Feature Improvement: Rule mining

[ihayet]

Rules are associations between error messages and action patterns. Existing rules have been mined by capturing patterns in the semantics of the error messages. An important criterion of impact while mining rules is to ensure that a rule is as general as possible and that the rule addresses a substantial number of unresolved errors. The rolling expectation for this task is to continue monitoring unresolved error messages to mine emerging rules.

When writing a new rule, we try to associate an error class with the appropriate action class. An error class directly corresponds to an error type. If an error class already exists for a certain error type, we should insert the error message as a key of the mappings dictionary inside the error class. The corresponding value of this entry in the mappings dictionary should be a list of action classes. If an error class does not exist for a certain error type, we should ... (to be continued)