Initial and final markings when connection Petrinet to eventlog

Stijn_de_Laat

Hi all,

We have mined a Heuristics model and made a Petri net from this model. Next, we tried doing a "Replay a log on Petri net for conformance checking". When we do this, we get two warnings, namely one mentioning that the initial marking is an empty marking and the second warning is that 'no final marking is found'.

We have some issue interpreting these markings and are thus unable to correctly run the conformance check. Could someone please explain to us what the initial and final markings are?

Our data is as follows: we have meetings in which we have coded utterances. Every meeting is a case and every utterance code is an event. There are quite some utterances in each meeting and thus the resulting heuristic model has no true beginning or an end, but contains the 'conversational flow'.

Thank you in advance for your help.

Best,

Stijn and Marijn

hverbeek

Dear Stijn and Marijn,

The initial marking corresponds to the initial state of the Petri net: Which places contains initially how many tokens? If your net does not need any initial tokens, then an empty initial marking would do. But if every transition has at least one input place, you do need initial tokens as otherwise every transition would be dead, and nothing can happen.

A final marking corresponds to a possible final state of the Petri net. If a final marking is reached, the execution of the Petri net is said to be successful. For conformance this can be important, as you might want to take only executions that end in a final marking into account.

Kind regards,

Eric.

Stijn_de_Laat

Dear Eric,

Thank you for your response. We still have a question regarding these states.

Our data consists of utterances in conversations and it is random with what type of utterance the conversation starts or ends. Also, every type of utterance will appear several times during the conversation, resulting in a network with lots of loops. Does this mean that we cannot provide the initial and final tokens? If so, are we unable to make a petri net out of this type of data?

Could it be a solution to add a 'start'-utterance and an 'end'-utterance to every eventlog. Thus creating start and end places for tokens in the petri net? If so, would this have any consequences for the conformance analysis.

Best regards,

Stijn and Marijn

hverbeek

Dear Stijn and Marijn,

Whether you need such artificial start and end utterances depends on the miner you are using. But adding them will not hurt, so if you want to play safe, adding them would be a good idea.

A Petri net typically allows for so-called silent (or invisible) transitions. These transitions are not related to any utterance, but are added for routing purposes only. They are typically visualized using a black (or grey) fill. Using such a transition, it is possible to skip utterances, or jump from one utterance to the next. Some miners can add these silent transitions, although some others cannot. So, this again depends on the miner.

The miners typically discover workflow nets (WF nets), which have a single source place (no incoming arcs) and a single sink place (no outgoing arcs) such that every node (transition or place) is on some path from the source place to the sink place. The initial marking is the one token in the source place, and the only final marking is one token in the sink place. After adding the artificial start and end utterances, you may expect that the source place is the only input for the start utterance, and the sink place is the only outplace for the end utterance, but again this depends on the miner. Many miners will do this, but some may not (also depends on the log at hand).

Kind regards,

Eric.

Stijn_de_Laat

Dear Eric,

Thank you for your response. We are using the Heuristic Miner, would that be a miner that does this?

How can we practically add these markings in ProM once we have added them in the logfile?

Best,

Stijn and Marijn

hverbeek

Dear Stijn and Marijn,

The Heuristics miner discovers a heuristics net from the log, which is not a Petri net. But you can convert a heuristics net to a Petri net. This Petri net may not be a workflow net, so I would add the artificial start and end utterances. After that, it will discover a workflow net. The conversion also introduces silent transitions where needed. Apart from the fact that the net may not be sound (some reachable states may not be desired), you should be fine.

Kind regards,

Eric.

Stijn_de_Laat

Dear Eric,

Thank you for your help! The Petri net and following conformance check ran without the warnings when we included the start and stop utterances.

Best,

Stijn and Marijn