def train(n, k, firstRun, epochs_no, starting_epoch_no, last_model_file = None):
if firstRun:
total_training_batches_no = b.divide_into_files(True)
total_test_batches_no = b.divide_into_files(False)
else:
total_training_batches_no = 707
total_test_batches_no = 177
cvX, cvY1, cvY2, cvY3 = b.load_cross_validation()
# put appropriate Y value here depending on the category you want to run for
cvY = tflearn.data_utils.to_categorical(cvY3, b.CLASS_3_NUMBER)
# Training
model = buildNetwork(n, k)
if (starting_epoch_no > 0):
model.load(b.PATH + last_model_file)
for j in range(starting_epoch_no, epochs_no):
validation_set = None
snapshot_epoch = False
for i in range(0, total_training_batches_no):
X, Y, Y1, Y2 = b.load_batch(i, True)
Y = tflearn.data_utils.to_categorical(Y, b.CLASS_3_NUMBER)
if i == total_training_batches_no - 1 or i % 50 == 0:
validation_set = cvX, cvY
snapshot_epoch = True
model.fit(X, Y, n_epoch=1, validation_set=validation_set, snapshot_epoch=snapshot_epoch, show_metric=True,
run_id='wideresnet')
if i % 50 == 0:
model.save(b.PATH + 'ALLmodel_wideresnet_batch_' + str(i) + 'epoch_' + str(j) + '.txt')
model.save(b.PATH + 'ALLmodel_wideresnet_epoch_' + str(j) + '.txt')
return [model, total_test_batches_no]
def test(model, total_test_batches_no):
predictions = []
for i in range(0, total_test_batches_no):
testX, product_id, Y3, Y4 = b.load_batch(i, False)
size = len(testX) / 10
for i in range(0, 10):
testXpart = testX[i * size, min((i + 1) * size, len(testX))]
predictions.extend(model.predict_label(testXpart))
with open(b.PATH + 'ALL' + 'resNextPredictions.txt', 'wb') as pickleFile:
pickle.dump(predictions, pickleFile)
net = tflearn.conv_2d(net, 16, 3, regularizer='L2', weight_decay=0.0001)
#resnet part
net = residual_block(net, n, 16, downsample=True)
net = residual_block(net, 1, 32, downsample=True)
net = residual_block(net, n-1, 32, downsample=True)
net = residual_block(net, 1, 64, downsample=True)
net = residual_block(net, n-1, 64)
net = tflearn.batch_normalization(net)
net = tflearn.activation(net, 'relu')
net = tflearn.global_avg_pool(net)
# Regression
net = tflearn.fully_connected(net, b.CLASS_3_NUMBER, activation='softmax')
mom = tflearn.Momentum(0.1, lr_decay=0.1, decay_step=32000, staircase=True)
net = tflearn.regression(net, optimizer=mom, loss='categorical_crossentropy')
# Training
model = tflearn.DNN(net, tensorboard_verbose=0, tensorboard_dir=b.PATH + 'log')
model.load(b.PATH + last_model_file)
predictions = []
for i in range(0, test_batches_no):
testX, product_id, Y3, Y4 = b.load_batch(i, False)
size = len(testX) / 10
for i in range(0, 10):
testXpart = testX[i * size, min((i + 1) * size, len(testX))]
predictions.extend(model.predict_label(testXpart))
with open(b.PATH + 'ALL' + str(n) + 'resNetPredictions.txt', 'wb') as pickleFile:
pickle.dump(predictions, pickleFile)
net = wideresnet_block(net, 1, 64, 2, downsample=True)
net = wideresnet_block(net, n-1, 64, k, downsample=True)
net = tflearn.batch_normalization(net)
net = tflearn.activation(net, 'relu')
net = tflearn.global_avg_pool(net)
# Regression
net = tflearn.fully_connected(net, b.CLASS_3_NUMBER, activation='softmax')
mom = tflearn.Momentum(0.1, lr_decay=0.1, decay_step=32000, staircase=True)
net = tflearn.regression(net, optimizer=mom,
loss='categorical_crossentropy')
# Training
model = tflearn.DNN(net, tensorboard_verbose=0, clip_gradients=0., tensorboard_dir=b.PATH + 'log')
if (starting_epoch_no > 0):
model.load(b.PATH + last_model_file)
for j in range(starting_epoch_no, epochs_no):
validation_set = None
snapshot_epoch = False
for i in range(0, training_batches_no):
X, Y, Y1, Y2 = b.load_batch(i, True)
Y = tflearn.data_utils.to_categorical(Y, b.CLASS_3_NUMBER)
if i == training_batches_no - 1 or i % 50 == 0:
validation_set = cvX, cvY
snapshot_epoch = True
model.fit(X, Y, n_epoch=1, validation_set=validation_set, snapshot_epoch=snapshot_epoch, show_metric=True, run_id='wideresnet')
if i % 50 == 0:
model.save(b.PATH + 'ALLmodel_wideresnet_batch_' + str(i) + 'epoch_' + str(j) + '.txt')
model.save(b.PATH + 'ALLmodel_wideresnet_epoch_' + str(j) + '.txt')
net = tflearn.regression(net, optimizer=mom, loss='categorical_crossentropy')
with open(b.PATH + 'ALLResNetNet created.txt', 'wb') as pickleFile:
pickle.dump([], pickleFile)
# Training
model = tflearn.DNN(net, tensorboard_verbose=0, tensorboard_dir=b.PATH + 'log')
for j in range(0, epochs_no):
with open(b.PATH + 'ALL' + str(j) + 'ResNetEpoch.txt', 'wb') as pickleFile:
pickle.dump([], pickleFile)
validation_set = None
snapshot_epoch = False
for i in range(0, training_batches_no):
with open(b.PATH + 'ALL' + str(i) + 'ResNetBatch.txt',
'wb') as pickleFile:
pickle.dump([], pickleFile)
X, Y, Y1, Y2 = b.load_batch(i, True)
Y = tflearn.data_utils.to_categorical(Y, b.CLASS_3_NUMBER)
if i == training_batches_no - 1 or i % 50 == 0:
validation_set = cvX, cvY
snapshot_epoch = True
model.fit(X,
Y,
n_epoch=1,
validation_set=validation_set,
snapshot_epoch=snapshot_epoch,
show_metric=True,
run_id='resnet')
if i % 50 == 0:
model.save(b.PATH + 'ALLmodel_resnet_batch_' + str(i) + 'epoch_' +
str(j) + '.txt')
model = build_network()
after_first_run_flag= False
val_x, cat1_y, cate2_y, cate3_y = load_cross_validation()
# ***************************************************************************
for epoch in range(0, EPOCH_NUM ):
for batch_index in range(0, total_train_batch_file):
if (after_first_run_flag):
model.load( MODEL_NAME )
after_first_run_flag = True
val_set = None
# We need to change here for label selection based on category
train_data, train_labels, Y2, Y3 = load_batch(batch_index, True)
#converting Y list into Matrix
train_labels = to_categorical( train_labels, CLASS_NUMBER)
val_y_labels = to_categorical( cate3_y, CLASS_NUMBER) # We need to change here
if (batch_index == total_train_batch_file-1 ):
val_set =(val_x, val_y_labels)
model.fit( train_data, train_labels, n_epoch= 1 , shuffle=True, validation_set= val_set ,
show_metric=True, batch_size=96, run_id=MODEL_NAME ) # We need to change here
model.save( MODEL_NAME )
print ("Finished load: ", batch_index )
print ("Finished epoch: ", epoch )
