python - learning - ¿Cómo cargar un modelo desde un archivo HDF5 en Keras?

keras python documentation

Question

¿Cómo cargar un modelo desde un archivo HDF5 en Keras?

Lo que probé:

model = Sequential()
model.add(Dense(64, input_dim=14, init=''uniform''))
model.add(LeakyReLU(alpha=0.3))
model.add(BatchNormalization(epsilon=1e-06, mode=0, momentum=0.9, weights=None))
model.add(Dropout(0.5))
model.add(Dense(64, init=''uniform''))
model.add(LeakyReLU(alpha=0.3))
model.add(BatchNormalization(epsilon=1e-06, mode=0, momentum=0.9, weights=None))
model.add(Dropout(0.5))
model.add(Dense(2, init=''uniform''))
model.add(Activation(''softmax''))
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss=''binary_crossentropy'', optimizer=sgd)
checkpointer = ModelCheckpoint(filepath="/weights.hdf5", verbose=1, save_best_only=True)
model.fit(X_train, y_train, nb_epoch=20, batch_size=16, show_accuracy=True, validation_split=0.2, verbose = 2, callbacks=[checkpointer])

El código anterior guarda con éxito el mejor modelo en un archivo llamado weights.hdf5. Lo que quiero hacer es cargar ese modelo. El siguiente código muestra cómo intenté hacerlo:

model2 = Sequential()
model2.load_weights("/Users/Desktop/SquareSpace/weights.hdf5")

Este es el error que obtengo:

IndexError Traceback (most recent call last)
<ipython-input-101-ec968f9e95c5> in <module>()
1 model2 = Sequential()
----> 2 model2.load_weights("/Users/Desktop/SquareSpace/weights.hdf5")
/Applications/anaconda/lib/python2.7/site-packages/keras/models.pyc in load_weights(self, filepath)
582 g = f[''layer_{}''.format(k)]
583 weights = [g[''param_{}''.format(p)] for p in range(g.attrs[''nb_params''])]
--> 584 self.layers[k].set_weights(weights)
585 f.close()
586
IndexError: list index out of range

Answer 1

Consulte el siguiente código de muestra sobre cómo construir un modelo básico de red neuronal Keras, guardar el modelo (JSON) y los pesos (HDF5) y cargarlos:

# create model
model = Sequential()
model.add(Dense(X.shape[1], input_dim=X.shape[1], activation=''relu'')) #Input Layer
model.add(Dense(X.shape[1], activation=''relu'')) #Hidden Layer
model.add(Dense(output_dim, activation=''softmax'')) #Output Layer
# Compile & Fit model
model.compile(loss=''binary_crossentropy'', optimizer=''adam'', metrics=[''accuracy''])
model.fit(X,Y,nb_epoch=5,batch_size=100,verbose=1)
# serialize model to JSON
model_json = model.to_json()
with open("Data/model.json", "w") as json_file:
json_file.write(simplejson.dumps(simplejson.loads(model_json), indent=4))
# serialize weights to HDF5
model.save_weights("Data/model.h5")
print("Saved model to disk")
# load json and create model
json_file = open(''Data/model.json'', ''r'')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights("Data/model.h5")
print("Loaded model from disk")
# evaluate loaded model on test data
# Define X_test & Y_test data first
loaded_model.compile(loss=''binary_crossentropy'', optimizer=''adam'', metrics=[''accuracy''])
score = loaded_model.evaluate(X_test, Y_test, verbose=0)
print ("%s: %.2f%%" % (loaded_model.metrics_names[1], score[1]*100))

Answer 2

De acuerdo con la documentación oficial https://keras.io/getting-started/faq/#how-can-i-install-hdf5-or-h5py-to-save-my-models-in-keras

tu puedes hacer :

primera prueba si tiene instalado h5py ejecutando el

importar h5py

si no tiene errores al importar h5py, puede guardarlos:

from keras.models import load_model
model.save(''my_model.h5'') # creates a HDF5 file ''my_model.h5''
del model # deletes the existing model
# returns a compiled model
# identical to the previous one
model = load_model(''my_model.h5'')

Si necesita instalar h5py http://docs.h5py.org/en/latest/build.html

Answer 3

Lo hice de esta manera

from keras.models import Sequential
from keras_contrib.losses import import crf_loss
from keras_contrib.metrics import crf_viterbi_accuracy
# To save model
model.save(''my_model_01.hdf5'')
# To load the model
custom_objects={''CRF'': CRF,''crf_loss'': crf_loss,''crf_viterbi_accuracy'':crf_viterbi_accuracy}
# To load a persisted model that uses the CRF layer
model1 = load_model("/home/abc/my_model_01.hdf5", custom_objects = custom_objects)

Answer 4

Si almacenó el modelo completo, no solo los pesos, en el archivo HDF5, entonces es tan simple como

from keras.models import load_model
model = load_model(''model.h5'')

Answer 5

load_weights solo establece los pesos de su red. Aún necesita definir su arquitectura antes de llamar a load_weights :

def create_model():
model = Sequential()
model.add(Dense(64, input_dim=14, init=''uniform''))
model.add(LeakyReLU(alpha=0.3))
model.add(BatchNormalization(epsilon=1e-06, mode=0, momentum=0.9, weights=None))
model.add(Dropout(0.5))
model.add(Dense(64, init=''uniform''))
model.add(LeakyReLU(alpha=0.3))
model.add(BatchNormalization(epsilon=1e-06, mode=0, momentum=0.9, weights=None))
model.add(Dropout(0.5))
model.add(Dense(2, init=''uniform''))
model.add(Activation(''softmax''))
return model
def train():
model = create_model()
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss=''binary_crossentropy'', optimizer=sgd)
checkpointer = ModelCheckpoint(filepath="/tmp/weights.hdf5", verbose=1, save_best_only=True)
model.fit(X_train, y_train, nb_epoch=20, batch_size=16, show_accuracy=True, validation_split=0.2, verbose=2, callbacks=[checkpointer])
def load_trained_model(weights_path):
model = create_model()
model.load_weights(weights_path)