Resuming training from Python Interface

[Sharif Amit]

I'm trying to resume my training from a snapshot but i don't know how to do it.
Read somewhere to use solver.restore(). But can someone tell me what to pass inside the parenthesis as parameters?

[沈仕卿]

I have not found the python interface, but find the c++ interface in solver.cpp

The parameter is the path of the '.solverstate' file

在 2016年11月5日星期六 UTC+8上午1:02:56，Sharif Amit写道：

[Grayson Adkins]

You need both the .caffemodel and .solverstate files. Caffemodel is the model itself (i.e. weights and biases for the network architecture) and solverstate contains all other state information required by the network, including hyperparamters.

Try:

solver.net.copy_from('<filename>.caffemodel')

solver.restore('<filename>.solverstate')

[Sharif Amit]

Thank you.