Intelligent Habitat Restoration Under Uncertainty
Authors: Tommaso Urli, Jana Brotánková, Philip Kilby, Pascal Van Hentenryck
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results Our solver was implemented in GECODE 4.4.0 (Schulte, Tack, and Lagerkvist 2015), and has been tested by exploring the effects of different budgeting scenarios on a master instance involving 13 islands, 46 features, and 19 threats. The chosen horizon for this validation was 10 periods. The instance is generated from incomplete survey data that will be fully available only during the next year, and represents only about 10% of the extent of the real problem faced by conservation managers. The missing data, namely the impact of actions on threats , and the effect of threats on features , have been replaced with synthetic values. The validation experiments were run on a Linux cluster of 2 2.8 GHz AMD 6-Core Opteron 4184 nodes with 64 GB of RAM each, using a time limit of 1 hour per optimisation run. The solver parameters were tuned automatically through an F-Race (Birattari et al. 2010). The tuning experiments were run on 800 random extracts (subsets) of the master instance, involving 70 100% of the islands and budgets B {0.5, 1, 2, 5} millions. The tuning procedure fixed the parameters to iimax = tvar = 50. The detail of the tuning process are beyond the scope of this paper. The tuning experiments were run on a 2.9 GHz Intel Core i7 with 8 GB of RAM running Mac OS 10, using a time limit of 5 minutes. Figure 3 shows some preliminary results of this validation. |
| Researcher Affiliation | Academia | Tommaso Urli,1 Jana Brot ankov a,2 Philip Kilby,1 and Pascal Van Hentenryck1 1 Optimisation Research Group, NICTA Canberra Research Lab, 7 London Circuit, Canberra, 2601 ACT, Australia 2 ARC Centre of Excellence for Coral Reef Studies, James Cook University, 1 James Cook Drive, Townsville, 4811 QLD, Australia |
| Pseudocode | No | The paper describes the models and algorithms in prose but does not provide pseudocode or structured algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about making the source code available or links to a code repository. |
| Open Datasets | No | The instance is generated from incomplete survey data that will be fully available only during the next year, and represents only about 10% of the extent of the real problem faced by conservation managers. The missing data, namely the impact of actions on threats , and the effect of threats on features , have been replaced with synthetic values. |
| Dataset Splits | No | The paper mentions that |
| Hardware Specification | Yes | The validation experiments were run on a Linux cluster of 2 2.8 GHz AMD 6-Core Opteron 4184 nodes with 64 GB of RAM each, using a time limit of 1 hour per optimisation run. The tuning experiments were run on a 2.9 GHz Intel Core i7 with 8 GB of RAM running Mac OS 10, using a time limit of 5 minutes. |
| Software Dependencies | Yes | Our solver was implemented in GECODE 4.4.0 (Schulte, Tack, and Lagerkvist 2015) |
| Experiment Setup | Yes | The solver parameters were tuned automatically through an F-Race (Birattari et al. 2010). The tuning experiments were run on 800 random extracts (subsets) of the master instance, involving 70 100% of the islands and budgets B {0.5, 1, 2, 5} millions. The tuning procedure fixed the parameters to iimax = tvar = 50. |