D_multiple), 0), n_chains_multiple),
group_event_ndims=1)
beta = zs.Normal('beta',
tf.zeros([K, V]),
logstd=tf.ones([K, V]) * log_delta,
group_event_ndims=1)
return model
if __name__ == "__main__":
tf.set_random_seed(1237)
# Load nips dataset
data_name = 'nips'
data_path = os.path.join(conf.data_dir, data_name + '.pkl.gz')
X, vocab = dataset.load_uci_bow(data_name, data_path)
X_train = X[:1200, :]
X_test = X[1200:, :]
# Define model training/evaluation parameters
D = 100
K = 100
V = X_train.shape[1]
n_chains = 1
num_e_steps = 5
hmc = zs.HMC(step_size=1e-3,
n_leapfrogs=20,
adapt_step_size=True,
target_acceptance_rate=0.6)
epochs = 100
X = X.tocoo()
indices = np.transpose(np.array([X.row, X.col])).astype(np.int64)
values = X.data
p = np.lexsort((indices[:,1], indices[:,0]))
#print(p)
return indices[p,:], values[p]
if __name__ == "__main__":
tf.set_random_seed(1237)
# Load nips dataset
data_name = 'enron'
data_path = os.path.join(conf.data_dir, data_name + '.pkl.gz')
# TODO write a sparse datareader: FINISHED
X, vocab = dataset.load_uci_bow(data_name, data_path, is_sparse=True)
X_train = X[:30000, :]
X_test = X[30000:, :]
X_test_indices, X_test_values = get_indices_and_values(X_test)
# Define model training/evaluation parameters
D = 2500
K = 100
V = X_train.shape[1]
n_chains = 1
num_e_steps = 5
hmc = zs.HMC(step_size=1e-3, n_leapfrogs=20, adapt_step_size=True,
target_acceptance_rate=0.6)
epochs = 100
learning_rate_0 = 1.0