Log Filtering with Linear Temporal Properties

The filtering of a log according to Linear Temporal Logic properties is implemented with an LTLf checker based on automata. First of all we create an D4PyEventLog object that wraps the log.

import os

from Declare4Py.D4PyEventLog import D4PyEventLog

log_path = os.path.join("../../../", "tests", "test_logs","Sepsis Cases.xes.gz")
event_log = D4PyEventLog()
event_log.parse_xes_log(log_path)

/usr/lib/python3/dist-packages/pyparsing.py:108: DeprecationWarning: module 'sre_constants' is deprecated
  import sre_constants
/usr/local/lib/python3.11/dist-packages/lark/utils.py:163: DeprecationWarning: module 'sre_parse' is deprecated
  import sre_parse

/usr/local/lib/python3.11/dist-packages/pm4py/utils.py:486: UserWarning: the EventLog class has been deprecated and will be removed in a future release.
  warnings.warn("the EventLog class has been deprecated and will be removed in a future release.")

The classes LTLTemplate and LTLModel are responsible for the LTLf template definitions and their instantiations into LTLf models given some activities and/or payload names:

from Declare4Py.ProcessModels.LTLModel import LTLTemplate
from Declare4Py.ProcessModels.LTLModel import LTLModel

template : LTLTemplate = LTLTemplate('is_first_state_a')
model_1 : LTLModel= template.fill_template(['ER Registration'], attr_type=['concept:name'])

template: LTLTemplate = LTLTemplate('eventually_a')
model_2 : LTLModel = template.fill_template(['Leucocytes'], attr_type=['concept:name'])

The class LTLAnalyzer is therefore needed to analyze the log and filtering its traces according to multiple input LTLf models.

from Declare4Py.ProcessMiningTasks.ConformanceChecking.LTLAnalyzer import LTLAnalyzer

analyzer = LTLAnalyzer(event_log, [model_1, model_2])
conf_check_res_df = analyzer.run_multiple_models(minimize_automaton=False)

The output is a Pandas DataFrame that can be easily queried.

print(f"Accepted traces: {len(conf_check_res_df[conf_check_res_df['accepted'] == True])}")
conf_check_res_df

Accepted traces: 957

	case:concept:name	accepted
0	A	True
1	B	True
2	C	True
3	D	True
4	E	True
...	...	...
1045	HNA	True
1046	INA	False
1047	JNA	False
1048	KNA	True
1049	LNA	False

1050 rows × 2 columns

The same easy coding holds also for Branched-DECLARE templates:

template = LTLTemplate('alternate_response')
activities_a = ["Er Registration", "IV Liquid"]
activities_b = ["CRP", "IV Antibiotics"]
model_1 = template.fill_template(activities_a, activities_b)

template = LTLTemplate('not_precedence')
activities_a = ["ER Sepsis Triage", "CRP"]
activities_b = ["IV Antibiotics", "LacticAcid"]
model_2 = template.fill_template(activities_a, activities_b)

analyzer = LTLAnalyzer(event_log, [model_1, model_2])
conf_check_res_df = analyzer.run_multiple_models()

We therefore access the Pandas Dataframe

print(f"Accepted traces: {len(conf_check_res_df[conf_check_res_df['accepted'] == True])}")
conf_check_res_df

Accepted traces: 117

	case:concept:name	accepted
0	A	False
1	B	False
2	C	False
3	D	False
4	E	False
...	...	...
1045	HNA	False
1046	INA	False
1047	JNA	False
1048	KNA	False
1049	LNA	False

1050 rows × 2 columns

More information about managing process models and the LTLf or B-Declare templates can be found in tutorials 2 and 3.