def get_HMC_models():
_, encoder, decoder = define_VAE()
encoder.load_weights('save/enc_weights.h5')
decoder.load_weights('save/dec_weights.h5')
input_shape = (28, 28, 1)
num_classes = 10
def modeldef():
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
activation='relu',
input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.SGD(),
metrics=['accuracy'])
return model
class HMC_model():
def __init__(self, ensemble_weights):
self.m = modeldef()
self.ws = ensemble_weights
self.ms = [modeldef() for _ in self.ws]
for model, wlist in zip(self.ms,self.ws):
for tensor, weight in zip(model.weights, wlist):
K.set_value(tensor, weight)
def get_results(self,X):
mc_preds = mcmc.HMC_ensemble_predict(self.m, self.ws, X)
preds = np.mean(mc_preds, axis=0)
predictive_entropy = H(preds)
expected_entropy = np.mean(H(mc_preds), axis=0)
minfo = predictive_entropy - expected_entropy
return preds, predictive_entropy, minfo
def predict(self,X):
mc_preds = mcmc.HMC_ensemble_predict(self.m, self.ws, X)
preds = np.mean(mc_preds, axis=0)
return preds
def __call__(self, X):
"""get predictions on a symbolic tensor; this is a little inefficient in terms of space"""
return K.mean( K.stack([ model(X) for model in self.ms]), axis=0)
#load Y's weights
with open('save/mnist_hmc_ensemble_run_1.pkl', 'rb') as pkl:
weights = pickle.load(pkl)
with open('save/mnist_hmc_ensemble_run.pkl', 'rb') as pkl:
weights += pickle.load(pkl)
hmc_model = HMC_model(weights)
return hmc_model, encoder, decoder
def get_models():
_, encoder, decoder = define_VAE()
encoder.load_weights('save/enc_weights.h5')
decoder.load_weights('save/dec_weights.h5')
model = keras.models.load_model('save/mnist_cnn_run_3.h5')
mc_model = U.MCModel(model, model.input, n_mc=50)
#we have been using more mc samples elsewhere, but save time for now
return mc_model, encoder, decoder
def get_models_3s_7s():
_, encoder, decoder = define_VAE()
encoder.load_weights('save/enc_weights_3s_7s.h5')
decoder.load_weights('save/dec_weights_3s_7s.h5')
model = define_cdropout_3s_7s()
model.load_weights('save/mnist_cdrop_3s_7s.h5')
mc_model = U.MCModel(model, model.input, n_mc=50)
#we have been using more mc samples elsewhere, but save time for now
return mc_model, encoder, decoder
def get_model_ensemble(n_mc=10):
_, encoder, decoder = define_VAE()
encoder.load_weights('save/enc_weights.h5')
decoder.load_weights('save/dec_weights.h5')
models = []
for name in filter(lambda x: 'mnist_cnn' in x, os.listdir('save')):
print('loading model {}'.format(name))
model = load_drop_model('save/' + name)
models.append(model)
mc_model = U.MCEnsembleWrapper(models, n_mc=10)
return mc_model, encoder, decoder
def get_ML_models():
_, encoder, decoder = define_VAE()
encoder.load_weights('save/enc_weights.h5')
decoder.load_weights('save/dec_weights.h5')
model = keras.models.load_model('save/mnist_cnn.h5')
def get_results(X):
preds = model.predict(X)
ent = - np.sum(preds * np.log(preds + 1e-10), axis=-1)
return preds, ent, np.zeros(ent.shape)
model.get_results = get_results
return model, encoder, decoder
def get_ML_ensemble():
_, encoder, decoder = define_VAE()
encoder.load_weights('save/enc_weights.h5')
decoder.load_weights('save/dec_weights.h5')
K.set_learning_phase(False)
ms = []
for name in filter(lambda x: 'mnist_cnn' in x, os.listdir('save')):
print('loading model {}'.format(name))
model = load_model('save/' + name)
ms.append(model)
model = U.MCEnsembleWrapper(ms, n_mc=1)
return model, encoder, decoder