training_rdd=train_data,
criterion=ClassNLLCriterion(),
optim_method=SGD(learningrate=0.4, learningrate_decay=0.0002),
end_trigger=MaxEpoch(5),
batch_size=2048)
# Set the validation logic
optimizer.set_validation(
batch_size=2048,
val_rdd=test_data,
trigger=EveryEpoch(),
val_method=[Top1Accuracy()]
)
# generate summaries and start tensortboard if needed
(train_summary, val_summary) = generate_summaries('/home/cdsw/tmp/bigdl_summaries', 'cnn')
optimizer.set_train_summary(train_summary)
optimizer.set_val_summary(val_summary)
# * Train the network. Wait some time till it finished.. Voila! You've got a trained model
trained_model = optimizer.optimize()
# ## 4. Predication on test data
def map_predict_label(l):
return np.array(l).argmax()
def map_groundtruth_label(l):
return l[0] - 1
# 4. Optimizer setup
# Create an Optimizer
optimizer = Optimizer(
model=model,
training_rdd=train_data,
criterion=MSECriterion(),
optim_method=Adam(),
end_trigger=MaxEpoch(2),
batch_size=batch_size)
# generate summaries and start tensortboard if needed
(train_summary, val_summary) = generate_summaries('/home/cdsw/tmp/bigdl_summaries', 'autoencoder')
optimizer.set_train_summary(train_summary)
optimizer.set_val_summary(val_summary)
#Train model
trained_model = optimizer.optimize()
# 5. Loss visualization
# Let's draw the performance curve during optimization.
loss = np.array(train_summary.read_scalar("Loss"))
plt.figure(figsize = (12,12))
plt.plot(loss[:,0],loss[:,1],label='loss')
training_rdd=train_data,
criterion=ClassNLLCriterion(),
optim_method=SGD(learningrate=learning_rate),
end_trigger=MaxEpoch(training_epochs),
batch_size=batch_size)
# Set the validation logic
optimizer.set_validation(
batch_size=batch_size,
val_rdd=test_data,
trigger=EveryEpoch(),
val_method=[Top1Accuracy()]
)
# generate summaries and start tensortboard if needed
(train_summary, val_summary) = generate_summaries('/home/cdsw/tmp/bigdl_summaries', 'multilayer_perceptron')
optimizer.set_train_summary(train_summary)
optimizer.set_val_summary(val_summary)
get_ipython().run_cell_magic(u'time', u'', u'# Boot training process\ntrained_model = optimizer.optimize()\nprint "Optimization Done."')
# 5. Loss visualization
# After training, we can draw the preformance curves from the previous `train_summary` and `val_summary` variables.
loss = np.array(train_summary.read_scalar("Loss"))
top1 = np.array(val_summary.read_scalar("Top1Accuracy"))
def plotLoss():
plt.figure(figsize = (12,12))
plt.subplot(2,1,1)
# Create an Optimizer
optimizer = Optimizer(
model=lenet_model,
training_rdd=train_data,
criterion=ClassNLLCriterion(),
optim_method=SGD(learningrate=0.4, learningrate_decay=0.0002),
end_trigger=MaxEpoch(5),
batch_size=256)
# Set the validation logic
optimizer.set_validation(
batch_size=256,
val_rdd=test_data,
trigger=EveryEpoch(),
val_method=[Top1Accuracy(), Loss()]
)
app_name='lenet-'+dt.datetime.now().strftime("%Y%m%d-%H%M%S")
# create TrainSummary
(train_summary,val_summary) = generate_summaries('/tmp/bigdl_summaries', app_name)
optimizer.set_train_summary(train_summary)
optimizer.set_val_summary(val_summary)
# Boot training process
trained_model = optimizer.optimize()
print "Optimization Done."