struct.use_dropout = True
struct.dropout_value = 0.1 * (
j + 1) # Pour avoir des dropouts à 0.1, puis 0.2, puis 0.3
struct.dropout_indexes = [
x + 1 for x in range(struct.nb_hidden_layers)
]
model = [create_custom_mlp(struct)]
desc = [getMlpStructAsString(struct)]
print(desc[0])
test_models('mlp_best_models_with_l2_and_dropout',
model,
desc,
train_data,
train_labels,
val_data,
val_labels,
epochs_p=epochs,
batch_size_p=8192)
clear_session() #clear tf GPU memory
for j in range(3):
struct.use_l1l2_regularisation_hidden_layers = True
struct.regulization_indexes = [
x + 1 for x in range(struct.nb_hidden_layers)
]
struct.l2_value = 0.002
struct.l1_value = 0.0
struct.use_dropout = True
from cifar10.models.ModelTester import test_models
from tensorflow.keras.datasets import cifar10
from cifar10.models.UNet import *
from tensorflow.keras.backend import clear_session
if __name__ == "__main__":
(train_data, train_labels), (val_data, val_labels) = cifar10.load_data()
epochs = [40]
for i in range(3):
struct = generateRandoUNetStruc(min_nb_layers=7, max_nb_layers=7)
struct.use_MaxPooling2D = True
struct.MaxPooling2D_position = [1, 2, 3]
model = [create_unet(struct)]
desc = [getUNetStructAsString(struct)]
print(desc[0])
test_models('model_3',
model,
desc,
train_data,
train_labels,
val_data,
val_labels,
epochs_p=epochs,
batch_size_p=256,
save_image=True,
save_model=True)
clear_session()
# [0.025], [True], [True], [0.012], [0.025], [[6]],
# ['sparse_categorical_crossentropy'], ['Adam'], [['accuracy']])
#
# test_models('cnn_8_8_OOOO_32_128', modelcnn, desccnn, train_images, train_labels, test_images, test_labels,
# epochs_p=epochs,
# batch_size_p=256, save_image=True)
# clear_session()
#testes 28/01/2020 08h22
epochs=[200]
(modelcnn, desccnn) = generateCNNModels([8], [[(128, 4),(128, 4),(64, 4),(64, 4),(64, 4),(32, 4),(32, 4),(32, 4)]],
['relu'], ['softmax'], ['same'], [True], [[1, 4]], [4], [True], [[0, 4, 6]],
[0.025], [True], [True], [0.012], [0.025], [[2,7]],
['sparse_categorical_crossentropy'], ['Adam'], [['accuracy']])
test_models('cnn_8_8_OOOO_32_128', modelcnn, desccnn, train_images, train_labels, test_images, test_labels,
epochs_p=epochs,
batch_size_p=256, save_image=True)
clear_session()
(modelcnn, desccnn) = generateCNNModels([8], [
[(128, 4), (128, 4), (64, 4), (64, 4), (64, 4), (32, 4), (32, 4), (32, 4)]],
['relu'], ['softmax'], ['same'], [True], [[1, 4]], [4], [True], [[0, 4, 6]],
[0.028], [True], [True], [0.012], [0.025], [[2,3,7]],
['sparse_categorical_crossentropy'], ['Adam'], [['accuracy']])
test_models('cnn_8_8_OOOO_32_128', modelcnn, desccnn, train_images, train_labels, test_images, test_labels,
epochs_p=epochs,
batch_size_p=256, save_image=True)
clear_session()
(modelcnn, desccnn) = generateCNNModels([8], [
clear_session()
struct = generateRandoCNNStruc(use_maxpool=False,
use_dropout=False,
use_l1l2_conv=True,
use_l1l2_output=False,
min_nb_layers=3,
max_nb_layers=8,
min_filter_size=32,
max_filter_size=64)
model = [create_CNN_model(struct)]
desc = [getcnnStructAsString(struct)]
test_models('cnn_3_8_NNON_32_64',
model,
desc,
train_images,
train_labels,
test_images,
test_labels,
epochs_p=epochs,
batch_size_p=256,
save_image=True)
del model
del desc
# ######################## CNN 3-8 O O O O 32 64 #########################
for i in range(3):
clear_session()
struct = generateRandoCNNStruc(use_maxpool=True,
use_dropout=True,
use_l1l2_conv=True,
use_l1l2_output=True,
min_nb_layers=3,
max_nb_layers=8,
from tensorflow.keras.datasets import *
from cifar10.models.LIN import create_lin_model, getLinStructAsString
from cifar10.models.ModelTester import test_models
from cifar10.models.structurer.LinearStructurer import LinearStructurer
if __name__ == "__main__":
(train_data, train_labels), (val_data, val_labels) = cifar10.load_data()
classes = [
'plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship',
'truck'
]
epochs = [100]
struct = LinearStructurer()
desc = [getLinStructAsString(struct)]
linear_model = [create_lin_model(struct)]
test_models('lin',
linear_model,
desc,
train_data,
train_labels,
val_data,
val_labels,
epochs_p=epochs,
batch_size_p=1024,
save_model=True)
'airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog',
'horse', 'ship', 'truck'
]
#script du 06/02/2020:
for e in (30, 35):
(modelcnn, desccnn) = generateCNNModels(
[3], [[(128, 3), (64, 3), (32, 3)]], ['relu'], ['softmax'],
['same'], [True], [[2]], [3], [True], [[0]], [0.25], [True],
[True], [0.030], [0.090], [[1, 3]],
['sparse_categorical_crossentropy'], ['Adam'], [['accuracy']])
test_models('cnn_1_3_OOOO_32_128',
modelcnn,
desccnn,
train_images,
train_labels,
test_images,
test_labels,
epochs_p=[e],
batch_size_p=256,
save_image=True)
clear_session()
# ##Scirpt du 26 01 2020 11h40:
# # ######################## CNN 1-3 N N N N 32 64 #########################
# for i in range(3):
# clear_session()
# struct = generateRandoCNNStruc(use_maxpool=False, use_dropout=False, use_l1l2_conv=False, use_l1l2_output=False,
# min_nb_layers=1, max_nb_layers=3, min_filter_size=32, max_filter_size=64)
# model = [create_CNN_model(struct)]
# desc = [getcnnStructAsString(struct)]
# test_models('cnn_1_3_NNNN_32_64', model, desc, train_images, train_labels, test_images, test_labels,
from __future__ import absolute_import, division, print_function, unicode_literals
from tensorflow.keras import layers
from tensorflow.keras.backend import clear_session
import tensorflow as tf
from cifar10.models.CNN import *
from cifar10.models.ModelTester import test_models
if __name__ == "__main__":
#déclaration des param:
epochs=[30]
#importation des données:
(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.cifar10.load_data()
#définition des noms des classes:
class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer',
'dog', 'frog', 'horse', 'ship', 'truck']
epochs=[50]
for i in range(3):
clear_session()
cnn_str=generateRandoCNNStruc(use_maxpool=False,use_l1l2_conv=False,use_l1l2_output=True,use_dropout=False,min_nb_layers=3,max_nb_layers=4,min_filter_size=32,max_filter_size=128)
cnn_model=[create_CNN_model(cnn_str)]
cnn_desc=[getcnnStructAsString(cnn_str)]
test_models('cnn_3_4_N_N_O_N_32_128', cnn_model, cnn_desc, train_images, train_labels, test_images, test_labels,
epochs_p=epochs,
batch_size_p=1024, save_image=True)
del cnn_model
del cnn_desc
# test_models('cnn_5_5_ONON_32128', modelcnn, desccnn, train_images, train_labels, test_images, test_labels,
# epochs_p=[e],
# batch_size_p=256, save_image=True)
# clear_session()
#tested on 8/2/2020 à 12h08
#5;(64, 3) (128, 3) (64, 3) (32, 3) (32, 3);selu;softmax;same;True;3;2 4;False;0;0;True;False;0.005;0.06;1 3
e=[45]
for i in (0.15,0.25,0.35,0.45):
(modelcnn, desccnn) = generateCNNModels([5], [
[(64, 3), (128, 3),(64, 3), (32, 3) ,(32, 3)]],
['selu'], ['softmax'], ['same'], [True], [[2,4]], [3], [False], [[0]],
[0], [True], [True], [i], [0.099], [[1,3]],
['sparse_categorical_crossentropy'], ['Adam'], [['accuracy']])
test_models('cnn_5_5_OOON_32128', modelcnn, desccnn, train_images, train_labels, test_images, test_labels,
epochs_p=[e],
batch_size_p=256, save_image=True)
clear_session()
for e in (30,35,40,45,50):
(modelcnn, desccnn) = generateCNNModels([6], [
[(64, 3), (128, 3), (128, 3) ,(64, 3), (32, 3) ,(32, 3)]],
['relu'], ['softmax'], ['same'], [True], [[4,5]], [3], [True], [[0]],
[0.35], [True], [True], [0.05], [0.095], [[1,4]],
['sparse_categorical_crossentropy'], ['Adam'], [['accuracy']])
test_models('cnn_6_6_OOOO_32128', modelcnn, desccnn, train_images, train_labels, test_images, test_labels,
epochs_p=[e],
batch_size_p=256, save_image=True)
clear_session()
for e in (25,30,35,40,45,50):