Comparing time series data using context-based similarity

A system, comprising:

• one or more computer processors;
• and
• a memory containing a program which when executed by the one or more computer processors performs an operation, the operation comprising:
• receiving a first time series and a second time series;
• identifying a first observation window defining a period of time in the first time series;
• identifying a second observation window defining a period of time in the second time series;
• generating a preceding similarity measure by comparing a first portion of the first time series with a first portion of the second time series, wherein the first portion of the first time series precedes the first observation window and the first portion of the second time series precedes the second observation window;
• and
• determining a window similarity between the first observation window and the second observation window based at least in part on the generated preceding similarity measure.

A computer program product comprising a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:

• identifying a first observation window defining a period of time in a first time series, wherein the first observation window is preceded by a first portion of the first time series;
• training a first neural network based on the first portion of the first time series and the first observation window;
• extracting a first set of weights from one or more middle layers of the first neural network;
• generating a first feature vector for the first observation window, based at least in part on the first set of weights;
• identifying a second observation window defining a period of time in a second time series, wherein the second observation window is preceded by a first portion of the second time series;
• determining a second feature vector associated with the second observation window, wherein the second feature vector is based at least in part on the first portion of the second time series;
• and
• determining a window similarity between the first observation window and the second observation window based at least in part on comparing the first feature vector and the second feature vector.