#~ model.add(MaxPooling2D(pool_size=(2, 2)))
#~ model.add(Dropout(0.2))
#~ model.add(Flatten())
#~ model.add(Dense(128, activation='relu'))
#~ model.add(Dense(base.nb_classes))
#~ model.add(Dense(base.nb_classes, activation='softmax'))
print("Compilando...")
model.compile(loss='categorical_crossentropy',
optimizer='adadelta',
metrics=['accuracy', top3])
model.summary()
#~ model.compile(loss='categorical_crossentropy',
#~ optimizer='adadelta',
#~ metrics=['accuracy' ])
trainer = base.Trainer('redMarianoVieja',
train_data=base.dataset("dataseth5/train.h5", "Train"),
valid_data=base.dataset("dataseth5/valid.h5", "Valid"),
test_data=base.dataset("dataseth5/test.h5", "Test"))
#~ trainer.train(model, nb_epoch=2, samples_per_epoch=10240, nb_val_samples=5000)
trainer.train(model, nb_epoch=100,
samples_per_epoch=269018) #usa todo el dataset
#~ trainer.train(model, nb_epoch=12, samples_per_epoch=269018) #usa todo el dataset
#~ trainer.train(model, nb_epoch=3, samples_per_epoch=100) #usa todo el dataset
#~ model = load_model('redMarianoPro--01-Nov-2016--10-40--model.h5')
trainer.evaluate(model)
from sys import argv
dropout = float(argv[1])
# number of convolutional filters to use
nb_filters = 32
# size of pooling area for max pooling
pool_size = (2, 2)
# convolution kernel size
kernel_size = (3, 3)
#~ trainer = base.Trainer('red_dropout'+ '-' + str(dropout), train_data=base.dataset("dataset/train", "Train"),
#~ valid_data=base.dataset("dataset/valid", "Valid"),
#~ test_data=base.dataset("dataset/test", "Test"))
trainer = base.Trainer('red_dropout' + '-' + str(dropout),
train_data=base.dataset("dataseth5/train.h5", "Train"),
valid_data=base.dataset("dataseth5/valid.h5", "Valid"),
test_data=base.dataset("dataseth5/test.h5", "Test"))
#~ trainer.train_data.preview()
print("Armando red...")
model = Sequential()
model.add(
Convolution2D(nb_filters,
kernel_size[0],
kernel_size[1],
border_mode='valid',
input_shape=base.input_shape))
model.add(Activation('relu'))
model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1]))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=pool_size))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(256))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(base.nb_classes))
model.add(Activation('softmax'))
print("Compilando...")
model.compile(loss='categorical_crossentropy',
optimizer='adadelta',
metrics=['accuracy', top3])
model.summary()
trainer = base.Trainer('red_martin',
train_data=base.dataset("dataseth5-featMean/train.h5",
"Train"),
valid_data=base.dataset("dataseth5-featMean/valid.h5",
"Valid"),
test_data=base.dataset("dataseth5-featMean/test.h5",
"Test"))
trainer.train(model, nb_epoch=30,
samples_per_epoch=269018) #usa todo el dataset
#~ trainer.train(model, nb_epoch=3, samples_per_epoch=128) #usa todo el dataset
trainer.save_last_train_history()
trainer.evaluate(model)
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution2D, MaxPooling2D
# number of convolutional filters to use
nb_filters = 32
# size of pooling area for max pooling
pool_size = (2, 2)
# convolution kernel size
kernel_size = (3, 3)
#~ trainer = base.Trainer('red_mas_densa', train_data=base.dataset("dataset/train", "Train"),
#~ valid_data=base.dataset("dataset/valid", "Valid"),
#~ test_data=base.dataset("dataset/test", "Test"))
trainer = base.Trainer('red_mas_densa', train_data=base.dataset("dataseth5/train.h5", "Train"),
valid_data=base.dataset("dataseth5/valid.h5", "Valid"),
test_data=base.dataset("dataseth5/test.h5", "Test"))
#~ trainer.train_data.preview()
print("Armando red...")
model = Sequential()
model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1],
border_mode='valid',
input_shape=base.input_shape))
model.add(Activation('relu'))
model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1]))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=pool_size))
model.add(Dropout(0.25))