Closed MicheleCancilla closed 3 years ago
The last eddl release provides an example (8_cifar_pretrained.cpp) that:
dense
However, the lines: https://github.com/deephealthproject/eddl/blob/690f3b6cd8b6287247e850204de0c7a4911486a1/examples/onnx/nn/2_cifar10/8_cifar_pretrained.cpp#L51-L58 do not contain the init_weights = false flag that avoids the random weight initialization.
init_weights = false
Is the false flag required in this case?
false
Yes, but then we will not initialize the new layers... so i have to think how to fix that.
The last eddl release provides an example (8_cifar_pretrained.cpp) that:
dense
layerdense
by means of networks mergingHowever, the lines: https://github.com/deephealthproject/eddl/blob/690f3b6cd8b6287247e850204de0c7a4911486a1/examples/onnx/nn/2_cifar10/8_cifar_pretrained.cpp#L51-L58 do not contain the
init_weights = false
flag that avoids the random weight initialization.Is the
false
flag required in this case?