ALL_CNN_C_solver.prototxt

# The train/test net protocol buffer definition
net: "ALL_CNN_C_train_val.prototxt"
# test_iter specifies how many forward passes the test should carry out.
# We have test batch size 200 and 50 test iterations,
# covering the full 10,000 testing images for cifar-10.
test_iter: 50
# Carry out testing every 200 training iterations.
test_interval: 200
# Optimizer, the base learning rate, momentum and the weight decay of the network.
type: "SGD"
momentum: 0.9
base_lr: 0.05
weight_decay: 0.001
# The learning rate policy
lr_policy: "multistep"
gamma: 0.1
# stepvalue: 200 epochs
stepvalue: 40000
# stepvalue: 250 epochs
stepvalue: 50000
# stepvalue: 300 epochs
stepvalue: 60000
# Display every 200 iterations
display: 200
# The maximum number of iterations
# max epochs: 350 ~ 70000 iterations with 256 examples per batch
max_iter: 70000
# snapshot intermediate results every 2000 iterations ~ 10 epochs
snapshot: 2000
snapshot_prefix: "ALL_CNN_C"
# solver mode: CPU or GPU
solver_mode: GPU
# debug mode
debug_info: false