If you use the "official" vanilla implementation of CMA-ES in Python/Numpy solving the BBOB problem suite through the COCO plateform, running the whole benchmark will take approximately 10 minutes on a single Intel Core i5 @ 2.50GHz with an solid state disk.
+
The same experiment, running the Paradiseo implementation using the seamless binding to the IOHprofiler BBOB implementation, will take 1 minute.
+
+ Thus, you can expect that an algorithm implemented with Paradiseo will be up to 10 times faster than its (heavily optimized) Python counterpart.
+ As indicated in the previous section, this speed is crucially useful for algorithm design and algorithm selection.
+
+
+
+
+
CMA-ES solves numeric problems with quadratic complexity regarding the number of dimensions. If you target simpler algorithms, they will be faster but one can expect a similar behaviour.
+
The pycma used for the comparison rely on numpy for the heavy parts of the computations. If you use a pure Python implementation, the difference will be far greater.