an alert tuner including instructions that when executed by the one or more processors provides the quantized results for expert analysis, receives identifications of one or more errors in the quantized results, and automatically adjusts one or more parameters of the reusable operators to correct the identified errors; and
wherein the rule engine further analyzes the metric stream by applying the ensemble analytical rule with the one or more parameters as automatically adjusted.
2. The system of claim 1 , further implementing actions transmitting an alert to a human operator when non-normal comparison values are repeatedly detected over a predetermined time.
3. The system of claim 1 , further implementing actions transmitting an alert to a human operator when repeated normal comparison values are repeatedly detected over a predetermined time.
4. The system of claim 1 , wherein at least one of the reusable operators applies two or more inputs selected from a set consisting of a metric stream and an output stream of another operator.
5. The system of claim 1 , wherein two or more operator determined parameters of two or more of the reusable operators in the multiple stages are initially determined from a user entry or a user acceptance of default values.
6. The system of claim 1 , wherein operators are organized into a decision tree rooted by a single operator output.
7. The system of claim 6 , further implementing actions of estimating using the decision tree, a severity of anomalous data points in the metric stream.
8. The system of claim 1 , wherein expert analysis includes application of one or more domain rules that incorporate specific and targeted expert domain knowledge optimized according to metric class of performance metrics.
9. The system of claim 1 , wherein the metric stream measures latency of a first network computing device responding to a message from a second network computing device.
10. A method of building an ensemble analytical rule using reusable operators, the method including:
connecting, by a rule instantiator, an ensemble analytical rule to a metric stream of performance information of a network selected for metrics of the metric stream based upon a metric class received from a metric classifier UI;
analyzing, by a rule engine, the metric stream by applying the ensemble analytical rule to process a time series of metrics through multiple stages of reusable operators;
quantizing one or more results from the ensemble analytical rule into comparison values including at least normal and non-normal values;
providing the quantized results for expert analysis;
receiving identifications of one or more errors in the quantized results; and
automatically adjusting one or more parameters of the reusable operators to correct the identified errors;
further analyzing the metric stream by applying the ensemble analytical rule with the one or more parameters as automatically adjusted.
11. The method of claim 10 , further including transmitting an alert to a human operator when non-normal comparison values are repeatedly detected over a predetermined time.
12. The method of claim 10 , further including transmitting an alert to a human operator when repeated normal comparison values are repeatedly detected over a predetermined time.
13. The method of claim 10 , wherein at least one of the reusable operators applies two or more inputs selected from a set consisting of a metric stream and an output stream of another operator.
14. The method of claim 10 , wherein two or more operator determined parameters of two or more of the reusable operators in the multiple stages are initially determined from a user entry or a user acceptance of default values.
15. The method of claim 10 , wherein operators are organized into a decision tree rooted by a single operator output.
16. The method of claim 15 , further implementing actions of estimating using the decision tree, a severity of anomalous data points in the metric stream.
17. The method of claim 10 , wherein expert analysis includes application of one or more domain rules that incorporate specific and targeted expert domain knowledge optimized according to metric class of performance metrics.
18. The method of claim 10 , wherein the metric stream measures latency of a first network computing device responding to a message from a second network computing device.
19. A non-transitory computer readable storage medium impressed with computer program instructions to build an ensemble analytical rule using reusable operators, the computer program instructions, when executed on one or more processors, implement a method comprising:
connecting, by a rule instantiator, an ensemble analytical rule to a metric stream of performance information of a network selected for metrics of the metric stream based upon a metric class received from a metric classifier UI;
analyzing, by a rule engine, the metric stream by applying the ensemble analytical rule to process a time series of metrics through multiple stages of reusable operators;
quantizing one or more results from the ensemble analytical rule into comparison values including at least normal and non-normal values;
providing the quantized results for expert analysis;
receiving identifications of one or more errors in the quantized results; and
automatically adjusting one or more parameters of the reusable operators to correct the identified errors;
further analyzing the metric stream by applying the ensemble analytical rule with the one or more parameters as automatically adjusted.