var_unet = Model(inputs, outputs)
var_unet.summary()
#[Keras Optimizer, Loss & Metrics] ------------------------------------------
optimizer = {
"optimizer" : keras.optimizers.Adam(lr=1e-3),
"loss" : ezlosses.vae_loss(z_mean,z_log_var,train.X.shape[1:]),
# "metrics" : [ezlosses.reconstruction_loss]
}
# # [EZMODEL] ------------------------------------------------------------------
ez = ezmodel(
train = train,
test = test,
network = var_unet,
optimizer = optimizer,
transformers = transformers
)
# # # Training --------------------------------------------------------------------
parameters = {
"epochs" : 40,
# "validation_split": 0.2
}
ez.train(parameters)
p = ez.predict()
r = show_images(p,4)
import sys, os
sys.path.append(os.path.abspath('..\\..\\ezmodel'))
from ezmodel.ezmodel import ezmodel
from ezmodel.ezdata import ezdata
from ezmodel.eztrainer import eztrainer
# [EZ MODEL initialization]
ez_model = ezmodel(type="classification")
print(ez_model.type)
# ---------------------------- [EZ Data] -----------------------------------
# Table Classification from a csv file:
# - One subdirectory by Class
parameters = {
"name": "Bacteria CSV",
"path": "C:\\Users\\daian\\Desktop\\DATA\\bacteria_csv\\bacteria.csv",
"table.target.column": "Label",
}
ez_data = ezdata(parameters)
ez_data.gen_test(size=0.2)
ez_data.preprocess(X="standard",
y="categorical") #on crée les scaler dans ez_data aussi
# -------------------------- [EZ Trainer] ------------------------------------
ez_trainer = eztrainer()
ez_trainer.gen_trainval(ez_data, size=0.2)
def load_ezmodel(filename):
if not os.path.isfile(filename + ".zip"):
raise Exception("[Fail] ezmodel(load) : ", filename,
".zip has not been found !")
ez = ezmodel(empty=True)
zip_ref = ZipFile(filename + ".zip", 'r')
zip_ref.extractall(".")
zip_ref.close()
filehandler = open(filename + ".data.pkl", 'rb')
tmp = pickle.load(filehandler)
filehandler.close()
ez.data_train = tmp[0]
ez.data_test = tmp[1]
filehandler = open(filename + ".trans.pkl", 'rb')
tmp = pickle.load(filehandler)
filehandler.close()
ez.transformerX = tmp[0]
ez.transformerY = tmp[1]
filehandler = open(filename + ".params.pkl", 'rb')
tmp = pickle.load(filehandler)
filehandler.close()
ez.model_parameters = tmp
filehandler = open(filename + ".hist.pkl", 'rb')
tmp = pickle.load(filehandler)
filehandler.close()
ez.history = tmp
filehandler = open(filename + ".aug.pkl", 'rb')
tmp = pickle.load(filehandler)
filehandler.close()
ez.augmentation = tmp
#Network + optimizer + custom losses
import ezmodel.ezlosses as tmp
import types
l = [
getattr(tmp, a) for a in dir(tmp)
if isinstance(getattr(tmp, a), types.FunctionType)
]
values = l
keys = []
for i in l:
keys.append((i.__name__))
custom_objects = dict(zip(keys, values))
ez.network = load_model(filename + ".h5", custom_objects=custom_objects)
os.remove(filename + ".h5")
os.remove(filename + ".data.pkl")
os.remove(filename + ".trans.pkl")
os.remove(filename + ".params.pkl")
os.remove(filename + ".hist.pkl")
os.remove(filename + ".aug.pkl")
print("[X]Ezmodel loaded successfully !")
return ez
import sys, os
sys.path.append(os.path.abspath('..\\..\\ezmodel'))
from ezmodel.ezmodel import ezmodel
from ezmodel.ezdata import ezdata
from ezmodel.eztrainer import eztrainer
# [EZ MODEL loading]
ez_model = ezmodel(load = "model0")
print("Model parameters = ",ez_model.model_parameters)
# [EZ evaluation]
ez_model.evaluate()
optimizer = {
"optimizer": keras.optimizers.Adam(lr=1e-4),
"loss": keras.losses.categorical_crossentropy,
"metrics": [keras.metrics.categorical_accuracy]
}
# [EZMODEL] ------------------------------------------------------------------
augmentation_parameters = {
"rotation_range": 15,
"width_shift_range": .15,
"height_shift_range": .15,
"horizontal_flip": True
}
ez = ezmodel(train=train,
test=test,
network=net,
optimizer=optimizer,
transformers=transformers,
augmentation=augmentation_parameters)
# Training --------------------------------------------------------------------
parameters = {
"epochs": 200,
# "validation_split": 0.2
}
ez.train(parameters)
# Evaluation ------------------------------------------------------------------
ez.evaluate()
# Evaluation already defined augmentation for Test Time Augmentation (TTA)
ez.evaluate(tta=10)
# sys.exit()
# # [EZNETWORK] ----------------------------------------------------------------
from ezmodel.eznetwork import LeNet5
net = LeNet5(input=train)
net.summary()
# [Keras Optimizer, Loss & Metrics] ------------------------------------------
optimizer = {
"optimizer": keras.optimizers.Adam(lr=1e-4),
"loss": keras.losses.categorical_crossentropy,
"metrics": [keras.metrics.categorical_accuracy]
}
# net.compile(**optimizer)
# [EZMODEL] ------------------------------------------------------------------
ez = ezmodel(train=train, test=test, network=net, optimizer=optimizer)
# Training --------------------------------------------------------------------
parameters = {
"epochs": 50,
}
ez.train(parameters)
# Evaluation ------------------------------------------------------------------
#ez.evaluate()
# p = ez.network.predict_generator(ez.data_test.generator,steps=ez.data_test.generator.n//ez.data_test.generator.batch_size)
# print(p.argmax(axis=1))
# p=net.predict_generator(test_gen,steps=test_gen.n//test_gen.batch_size)
# print(p)