def _train(cls, model_param, xys_2d):
"""
:param model_param:
:param xys_2d:
:return:
"""
kernel = model_param['kernel'] if 'kernel' in model_param else 'rbf'
gamma = model_param['gamma'] if 'gamma' in model_param else 0.0
C = model_param['C'] if 'C' in model_param else 1.0
nu = model_param['nu'] if 'nu' in model_param else 0.5
cache_size = model_param[
'cache_size'] if 'cache_size' in model_param else 200
try:
svmutil
except NameError:
from vmaf import svmutil
if kernel == 'rbf':
ktype_int = svmutil.RBF
elif kernel == 'linear':
ktype_int = svmutil.LINEAR
elif kernel == 'poly':
ktype_int = svmutil.POLY
elif kernel == 'sigmoid':
ktype_int = svmutil.SIGMOID
else:
assert False, 'ktype = ' + str(kernel) + ' not implemented'
param = svmutil.svm_parameter([
'-s', 4, '-t', ktype_int, '-c', C, '-g', gamma, '-n', nu, '-m',
cache_size
])
f = list(xys_2d[:, 1:])
for i, item in enumerate(f):
f[i] = list(item)
prob = svmutil.svm_problem(xys_2d[:, 0], f)
model = svmutil.svm_train(prob, param)
return model
def _train(cls, model_param, xys_2d):
"""
:param model_param:
:param xys_2d:
:return:
"""
kernel = model_param['kernel'] if 'kernel' in model_param else 'rbf'
gamma = model_param['gamma'] if 'gamma' in model_param else 0.0
C = model_param['C'] if 'C' in model_param else 1.0
nu = model_param['nu'] if 'nu' in model_param else 0.5
cache_size = model_param['cache_size'] if 'cache_size' in model_param else 200
if kernel == 'rbf':
ktype_int = svmutil.RBF
elif kernel == 'linear':
ktype_int = svmutil.LINEAR
elif kernel == 'poly':
ktype_int = svmutil.POLY
elif kernel == 'sigmoid':
ktype_int = svmutil.SIGMOID
else:
assert False, 'ktype = ' + str(kernel) + ' not implemented'
param = svmutil.svm_parameter([
'-s', 4,
'-t', ktype_int,
'-c', C,
'-g', gamma,
'-n', nu,
'-m', cache_size
])
f = list(xys_2d[:, 1:])
for i, item in enumerate(f):
f[i] = list(item)
prob = svmutil.svm_problem(xys_2d[:, 0], f)
model = svmutil.svm_train(prob, param)
return model
