def evaluate(individual, x_train_reduced, y_train_reduced, x_test_reduced,
y_test_reduced, x_val_reduced, y_val_reduced, MATRIX):
model = cifar100vgg(train=0) # build the model
model.load(
'R:\\coco_dataset\\dati_salvati\\cifar100_baseline.h5'
) #load the pre trained model, change the path to the saved model
_ = model.selection_pruning(individual) #
score_val, score_test = model.train(x_train_reduced, y_train_reduced,
x_test_reduced, y_test_reduced,
x_val_reduced, y_val_reduced, 3,
'R:\\')
total_score = 0.6 * score_val + 0.4 * (sum(individual))
# total_score=5
# score_val=2
# score_test=3
MATRIX.append([individual, score_val, score_test, total_score])
# gen_matrix.append([mega_list,score,sum(individual),total_score])
del model # delete the model and clear the memory
K.clear_session()
tf.reset_default_graph()
return total_score, MATRIX
def random_pruning(P,num): # total random pruning
matrix=[]
# model.train(model,x_train_reduced,y_train_reduced,x_test_reduced,y_test_reduced,3)
for i in range(num):
print('iteration number:',i)
model = cifar100vgg(train=0) # build the model
model.load('R:\mygithub\\'+'cifar100_baseline.h5') #load the pre trained model, change the path to the saved model
matrix=model.random_pruning(matrix,P)
score=model.evaluate(x_train_reduced,y_train_reduced,x_test_reduced,y_test_reduced,x_val_reduced,y_val_reduced,3)
_,score=model.train(model,x_train_reduced,y_train_reduced,x_test_reduced,y_test_reduced,x_val_reduced,y_val_reduced,3)
matrix[-1].append(score)
del model # delete the model and clear the memory
K.clear_session()
tf.reset_default_graph()
path='R:\\matrix_{}'.format(int(P*100)) #path to the saved matrix folder
if not os.path.exists(path):
os.mkdir(path)
# np.save(path+'\\'+'matrix_{}_{}'.format(int(P*100),P,time.time()),matrix) # save the matrix containing all the filters and score to file
filters=[]
result=[]
for i in os.listdir(path):
final_matrix=np.load(os.path.join(path,i),allow_pickle=True)
for k in range(len(final_matrix)): # ciclo su tutte le prove
filters.append(final_matrix[k][0:-2])
result.append(final_matrix[k][-1])
idx=np.array(result)
idx=idx[:,-1]
idx=(-idx).argsort()
filterS=np.array(filters)[idx]
resulT=np.array(result)[idx]
new_path=path.replace('{}'.format(int(P*100)),'')
if not os.path.exists(new_path):
os.mkdir(new_path)
np.save(new_path+'\\'+'result_{}'.format(P),resulT)
np.save(new_path+'\\'+'filters_{}'.format(P),filterS)
time.sleep(2)
if debug == 1:
print('lavoro trovato, leggo files...')
time.sleep(2)
pop_list = np.load(os.path.join(operation_path, job_to_do),
allow_pickle=True) # carico la lista di lavoro
job_to_do = job_to_do.replace('.npy', '')
var = job_to_do[-3::]
if debug == 1:
print('var=', var)
os.remove(os.path.join(
operation_path, job_to_do +
'.npy')) # ho letto la lista del lavoro da fare, quindi la elimino
for idx, i in enumerate(pop_list):
model = cifar100vgg(train=0)
model.load(
os.path.join(operation_path, 'cifar100_baseline.h5')
) #load the pre trained model, change the path to the saved model
_ = model.selection_pruning(i)
model.give_name(
str(i)
) # i è un individuo che è stato castato a lista, tuttavia funziona list(individuo)==individuo
model.save(
os.path.join(operation_path, 'job_{}_{}.h5'.format(var, idx)))
del model # delete the model and clear the memory
K.clear_session()
tf.reset_default_graph()
pop_list)
for idx, i in enumerate(pop):
while not os.path.isfile(
os.path.join(operation_path, 'job_{}_{}.h5'.format(var, idx))):
time.sleep(2) # aspetto che il secondo kernel effettui il pruning
if debug == 1:
print('attendo svolgimento lavoro {}'.format(idx))
if debug == 1:
print('lavoro {} svolto'.format(idx))
print('individuo numero{} su {}'.format(idx, len(pop)))
model = cifar100vgg(train=0) # build the model
model.load(
os.path.join(operation_path, 'job_{}_{}.h5'.format(var, idx)))
ind_name = model.return_name()
# if debug==1:
# if str(list(i)) !=ind_name:
# print('ATTENZIONE INDIVIDUO DIVERSO')
# time.sleep(100)
# elif str(list(i)) ==ind_name:
# print('individuo corretto')
# time.sleep(5)
# ind_name=model.return_name()
# indice=pop.index(ind_name)
# print('indice=',indice)