def synthesize_tensor(dims, datatype, using_ratio, noise):
real_dim = 100
means = [
np.ones((real_dim, )) * 5,
np.ones((real_dim, )) * 10,
np.ones((real_dim, )) * 2
]
covariances = [
np.eye(real_dim) * 2,
np.eye(real_dim) * 3,
np.eye(real_dim) * 2
]
if datatype == "binary":
tensor = binary_tensor()
elif datatype == "real":
tensor = RV_tensor()
elif datatype == "count":
tensor = count_tensor()
"""
"""
tensor.synthesize_data(dims, means, covariances, real_dim, \
train=0.8, sparsity=1, noise=noise, noise_ratio=using_ratio)
# tensor.reduce_train_size(0.1)
return tensor
def synthesize_matrix(dims, datatype, noise_ratio, noise_amount):
real_dim = 100
means = [np.ones((real_dim,)) * 5, np.ones((real_dim,)) * 2]
covariances = [np.eye(real_dim) * 2, np.eye(real_dim) * 3]
if datatype == "binary":
tensor = binary_tensor()
elif datatype == "real":
tensor = RV_tensor()
else:
tensor = count_tensor()
tensor.synthesize_data(dims, means, covariances, real_dim, \
train=0.8, sparsity=1, noise=noise_amount, noise_ratio=noise_ratio)
return tensor
def test_learning_curve(datatype, model):
assert (datatype in ["real", "binary", "count"])
assert (model in ["deterministic", "simple", "robust"])
mean_update = "S"
cov_update = "N"
tensor = None
D = 20
if datatype == "real":
tensor = RV_tensor()
mean0 = np.ones((D, )) * 5
cov0 = np.eye(D)
elif datatype == "binary":
tensor = binary_tensor()
mean0 = np.zeros((D, ))
cov0 = np.eye(D)
elif datatype == "count":
tensor = count_tensor()
factorizer = None
fact_D = 20
if model == "deterministic":
factorizer = SSVI_TF_d(real_tensor, rank=fact_D, \
mean_update=mean_update, cov_update=cov_update, \
k1=64, k2=64, \
mean0=mean0, cov0=cov0)
elif model == "simple":
factorizer = SSVI_TF_simple(data, rank=D, \
mean_update=mean_update, cov_update=cov_update, \
k1=64, k2=64, \
mean0=mean0, cov0=cov0)
elif model == "robust":
factorizer = SSVI_TF_robust(data, rank=D, \
mean_update=mean_update, cov_update=cov_update, \
mean0=mean0, cov0=cov0, k1=128, k2=64, \
eta=1, cov_eta=0.1)
do_learning_curve(factorizer, tensor)
default=500)
parser.add_argument("-m",
"--model",
type=str,
help="model being used",
choices=["deterministic", "simple", "robust"])
args = parser.parse_args()
filename = args.filename
datatype = args.datatype
model = args.model
tensor = None
if datatype == "count":
tensor = count_tensor()
tensor.read_from_file(filename, 0.7, 0.1, 0.2)
print("min_count:", tensor.min_count)
print("max_count:", tensor.max_count)
D = 50
default_params = {
"mean_update": "S",
"cov_update": "N",
"rank": D,
"k1": 64,
"k2": 64
}