pm4rs / promi Goto Github PK

• look for unnecessary copies (I'm talking about you Observer)
• assess usage of concurrency for heavy operations
  • synced channels

Implementation of simple process mining algorithms & respective model structures. This includes

• directly follows graph (DFG), footprints (unweighted DFG)
• alpha / heuristic miner
• inductive miner

Give high level examples of how streaming can be done.

• counter
  Simply count elements in log, implemented as stream.
• stream stats
  Count elements in log and log components, implemented as handler.
• extension stats
  Loads extensions from registry that are declared in log, extracts respective attributes and summarizes them. Also implemented as handler.

There's not log/trace/event semantics defined by the XES standard. Instead, it introduces the extension concept which is intended to provide such semantics. On an abstract level, promi shall have a trait that defines such an extension representation. Further, it shall implement default extensions defined by the XES standard.

abstract

• validation
• statistics

implementation

• concept
• lifecycle
• organizational
• time
• semantic
• id
• cost

global extension registry
It would be neat to have a central place to look up all extension implementations available in order to apply abstract features automatically. E.g. statics would become automatically richer if further implementations are linked.

There's scenarios more common than other and users should be provided with high level interfaces saving them from writing the same code again and again. Such may cover

• XML-String -> Log
• .xes-File -> Log
• Log -> XML-String
• Log -> .xes-File
• load example

Partly, such functions exist for test modules. These shall be merged.

While other error cases cause a stream to crash in a very early stage already, e.g. while parsing, others won't. A validator will do a deeper inspection, including checks of

• presence of meta & element element order
• string based XML types
• globals
• extension specific

• token replay
• allignments

As the library grows, imports become very verbose even for small scripts. A prelude module could solve that by importing all the stuff average users want.

Sometimes one wants to assess a model with data different from that it was trained with. A modified Duplicator could perform such a split based on given criteria.

• frequent item sets
• association rules (FP-growth)

• application logging (log trait) to event stream
• trace factory: buffer events and aggregate them to traces
• random log / event stream factory
• declarative (e.g. yaml based) pipeline building
• more exotic targets such as embedded devices and web assembly (no std)
• advanced testing: compatibility of promi XES with OpenXES, PM4Py etc.
  • check against schema definition e.g. using xmllint

This should not be part of core promi but an separate script etc. that takes a dirty XES file and fixes some classes of recoverable errors such as string sanitation.

to be investigated

Potentially set up the same scenarios for promi and PM4Py.

A filter is a handler that owns condition closures the tell it whether a trace/event will be filtered out or not (logical disjunction). Trace conditions decide for a trace as a whole if it remains in the log while event conditions trim traces.

Multiple filters can be chained for logical conjunction.

Also, provide a factory that creates filter(s)/observer from CNF.