def AddTrainingOperators(model, softmax, label):
"""Adds training operators to the model."""
xent = model.LabelCrossEntropy([softmax, label], 'xent')
# compute the expected loss
loss = model.AveragedLoss(xent, "loss")
# track the accuracy of the model
model_defs.AddAccuracy(model, softmax, label)
# use the average loss we just computed to add gradient operators to the model
model.AddGradientOperators([loss])
# do a simple stochastic gradient descent
optimizer.build_sgd(
model,
base_learning_rate=0.1,
policy="step",
stepsize=1,
gamma=0.999,
)
# Validation model
val_model = model_helper.ModelHelper(name="val_net",
arg_scope=arg_scope,
init_params=False)
# Uncomment following two lines for GPU
val_model.param_init_net.RunAllOnGPU()
val_model.net.RunAllOnGPU()
data, label = model_defs.AddInput(val_model,
batch_size=validation_images,
db=validation_lmdb,
db_type='lmdb')
softmax = model_defs.AddUpgradedLeNetModel_GPU(val_model, data, num_classes,
image_height, image_width,
image_channels, device_opts)
model_defs.AddAccuracy(val_model, softmax, label)
# Testing model
test_model = model_helper.ModelHelper(name="test_net",
arg_scope=arg_scope,
init_params=False)
# Uncomment following two lines for GPU
test_model.param_init_net.RunAllOnGPU()
test_model.net.RunAllOnGPU()
data, label = model_defs.AddInput(test_model,
batch_size=testing_images,
db=testing_lmdb,
db_type='lmdb')
softmax = model_defs.AddUpgradedLeNetModel_GPU(test_model, data, num_classes,
image_height, image_width,
image_channels, device_opts)