def model_load(downloaded_path):
"""
Load list of three models and keras.layers.Input()
:param downloaded_path:
:return: List containing three models and keras.layers.Input()
"""
input_shape = (28, 28, 1)
# define input tensor as a placeholder
input_tensor = Input(shape=input_shape)
# load multiple models sharing same input tensor
train = not os.path.isfile(os.path.join(_deepxplore_mnist_dir,
'Model1.h5'))
model1 = Model1(input_tensor=input_tensor, train=train)
train = not os.path.isfile(os.path.join(_deepxplore_mnist_dir,
'Model2.h5'))
model2 = Model2(input_tensor=input_tensor, train=train)
train = not os.path.isfile(os.path.join(_deepxplore_mnist_dir,
'Model3.h5'))
model3 = Model3(input_tensor=input_tensor, train=train)
return model1, model2, model3, input_tensor
ganOutput = discriminator(x) gan = Model(inputs=ganInput, outputs=ganOutput) gan.compile(loss='binary_crossentropy', optimizer=adam) #print(gan.summary()) # # gen_img = generator.predict() # # # orig_img = gen_img.copy() adam2 = Adam(lr=args.step, beta_1=0.5) #actually we don't care all output of model1,2,3 but only the category that we focus on #model = Model( x) input_shape = (1, img_rows, img_cols) # define input tensor as a placeholder input_tensor = Input(shape=input_shape) model1 = Model1(input_tensor=input_tensor) model2 = Model2(input_tensor=input_tensor) model3 = Model3(input_tensor=input_tensor) model1.trainable = False model2.trainable = False model3.trainable = False orig_label = 1 # layer_name1, index1 = neuron_to_cover(model_layer_dict1) # layer_name2, index2 = neuron_to_cover(model_layer_dict2) # layer_name3, index3 = neuron_to_cover(model_layer_dict3) #
import torch.optim
import torch.backends.cudnn as cudnn
cudnn.benchmark = True
from Model1 import Model1
#dopo aver richiamato la classe per il model faccio la load
#load gensim model
model = gensim.models.KeyedVectors.load_word2vec_format(
'/media/daniele/AF56-12AA/GoogleNews-vectors-negative300.bin', binary=True)
#model = gensim.models.KeyedVectors.load_word2vec_format('./GoogleNews-vectors-negative300.bin', binary=True)
#load my model
checkpoint = torch.load('checkpoint-2.pth')
model_options = checkpoint["model_options"]
model2 = Model1(**model_options)
#hidden=checkpoint["h"]
model2.load_state_dict(checkpoint["model_state"])
#modello_h=modello.h
#modello_dict=modello.x
#prepare to test
zero = torch.FloatTensor(1, 1)
zero.fill_(0)
fineparola = torch.cat([zero, zero], 1)
h = torch.zeros(1, 1, 1024)
target_as_input = torch.zeros(1, 302)
#print(fineparola)
try:
domanda = input("you: ")
domanda = re.findall(r'\w+', domanda)
if ins_type=="VTL":
zeta2=0.2
zeta1=1
#Data.Maxtour= zeta1*math.ceil(float(NN)/M) * np.percentile(Data.distances.values(),50)
Data.Maxtour= zeta1*math.ceil(float(NN)/M[NN]) * np.percentile(list(Data.distances.values()),50)
#Data.Q= zeta2 * Data.G.node[0]['supply']/M # very tight capacity for instanc 6-10
Data.Q= zeta2 * Data.G.nodes[0]['supply']/M[NN]
#Data.Total_dis_epsilon= 0.85* M*Data.Maxtour#0.85 * M*Data.Maxtour
Data.Total_dis_epsilon= 0.85* M[NN]*Data.Maxtour
R=R_dic[File_name ]
Data.Q= max ( math.ceil(Data.total_demand / float(Data.M) ) , max(dict(Data.Gc.nodes(data='demand')).values()) )
start= time()
best_obj ,LB ,Runtime ,GAP = Model1(Data,R)
#best_obj ,LB ,Runtime ,GAP = Model2(Data,R)
#best_obj ,LB ,Runtime ,GAP = Model1_V2(Data,R)
#oldresult=read_object('G:\My Drive\\1-PhD thesis\equitable relief routing\Code\%s\%s_BnPresult' %(Case_name,File_name) )
results[File_name]=[best_obj ,LB ,Runtime ,GAP]
#oldresult[File_name][0]= best_obj
#results=oldresult
Model_runtime=time()-start
#save_object(results,'G:\My Drive\\1-PhD thesis\\2 - equitable relief routing\Code\%s\%s_NewModel' %(Case_name,File_name) )
#save_object(results,'G:\My Drive\\1-PhD thesis\\2 - equitable relief routing\Code\%s\%s_Modelresult' %(Case_name,File_name) )
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
x_train = x_train.astype('float32')
x_train /= 255
x_test = x_test.astype('float32')
x_test /= 255
input_tensor = Input(shape=input_shape)
if args.model == 'Model1':
model = Model1(input_tensor=input_tensor)
elif args.model == 'Model2':
model = Model2(input_tensor=input_tensor, retrain_num=retrain_num)
elif args.model == 'Model3':
model = Model3(input_tensor=input_tensor, retrain_num=retrain_num)
elif args.model == 'Model4':
model = Model4(input_tensor=input_tensor, retrain_num=retrain_num)
elif args.model == 'Model5':
model = Model5(input_tensor=input_tensor, retrain_num=retrain_num)
# if args.model == 'Similar_Model1':
# model = Similar_Model1(input_tensor=input_tensor)
# elif args.model == 'Similar_Model2':
# model = Similar_Model2(input_tensor=input_tensor)
# elif args.model == 'Similar_Model3':
# model = Similar_Model3(input_tensor=input_tensor)
try:
videoLen_batch, labelLen_batch, video_batch, label_batch = sess.run(
[videoLens, labelLens, videos, labels])
label_batch = converLabelsToInt(utilDict, label_batch)
feedDict = m.get_feed_dict(videoLen_batch,
labelLen_batch,
video_batch,
label_batch,
isTrain=True)
loss, _, cer = sess.run([m.cost, m.train_op, m.cer],
feed_dict=feedDict)
print('Train: epoch:{}, step:{}, loss:{}, cer:{}'.format(
epoch, step, loss, cer))
step += 1
except tf.errors.OutOfRangeError:
if epoch and epoch % 5 == 0:
saver.save(sess,
ckptPath,
write_meta_graph=True,
global_step=epoch)
break
MAX_EPOCH = 40000
REPEAT = 1
BATCH_SIZE = 50
TRAIN_TFRECORD = 'put your tfrecord file here'
utilDict = rloader.loadUtilDict('utilDict.pkl')
m = Model1(utilDict)
train(TRAIN_TFRECORD, MAX_EPOCH, BATCH_SIZE, REPEAT, utilDict, m, '')