CODEWITHSUNDEEP
pythonneural-networkkeras
Jiageng Zhu
Jiageng Zhu

Reputation: 39

keras load_weights() can not load weights

I want to load weights from .hdf5 file, and use load_weights(). Error doesn't occur. But when I use the model to predict. The result is the same with the model before I load weights.

The load weights doesn't work. My keras version is 2.2.2 tensorflow version is 1.10.0

How can I solve this problem. Thanks

Upvotes: 2

Views: 6528

Answers (2)

Jordan TheDodger
Jordan TheDodger

Reputation: 11

If you can share code on how you are saving model in.hd5 then we would be able to assist you. Assuming you are trying to solve a "Computer Vision" problem. Also if needed you can print shape after input and pooling layer It is best practice to save weights instead of whole model as mentioned in Keras docs. Hope this helps

For eg: I have used modelcheckpoint to load a model (or feature extractor model). Also note below code is in functional API.

# create a base model(eg: EfficientNetB0 and so on)
base_model = tf.keras.applications.EfficientNetB0(include_top=False)
# include_top=False allow setting base model per your problem

# freeze layer
base_model.trainable=False

# create inputs
inputs = tf.keras.layers.Input(shape=(224,224,3), name="input_layer")

# pass inputs to base model
x = base_model(inputs)

# perform pooling
x = tf.keras.layers.GlobalAveragePooling2D(name="pooling_layer")(x)


# create outputs
outputs = tf.keras.layers.Dense(10, activation="softmax",name="output_layer")(x)
# need to modify classes per your problem

# combine inputs and outputs with base model
model = tf.keras.Model(inputs,outputs)

# compile model
.... 


#fit model
model.fit(train_data, epochs=5,validation_data=test_data,validation_steps=len(test_data),callbacks=[tf.keras.callbacks.ModelCheckpoint(ckhpt_filepath,save_weights=True,save_best_only=True,save_freq="epoch",verbose=1)

# load best weights and evaluate model
model.load_weights(ckhpt_filepath)
loaded_model_results = model.evaluate(test_data)

Upvotes: 0

B. Kanani
B. Kanani

Reputation: 646

Have you saved weight and load weight like this? 

from keras.models import Sequential
from keras.layers import Dense
from keras.models import model_from_json 

model = Sequential()
model.add(Dense(12, input_dim=8, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(1,activation='sigmoid'))

Compile model

model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

Fit the model & evaluate

model.fit(X, Y, epochs=150, batch_size=10, verbose=0)
scores = model.evaluate(X, Y, verbose=0)
print("%s: %.2f%%" % (model.metrics_names[1], scores[1]*100))

serialize model to JSON

model_json = model.to_json()
with open("model.json", "w") as json_file:
    json_file.write(model_json)

serialize weights to HDF5

model.save_weights("model.h5")
print("Saved model to disk")

later... load json and create model

json_file = open('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("model.h5")
print("Loaded model from disk")

evaluate loaded model on test data

loaded_model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
score = loaded_model.evaluate(X, Y, verbose=0)
print("%s: %.2f%%" % (loaded_model.metrics_names[1], score[1]*100))

Upvotes: 3

Related Questions