Diagnosis of problem causes using factorization

A method for diagnosing problem causes in a complex computing environment by using a set of labeled data and factorization of a plurality of performance features, comprising:

• identifying a plurality of computing devices having device weighting parameters, and the plurality of performance indication features associated with the respective computing device having feature weighting parameters, wherein an instance of the plurality of performance indication features is associated with an individual computing device and is stored in a computer memory;
• applying a fault label for an ensemble grouping of at least one computing device;
• linking the plurality of performance indication features using the feature weighting parameter and the device weighting parameter by applying a bilinear model using an analysis device in communication with the complex computing network;
• estimating weighting values for the device weighting parameters and the feature weighting parameters for use in a statistical model;
• finding meaningful feature parameters in the statistical model that are likely to be responsible for device faults;
• and
• reporting the computing devices and performance features estimated to be responsible for faults associated with the fault label.

A system for diagnosing problem causes in a datacenter having computing devices using a set of labeled data and factorization of a plurality of performance indication features, comprising:

• a discovery module configured to identify the computing devices and associated device weighting parameters, and the plurality of performance indication features associated with a respective computing device having performance feature weighting parameters, wherein each computing device has an instance of the plurality of performance indication features;
• a fault module configured to apply a fault label for an ensemble grouping having at least one computing device;
• a factorization module configured to link the plurality of performance indication features using the feature weighting parameter and the device weighting parameter by applying a bilinear model using an analysis device having a processor;
• and
• a statistical module configured to 1) estimate weighting values for the device weighting parameters and the feature weighting parameters for use in a statistical model and 2) find meaningful feature parameters in the statistical model that have a probability of being responsible for device faults using the analysis device having the processor.