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'])
x_train, y_train, x_test, y_test = U.get_mnist()
#train on a subset only
x_train = x_train[:1000]
y_train = y_train[:1000]
epsilon = 2e-4 # step size
tau = 200 # number of steps to take before the reject/accept step
burn_in = 100
sample_every = 30
N_ensemble = 20 #number of models to create
N_restarts = 5 #use multiple intitialisations
# use multiple intitialisations
ensemble = []
with open('save/tmp/losses.dat', 'w') as f:
print('', f)
Dense(128, activation=act_fn),
weight_regularizer=WEIGHT_REGULARIZER,
dropout_regularizer=DROPOUT_REGULARIZER,
))
model.add(
ConcreteDropout(
Dense(N_CLASSES, activation='softmax'),
weight_regularizer=WEIGHT_REGULARIZER,
dropout_regularizer=DROPOUT_REGULARIZER,
))
return model
if __name__ == "__main__":
x_train, y_train, x_test, y_test = mnist_to_3s_and_7s(U.get_mnist())
# mnist, scaled to the range 0,1.
epochs = 50
batch_size = 128
model = define_cdropout_3s_7s()
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
validation_data=(x_test, y_test),
callbacks=[TrackConcreteDropoutP(model)
import src.utilities as U
from latent_plots import get_models, visualise_latent_space
plt.rcParams['figure.figsize'] = 8, 5
#use true type fonts only
plt.rcParams['pdf.fonttype'] = 42
plt.rcParams['ps.fonttype'] = 42
if __name__ == '__main__':
model, encoder, decoder = get_models()
#move along a random line in latent space
_, _, mnist, label = U.get_mnist()
x1 = mnist[label.argmax(axis=1) == 6][200]
x2 = mnist[label.argmax(axis=1) == 8][200]
x_ims = np.stack([(1 - t) * x1 + t * x2 for t in np.linspace(0, 1, 15)])
x_preds, x_entropy, x_bald = model.get_results(x_ims)
z_begin = encoder.predict(x1[None, :]).flatten()
z_end = encoder.predict(x2[None, :]).flatten()
z_lin = np.stack([(1 - t) * z_begin + t * z_end
for t in np.linspace(0, 1, 15)])
z_ims = decoder.predict(z_lin)