A new version of the Xcavate ProM plugin has been released in version 6.16.65 of the DiSCover package. First, this new release fixes some issues with the token-based replay:
- The initial marking was assumed to contain only a single token, which is true for WF-nets but later versions of this plugin may add places that initially contain additional tokens (as shown in the picture below, green indicates initial marking, red indicates final marking, green-red indicates both).
- Reaching a final marking may need some silent transitions to fire first. In previous versions, it was assumed that this was not needed, but examples exist for which it is needed.
Second, the fitness value returned from the token-based replay has been changed slightly: If possible, it will first add 1 to both missing and remaining and then compute the fitness as usual. As a result, even though a trace may have maximal fitness, a longer trace may have a better fitness.
Third, a new feature has been added to allow the user to restrict the addition of equivalence places: If either an equivalence place or its counter place contains more than a given number of tokens, then these places will not be added. By default, this number is set to 1.
The results show that for the different PDC data sets this either has a positive effect, or no effect:
| Year | Max | Win | Xcavator 6.15 | Xcavator 6.16.65 |
|---|---|---|---|---|
| 2016 | 200 | 193 | 178 | 178 |
| 2017 | 200 | 197 | 168 | 168 |
| 2019 | 900 | 898 | 817 | 823 |
| 2020 | 100% | 76.2% | 89.1% | 89.3% |
| 2021 | 100% | 96.2% | 97.8% | 97.8% |
| 2022 | 100% | 89.5% | 89.5% | 95.6% |
| 2023 | 100% | 84.0% | 92.3% | 93.2% |
| 2024 | 100% | 90.9% | 92.9% | 94.8% |
| 2025 | 100% | 91.8% | 92.9% | 94.0% |
