why solver.prototxt is divided into solver_train.prototxt and solver_test.prototxt?

[ujsyehao]

I am confused about solver_train.prototxt and solver_test.prototxt.

• different base_lr set, lr=0.0001 in solver_train.prototxt and lr=0.001 in solver_test.prototxt
• different gamma set, gamma=0.5 in solver_train.prototxt and gamma=0.1 in solver_test.prototxt
• no weight decay in solver_train.prototxt, weight_decay=0.0005 in solver_test.prototxt
• different stepvalue set in solver_train.prototxt and solver_test.prototxt
• different optimization method, RMSProp method in solver_train.prototxt, SGD method in solver_test.prototxt

[ujsyehao]

@chuanqi305 Hi, Can you give me some advice? Thank you in advance.

[chuanqi305]

@ujsyehao The solver_test.prototxt is used for testing the model accuracy.
The train_iters=0, so the params for train will not be used.

[ujsyehao]

@chuanqi305 Thank for your reply!
solver_test.prototxt is used for testing model accuracy, But I am confused about lr, gamma, weight_decay parameters in solver_test.prototxt, Does caffe test model accuracy(use caffe train tools) require these parameters?