Open pyankoff opened 9 years ago
I've been focusing on imageNet, for MNIST, it should be similar to the CIFAR-10 example. Also, ccv's CPU support is incomplete (as a reference implementation to verify the correctness of GPU implementation, thus, the back prop for CPU is not optimized, and don't have dropout).
I want to train ccv_convnet on MNIST data set (using LeNet like layers). Could you please provide any suggestions on how to do that?