PyDREAM is a Python implementation of the Decay Replay Mining (DREAM) approach and the corresponding predictive algorithms (NAP and NAPr) similar to the ones described in the paper Decay Replay Mining to Predict Next Process Events and is based on PM4Py. The original Java implementation used for benchmarking can be found here. There exists also a ProM plugin here.

Please see example.py for an end-to-end example.

PyDREAM requires an event log and a corresponding PNML Petri net file, both imported through PM4Py.

from pm4py.objects.log.importer.xes import importer as xes_importer
from pm4py.objects.petri.importer import importer as pnml_importer

log = xes_importer.apply('YOUR_EVENTLOG.xes')
net, initial_marking, _ = pnml_importer.apply("YOURPETRINET.pnml")

The event log must be wrapped into a PyDREAM LogWrapper instance.

from pydream.LogWrapper import LogWrapper
from pm4py.objects.log.importer.xes import importer as xes_importer

log = xes_importer.apply('YOUR_EVENTLOG.xes')
log_wrapper = LogWrapper(log)

If you plan on using resources, for example to train a NAPr model, provide the relevant resource identifiers.

log_wrapper = LogWrapper(log, resources=["IDENTIFIER"])

The loaded Petri net can be enhanced as described in the paper as described subsequently. An EnhancedPN instance will automatically detect if a given LogWrapper objects encompasses resources.

from pydream.EnhancedPN import EnhancedPN

enhanced_pn = EnhancedPN(net, initial_marking)
enhanced_pn.enhance(log_wrapper)
enhanced_pn.saveToFile("YOURENHANCEDPN.json")

Timed State Samples through Decay Replay can be obtained by replaying an event log wrapped into an LogWrapper instance on the enhanced Petri net. The function returns two values. First, the resulting Timed State Samples in JSON format that can be saved to file directly. Second, a list of TimedStateSample instances that can be used for further processing. The Timed State Samples will contain resource counters if a given LogWrapper objects encompasses resources.

import json
from pydream.LogWrapper import LogWrapper
from pydream.EnhancedPN import EnhancedPN
from pm4py.objects.petri.importer import importer as pnml_importer
from pm4py.objects.log.importer.xes import importer as xes_importer

log = xes_importer.apply('YOUR_EVENTLOG.xes')
log_wrapper = LogWrapper(log)

net, initial_marking, _ = pnml_importer.import_net("YOURPETRINET.pnml")
enhanced_pn = EnhancedPN(net, initial_marking, decay_function_file="YOURENHANCEDPN.json")

tss_json, tss_objs = enhanced_pn.decay_replay(log_wrapper=log_wrapper)
with open("timedstatesamples.json", 'w') as f:
        json.dump(tss_json, f)

To replay also resources, call decay_replay() by listing all resources identifier that should be considered.

log_wrapper = LogWrapper(log, resources=["IDENTIFIER"])
tss_json, tss_objs = enhanced_pn.decay_replay(log_wrapper=log_wrapper, resources=["IDENTIFIER"])

A NAP or NAPr predictor can be trained in the following way.

from pydream.predictive.nap.NAP import NAP

algo = NAP(tss_train_file="timedstatesamples.json", tss_test_file="timedstatesamples.json", options={"n_epochs" : 100})
algo.train(checkpoint_path="model-path", name="MODEL-NAME", save_results=True)

The corresponding model will be stored automatically based on the provided checkpoint_path and name parameters. The implemented options include:

• "seed" : int
• "n_epochs" : str
• "n_batch_size" : int
• "dropout_rate" : float
• "eval_size" : float
• "activation_function" : str

One can load a trained model from file in the following way.

algo = NAP()
algo.loadModel(path="model-path", name="MODEL-NAME")

The following function predicts the next events by returning the raw NAP output and the event names in string format.

from pydream.util.TimedStateSamples import loadTimedStateSamples

tss_loaded_objs = loadTimedStateSamples("timedstatesamples.json")
nap_out, string_out = algo.predict(tss_loaded_objs)

PyDREAM is developed for Python 3.7 and is based on PM4Py v2.1.2. NAP and NAPr require tensorflow 2.4.0. The full list of requirements can be found in requirements.txt.

This code is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the corresponding license for more details.

@article{theis2019decay,
  title={Decay Replay Mining to Predict Next Process Events},
  author={Theis, Julian and Darabi, Houshang},
  journal={IEEE Access},
  volume={7},
  pages={119787--119803},
  year={2019},
  publisher={IEEE}
}