Closed krishpat closed 4 years ago
@krishpat Firstly, your objective is convex and you wouldn't need Alpine for this. :) That said, we are currently working on including convexity detection for objectives of this type.
Regarding your issues, currently the way we are evaluating the relative gap when upper bound is 0 is not fully right, which will be fixed soon.
Updated output:
====================================================================================================
| Iter | Incumbent | Best Incumbent | Lower Bound | Gap (%) | Time
| 1 | 0.0 | 0.0 | -0.0539 | 5.39029 | 0.55s
| 2 | 0.0 | 0.0 | -0.0135 | 1.34765 | 0.61s
| 3 | 0.0 | 0.0 | -0.0041 | 0.41449 | 0.67s
| 4 | 0.0045 | 0.0 | -0.0041 | 0.41449 | 0.79s
| 5 | 0.0045 | 0.0 | -0.0034 | 0.33699 | 0.89s
| 6 | 0.0 | 0.0 | -0.001 | 0.1037 | 0.99s
| 7 | 0.0011 | 0.0 | -0.001 | 0.1037 | 1.08s
| 8 | 0.0014 | 0.0 | -0.001 | 0.1037 | 1.23s
| 9 | 0.0018 | 0.0 | -0.001 | 0.1037 | 1.36s
| 10 | 0.002 | 0.0 | -0.001 | 0.1037 | 1.5s
| 11 | 0.0021 | 0.0 | -0.001 | 0.1037 | 1.64s
| 12 | 0.0021 | 0.0 | -0.001 | 0.1037 | 1.86s
| 13 | 0.0021 | 0.0 | -0.001 | 0.1037 | 2.04s
| 14 | 0.0021 | 0.0 | -0.001 | 0.1037 | 2.26s
| 15 | 0.0021 | 0.0 | -0.001 | 0.1037 | 2.49s
| 16 | 0.0022 | 0.0 | -0.001 | 0.1037 | 2.82s
| 17 | 0.0022 | 0.0 | -0.0008 | 0.08432 | 3.37s
| 18 | 0.0 | 0.0 | -0.0003 | 0.026 | 4.0s
| 19 | 0.0003 | 0.0 | -0.0003 | 0.026 | 4.5s
| 20 | 0.0003 | 0.0 | -0.0002 | 0.02116 | 5.14s
| finish | 0.0 | 0.0 | -0.0001 | 0.00657 | 5.77s
====================================================================================================
Consider this NLP:
Here is Alpine's log of the output :
LARGE
when the gaps are actually small? Is it because the upper bound (0) value is in the denominator of the gap?There may be some issue in the way relative gaps are evaluated.