Extraneous activity delays BPS model enhancer

Python implementation of the approach to enhance BPS models by modeling the extraneous activity delays with timer events. A preliminary version of this technique has been presented in the paper "Modeling Extraneous Activity Delays in Business Process Simulation", by David Chapela-Campa and Marlon Dumas. The complete version is presented in the paper "Enhancing Business Process Simulation Models with Extraneous Activity Delays", by David Chapela-Campa and Marlon Dumas.

The technique supports two simulation engines: QBP and Prosimos. The input consists of i) an event log (pd.DataFrame) recording the execution of the activities of a process (including resource information), and ii) a BPS model. In QBP, the BPS model is represented by a BPMN file (example here). In Prosimos, by a BPMN file with the process model structure, and JSON file with the simulation parameters (example here).

The approach uses the input event log to discover the waiting time caused by extraneous factors, and enhances the input BPS model with timer events to model such delays. The proposal consists of different configurations such as:

• Discover the extraneous delays using the Naive proposal (DiscoveryMethod.NAIVE), or the Eclipse-aware proposal (DiscoveryMethod.COMPLEX).
• Consider the extraneous delays to occur after an activity instance, i.e., ex-post configuration (TimerPlacement.AFTER), or to occur before an activity instance, i.e., ex-ante configuration (TimerPlacement.BEFORE).
• Enhance the BPS model with the discovered extraneous delays (DirectEnhancer()), or try to tune the discovered delays with a TPE hyper-optimization stage (HyperOptEnhancer())

For a more detailed explanation of the different variants of the approach, we refer to the paper "Enhancing Business Process Simulation Models with Extraneous Activity Delays".

Requirements

• Python v3.9.5+
• PIP v21.1.2+
• Python dependencies: All packages are listed in pyproject.toml.

Basic Usage

Check this test file for a simple example of how to run the technique with Prosimos, and config file for an explanation of the configuration parameters.

Here, we provide two example of the proposal:

Using Prosimos as simulation engine with no TPE optimization stage

import json

import pandas as pd
from lxml import etree
from pix_framework.io.event_log import DEFAULT_CSV_IDS

from extraneous_activity_delays.config import Configuration, SimulationModel, SimulationEngine
from extraneous_activity_delays.config import DiscoveryMethod, TimerPlacement, OptimizationMetric
from extraneous_activity_delays.enhance_with_delays import DirectEnhancer

# Set up default configuration
log_ids = DEFAULT_CSV_IDS
config = Configuration(
  log_ids=log_ids, process_name="prosimos-example",
  simulation_engine=SimulationEngine.PROSIMOS,
  discovery_method=DiscoveryMethod.COMPLEX,  # Eclipse-aware method
  timer_placement=TimerPlacement.BEFORE,  # ex-ante configuration
  optimization_metric=OptimizationMetric.RELATIVE_EMD
  # working_schedules=working_schedules  # Use this to consider resource unavailability
)
# Read event log
event_log = pd.read_csv("path_to_input_log.csv")
event_log[log_ids.start_time] = pd.to_datetime(event_log[log_ids.start_time], utc=True)
event_log[log_ids.end_time] = pd.to_datetime(event_log[log_ids.end_time], utc=True)
event_log[log_ids.resource].fillna("NOT_SET", inplace=True)
event_log[log_ids.resource] = event_log[log_ids.resource].astype("string")
# Read BPMN model
parser = etree.XMLParser(remove_blank_text=True)
bpmn_model = etree.parse("path_to_bps_model.bpmn", parser)
# Read simulation parameters
with open("path_to_bps_parameters.json") as json_file:
  simulation_parameters = json.load(json_file)
simulation_model = SimulationModel(bpmn_model, simulation_parameters)
# Enhance with hyper-parametrized activity delays with hold-out
enhancer = DirectEnhancer(event_log, simulation_model, config)
enhanced_simulation_model = enhancer.enhance_simulation_model_with_delays()
# Write enhanced BPS model (BPMN and parameters)
enhanced_simulation_model.bpmn_document.bpmn_document.write("path_of_enhanced_bps_model.bpmn", pretty_print=True)
with open("path_to_enhanced_bps_parameters.json") as json_file:
  json.dump(enhanced_simulation_model.simulation_parameters, json_file)

Using QBP as simulation engine with TPE optimization stage

import pandas as pd
from lxml import etree
from pix_framework.io.event_log import DEFAULT_CSV_IDS

from extraneous_activity_delays.config import Configuration, SimulationModel, SimulationEngine
from extraneous_activity_delays.config import DiscoveryMethod, TimerPlacement, OptimizationMetric
from extraneous_activity_delays.enhance_with_delays import HyperOptEnhancer

# Set up default configuration
log_ids = DEFAULT_CSV_IDS
config = Configuration(
  log_ids=log_ids, process_name="qbp-example",
  max_alpha=10.0, training_partition_ratio=0.5,
  num_iterations=100, simulation_engine=SimulationEngine.QBP,
  discovery_method=DiscoveryMethod.COMPLEX,  # Eclipse-aware method
  timer_placement=TimerPlacement.BEFORE,  # ex-ante configuration
  optimization_metric=OptimizationMetric.RELATIVE_EMD
  # working_schedules=working_schedules  # Use this to consider resource unavailability
)
# Read event log
event_log = pd.read_csv("path_to_input_log.csv")
event_log[log_ids.start_time] = pd.to_datetime(event_log[log_ids.start_time], utc=True)
event_log[log_ids.end_time] = pd.to_datetime(event_log[log_ids.end_time], utc=True)
event_log[log_ids.resource].fillna("NOT_SET", inplace=True)
event_log[log_ids.resource] = event_log[log_ids.resource].astype("string")
# Read BPMN model
parser = etree.XMLParser(remove_blank_text=True)
bpmn_model = etree.parse("path_to_bps_model.bpmn", parser)
simulation_model = SimulationModel(bpmn_model)
# Enhance with hyper-parametrized activity delays with hold-out
enhancer = HyperOptEnhancer(event_log, simulation_model, config)
enhanced_simulation_model = enhancer.enhance_simulation_model_with_delays()
# Write enhanced BPS model
enhanced_simulation_model.bpmn_document.write("path_of_enhanced_bps_model.bpmn", pretty_print=True)

Resource unavailability: working schedules format

from pix_framework.discovery.resource_calendar_and_performance.crisp.resource_calendar import RCalendar

weekly_calendars = [
    {
        "resource_name": "Jonathan",
        "time_periods": [
            {
                "from": "MONDAY",
                "to": "FRIDAY",
                "beginTime": "09:00:00.000",
                "endTime": "14:00:00.000"
            },
            {
                "from": "MONDAY",
                "to": "THURSDAY",
                "beginTime": "16:00:00.000",
                "endTime": "19:00:00.000"
            }
        ]
    }, {
        "resource_name": "DIO",
        "time_periods": [
            {
                "from": "MONDAY",
                "to": "SUNDAY",
                "beginTime": "08:00:00.000",
                "endTime": "20:00:00.000"
            }
        ]
    }
]


def from_weekly_calendar() -> dict:
    # Read calendars
    resource_calendars = {}
    for calendar in weekly_calendars:
        resource_name = calendar['resource_name']
        r_calendar = RCalendar("calendar_{}".format(resource_name))
        for slot in calendar["time_periods"]:
            r_calendar.add_calendar_item(
                slot["from"], slot["to"], slot["beginTime"], slot["endTime"]
            )
        resource_calendars[resource_name] = r_calendar
    # Return resource calendars
    return resource_calendars