Keep case identifier as attribute when converting to a TraceLog

[fmannhardt]

It is my understanding while reading the source code that the attribute according to which a TraceLog is built from an EventLog cannot be retrieved after conversion. At least I did not see a way to retrieve it.

If you would keep it in some kind of attribute as part of the log object that would make it easier to convert between the R eventlog and the pm4py TraceLog. I guess it would also facilitate some other operations which rely on a trace being identifiable.

[Javert899]

Hi,

Thanks for signaling.

Actually, the change you suggested has been implemented in the 'develop' branch, in the pm4py.objects.conversion.log.versions.to_trace_log.apply method, changing "include_case_attributes=" to True from False

[fmannhardt]

Thanks. I assume you mean the changes done in 27577e1a446ebf43e807e34236703aadc171fdcc?

However, what I meant is that it is impossible to know which attributes was used as trace identifier or case_glue as you denote it in PM4PY after having obtained the TraceLog. There might be multiple case attributes so it would not be clear (without computation) which one was used.

[Javert899]

Hi,

In the 'develop' branch now, when the trace log is converted to event log, and the include_case_attributes is (as default) True, then it is checked that the attribute "concept:name" is present, and if not is added considering the case id glue

A.k.a. the code changes as follows:

        if include_case_attributes:
            for k in event.keys():
                if k.startswith(case_attribute_prefix):
                    trace_attr[k.replace(case_attribute_prefix, '')] = event[k]
            if xes.DEFAULT_TRACEID_KEY not in trace_attr:
                trace_attr[xes.DEFAULT_TRACEID_KEY] = trace_attr[case_glue]

[fmannhardt]

Just trying this with the new v1.0.19 and now I get this error when trying to convert an EventLog with case_id patient to a TraceLog:

 Error in py_call_impl(callable, dots$args, dots$keywords) : 
   KeyError: 'patient'

 Detailed traceback: 
   File "C:\Users\felixm\AppData\Local\CONTIN~1\ANACON~1\envs\R-RETI~1\lib\site-packages\pm4py\objects\conversion\log\factory.py", line 14, in apply
     return VERSIONS[variant](log, parameters=parameters)
   File "C:\Users\felixm\AppData\Local\CONTIN~1\ANACON~1\envs\R-RETI~1\lib\site-packages\pm4py\objects\conversion\log\versions\to_trace_log.py", line 26, in apply
     case_attribute_prefix=case_pref, enable_deepcopy=enable_deepcopy)
   File "C:\Users\felixm\AppData\Local\CONTIN~1\ANACON~1\envs\R-RETI~1\lib\site-packages\pm4py\objects\conversion\log\versions\to_trace_log.py", line 66, in transform_event_log_to_trace_log
     trace_attr[xes.DEFAULT_TRACEID_KEY] = trace_attr[case_glue]

It seems that you assume that all case attributes have some kind of prefix? Could this limitation be dropped? In many cases attributes have a domain specific meaning and it would be weird having to rename it. It also prevents me from converting an R event log to PM4PY an back without changes.

Edit: Just to add, for my use case on converting from bupaR to PM4PY and back I can workaround this issue by simply adding an artificial attribute and not converting trace attributes. So, the comment is more on the general PM4PY design choice. :)

[Javert899]

Hi,

Sorry for that. The bug you are seeing is obviously an error introduced by my side on last release. Indeed, it should be trace_attr[xes.DEFAULT_TRACEID_KEY] = glue

I've pushed this hotfix into release 1.0.20

For the remainder of your observations, there could be two options:

• provide to the conversion to trace log a list of attributes to consider as trace attributes
• provide an utility to "move later" event attributes to trace level