def train(opt):
#####Compile mode####
input_size = (opt.row, opt.col, opt.ch)
SegM = SegModel(input_size)
model = SegM.model
_adam = optimizers.Adam(lr=opt.lr, beta_1=0.9, beta_2=0.999, decay=0.0)
model.compile(loss='binary_crossentropy',
optimizer=_adam,
metrics=['accuracy'])
###################load data###########
img_dim = (opt.row, opt.col)
[train_img, train_mask, test_img,
test_mask] = dataloader(opt.ad1, opt.ad2, opt.ad3, opt.ad5,
opt.img_format, img_dim)
####check point####
model_checkpoint = ModelCheckpoint(opt.chekp + '.hdf5',
monitor='val_acc',
verbose=1,
save_best_only=True)
###################train###########
hist = model.fit(train_img,
train_mask,
validation_data=(test_img, test_mask),
batch_size=opt.batch_size,
nb_epoch=opt.epochs,
verbose=1,
callbacks=[model_checkpoint])
# ######evaluate#####
model.load_weights(opt.chekp + '.hdf5')
y_pred = model.predict(test_img)
f = fscore(y_pred, test_img, test_mask)
print(f)
return hist, f
return prediction
def loss_function(real, preds):
# y_error = tf.subtract(real,preds)
# return tf.nn.l2_loss(y_error)
return tf.losses.absolute_difference(real, preds)
# return tf.losses.mean_squared_error(real,preds)
print("version of tensorflow:", tf.__version__)
tf.enable_eager_execution()
tms = utils.buildAllTMToATensor(xmlPath="../traffic-matrices/")
data_loader = dataloader(tms, EMA=True, alpha=0.9)
model = Model(out_size=23 * 23, units=23 * 23)
# Using adam optimizer with default arguments
lr = 0.01
optimizer = tf.train.AdamOptimizer(learning_rate=lr)
BATCH_SIZE = 32
seq_length = 20
ds_list = [0, 1, 1, 1, 1, 1, 1, 1, 1, 1]
model.build(tf.TensorShape([BATCH_SIZE, len(ds_list), 23 * 23]))
model.summary()
# Directory where the checkpoints will be saved
checkpoint_path = './training_checkpoints.ckpt'
loss_record = []
print("version of tensorflow:", tf.__version__)
# tf.enable_eager_execution()
# ds_1_end =list(range(1,10))
# ds_list=[0]
ds_list = list(range(0, 100, 2))
target_pos = 1
batchsize = 64
#**************prepare data*********************
tms = utils.buildAllTMToATensor(xmlPath="../traffic-matrices/")
# data_loader = dataloader(tms,EMA=False,alpha=0.9)
data_loader = dataloader(tms,
ratio_for_train=0.85,
EMA=False,
alpha=0.1,
BOX=True,
normalization=1)
train_x, train_y = [], []
# gen = data_loader.iter_for_train(ds_list = ds_list,batchsize=1,target_pos=1 )
for x, y in data_loader.iter_for_train(ds_list=ds_list,
batchsize=1,
target_pos=target_pos):
b, l, m, n = x.shape
b_, m_, n_ = y.shape
# tmp_x =
train_x.append(x.reshape([l * m * n]))
train_y.append(y.reshape([m_ * n_]))
train_x = np.array(train_x)