1

I'm trying to implement DNN (mnist) on pycaffe.

pretrained_solver = caffe.SGDSolver('mnist/lenet_solver.prototxt')

pretrained_solver = caffe.get_solver('mnist/lenet_solver.prototxt')

What is the difference between the two lines above? It seems that I get the same result with either of the lines...

User42
  • 970
  • 1
  • 16
  • 27
sandman
  • 13
  • 4

1 Answers1

1

According to https://github.com/BVLC/caffe/issues/3033,

The first (get_solver) allows you to specify the solver type in the prototxt.

The second (SGDSolver) is less reliable because if you specify a different solver type in the prototxt, then it will still use SGDSolver and not tell you that you have a mismatch.

yangjie
  • 6,619
  • 1
  • 33
  • 40