Closed tBuLi closed 4 years ago
Hessian and Jacobian models are only created when they are needed. For most use cases, this means the Hessian will no longer be calculated. This is a huge speedup gain, comparing to the code provided in #271, we get the following timings:
identical components, time (s) 1 0.005013465881347656 2 0.0010039806365966797 3 0.0010023117065429688 4 0.0010018348693847656 5 0.0010030269622802734 6 0.0010023117065429688 7 0.002006053924560547 8 0.0010023117065429688 9 0.002004861831665039 10 0.0020058155059814453
whereas in 0.5.1 it was
identical components, time (s) 1 0.0671529769897461 2 0.20554876327514648 3 0.8522927761077881 4 2.4235997200012207 5 5.880227088928223 6 13.202399969100952 7 25.026122331619263 8 42.70384192466736 ...
The penalty will still occur if you actually use a Hessian minimizer, so don't.
Hessian and Jacobian models are only created when they are needed. For most use cases, this means the Hessian will no longer be calculated. This is a huge speedup gain, comparing to the code provided in #271, we get the following timings:
whereas in 0.5.1 it was
The penalty will still occur if you actually use a Hessian minimizer, so don't.