the run test for each algorithm was updated to check for model convergence in the optimization of the sphere function (sum(x**2)) in the interval [0, 10] to a global minima value <= 2. Such a large margin was used to avoid failing CI tests due to poor random initial points, but at the same time checking if no code modification is breaking (at least to a large extent) any existing implementation. To ensure that these tests always work, the following changes were made:
The amount of agents or iterations was increased accordingly for each algorithm;
In some cases the search interval was reduced from [0, 10] to [0, 5].
With these changes in place, the tests were executed 100 times without failure (hence, the chance of a build failing due to poor initialization points is <1%. On the other hand, the overall execution time was increased by 0.8 seconds (from ~1.3 seconds to ~2.1 seconds) in a 2015 macbook pro.
This PR introduces the following modifications:
runtest for each algorithm was updated to check for model convergence in the optimization of the sphere function (sum(x**2)) in the interval[0, 10]to a global minima value<= 2. Such a large margin was used to avoid failing CI tests due to poor random initial points, but at the same time checking if no code modification is breaking (at least to a large extent) any existing implementation. To ensure that these tests always work, the following changes were made:[0, 10]to[0, 5].With these changes in place, the tests were executed 100 times without failure (hence, the chance of a build failing due to poor initialization points is
<1%. On the other hand, the overall execution time was increased by0.8seconds (from~1.3seconds to~2.1seconds) in a 2015 macbook pro.