def main():
units = 512
encoding_units = 128
z_k = 10
input_units = 28 * 28
pz_regularizer = BalanceRegularizer(1e-1)
reg_weight_encoding = 1e-1
gen_regularizer = L1L2(1e-2, 1e-2)
((x, y), (xt, yt)) = mnist.load_data()
x = np.float32(x) / 255.
x = np.reshape(x, (x.shape[0], -1))
initializer = RandomUniform(minval=-0.05, maxval=0.05)
classifier = MLP(input_units=input_units,
hidden_units=units,
output_units=z_k,
hidden_depth=2,
hidden_activation=leaky_relu,
initializer=initializer,
output_activation=softmax_nd)
encoder = MLP(input_units=units,
hidden_units=units,
output_units=encoding_units,
hidden_depth=2,
hidden_activation=leaky_relu,
initializer=initializer,
output_activation=leaky_relu)
generator = MLP(input_units=encoding_units,
hidden_units=units,
output_units=input_units,
hidden_depth=2,
hidden_activation=leaky_relu,
initializer=initializer,
output_activation=T.nnet.sigmoid)
model = MSEModel(
z_k=z_k,
mode=2,
classifier=classifier,
encoder=encoder,
generator=generator,
units=units,
encoding_units=encoding_units,
opt=Adam(1e-3),
input_units=input_units,
reg_weight_encoding=reg_weight_encoding,
activation=leaky_relu,
initializer=initializer,
pz_regularizer=pz_regularizer,
gen_regularizer=gen_regularizer
)
model.train(
x=x,
output_path='output/mnist_clustering_mse',
epochs=500,
batches=10000,
batch_size=128
)
def main():
units = 512
rng_units = 128
z_k = 10
pz_regularizer = BalanceRegularizer(1e-2)
iwgan_weight = 1e-1
initializer = RandomUniform(minval=-0.05, maxval=0.05)
((x, y), (xt, yt)) = mnist.load_data()
x = np.float32(x) / 255.
x = np.reshape(x, (x.shape[0], -1))
input_units = 28 * 28
classifier = MLP(input_units=input_units,
hidden_units=units,
output_units=z_k,
hidden_depth=2,
hidden_activation=leaky_relu,
initializer=initializer,
output_activation=softmax_nd)
generator = MLP(input_units=units,
hidden_units=units,
output_units=input_units,
hidden_depth=2,
hidden_activation=leaky_relu,
initializer=initializer,
output_activation=T.nnet.sigmoid)
discriminator = MLP(input_units=units,
hidden_units=units,
output_units=1,
hidden_depth=2,
initializer=initializer,
hidden_activation=leaky_relu)
model = FCGAN(
z_k=z_k,
classifier=classifier,
generator=generator,
discriminator=discriminator,
optd=Adam(1e-3),
optg=Adam(1e-3),
input_units=input_units,
rng_units=rng_units,
units=units,
activation=leaky_relu,
pz_regularizer=pz_regularizer,
initializer=initializer,
iwgan_weight=iwgan_weight
)
model.train(
x=x,
output_path='output/mnist_clustering',
epochs=500,
batches=512,
discriminator_batches=5,
batch_size=128
)
def main():
epochs = 10
batches = 4096
z_k = 1024
iters = 5
outputpath = "output/skipgram_flat-b"
cooccurrence = np.load('output/cooccurrence.npy').astype(np.float32)
weights = [1e1, 1, 1e-1, 1e-2, 1e-3, 5e-4, 1e-4, 5e-5, 1e-5, 1e-6, 1e-7]
regularizers = [BalanceRegularizer(w) for w in weights]
labels = ["b-{:.01e}".format(w) for w in weights]
train_flat_regularizer_battery(outputpath=outputpath,
cooccurrence=cooccurrence,
epochs=epochs,
batches=batches,
iters=iters,
z_k=z_k,
labels=labels,
regularizers=regularizers,
is_weight_regularizer=False,
kwdata={'weights': np.array(weights)})
def main():
epochs = 1000
batches = 4096
z_k = 256
outputpath = "output/skipgram_256_m1"
cooccurrence = np.load('output/cooccurrence.npy').astype(np.float32)
scale = 1e-2
opt = Adam(1e-3)
pz_regularizer = BalanceRegularizer(1e-7)
model = FlatModel(cooccurrence=cooccurrence,
z_k=z_k,
opt=opt,
mode=1,
pz_regularizer=pz_regularizer,
scale=scale)
model.train(outputpath, epochs=epochs, batches=batches)
return run_flat_validation(input_path=outputpath,
output_path=os.path.join(
outputpath, "validate.txt"),
cooccurrence=cooccurrence)
def main():
epochs = 10
iters = 1
batches = 4096
z_k = 2
z_depth = 10
outputpath = "output/skipgram_tree_b"
betas = [0.5, 0.85, 1.2, 2.0]
weights = [1, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5]
labels = ["b-{:.01e}".format(w) for w in weights]
regularizers = [BalanceRegularizer(w) for w in weights]
train_tree_regularizer_battery(betas=betas,
epochs=epochs,
iters=iters,
batches=batches,
z_k=z_k,
z_depth=z_depth,
outputpath=outputpath,
labels=labels,
is_weight_regularizer=False,
kwdata={'weights': np.array(weights)},
regularizers=regularizers)