def test_export_sklearn_kernel_neg(self):
x = numpy.array([[1, 2], [3, 4]], dtype=float)
kernel = ExpSineSquared(length_scale=1.2, periodicity=1.1)
def kernel_call_ynone(X, length_scale=1.2, periodicity=1.1, pi=3.141592653589793):
dists = squareform_pdist(X, metric='euclidean')
arg = dists / periodicity * pi
sin_of_arg = numpy.sin(arg)
K = numpy.exp((sin_of_arg / length_scale) ** 2 * (-2))
return K
exp = kernel(x, None)
got = kernel_call_ynone(x)
self.assertEqualArray(exp, got)
context = {'numpy.sin': numpy.sin, 'numpy.exp': numpy.exp,
'numpy_pi': numpy.pi, 'squareform_pdist': 'squareform_pdist'}
onnx_code = translate_fct2onnx(kernel_call_ynone, context=context,
output_names=['Z'])
self.assertIn(
"X, length_scale=1.2, periodicity=1.1, pi=3.14159", onnx_code)
self.assertIn("-2", onnx_code)
self.assertIn('metric="euclidean"', onnx_code)
def gp(train, test, t=132):
X_train, X_test = sklearn_formatting(train, test)
gp_kernel = 2**2 \
+ ExpSineSquared(1, 60000.0) \
+ ExpSineSquared(2, 120000.0) \
+ WhiteKernel(2.5)
gpr = GaussianProcessRegressor(kernel=gp_kernel)
gpr.fit(X_train, train.values)
y_fit = gpr.predict(X_train, return_std=False)
# predict a cycle
y_pred = gpr.predict(X_test, return_std=False)
rmse = error(test.values, y_pred)
return y_fit, y_pred, rmse
def test_gpr_rbf_fitted_return_std_exp_sine_squared_double(self):
gp = GaussianProcessRegressor(kernel=ExpSineSquared(),
alpha=1e-7,
n_restarts_optimizer=15,
normalize_y=True)
gp.fit(Xtrain_, Ytrain_)
# return_cov=False, return_std=False
options = {GaussianProcessRegressor: {"return_std": True}}
gp.predict(Xtrain_, return_std=True)
model_onnx = to_onnx(gp,
initial_types=[('X',
DoubleTensorType([None, None]))],
options=options,
dtype=np.float64)
self.assertTrue(model_onnx is not None)
dump_data_and_model(
Xtest_.astype(np.float64),
gp,
model_onnx,
verbose=False,
basename="SklearnGaussianProcessExpSineSquaredStdDouble-Out0-Dec4")
self.check_outputs(
gp,
model_onnx,
Xtest_.astype(np.float64),
predict_attributes=options[GaussianProcessRegressor],
decimal=4)
def test_kernel_ker2_exp_sine_squared(self):
ker = ExpSineSquared()
onx = convert_kernel(ker,
'X',
output_names=['Y'],
dtype=np.float32,
op_version=onnx_opset_version())
model_onnx = onx.to_onnx(inputs=[('X', FloatTensorType([None, None]))],
dtype=np.float32)
sess = InferenceSession(model_onnx.SerializeToString())
res = sess.run(None, {'X': Xtest_.astype(np.float32)})[0]
m1 = res
m2 = ker(Xtest_)
assert_almost_equal(m1, m2, decimal=4)
onx = convert_kernel(ker,
'X',
output_names=['Z'],
x_train=Xtest_ * 2,
dtype=np.float32,
op_version=onnx_opset_version())
model_onnx = onx.to_onnx(inputs=[('X', FloatTensorType([None, None]))],
dtype=np.float32)
sess = InferenceSession(model_onnx.SerializeToString())
res = sess.run(None, {'X': Xtest_.astype(np.float32)})[0]
m1 = res
m2 = ker(Xtest_, Xtest_ * 2)
assert_almost_equal(m1, m2, decimal=4)
def test_export_sklearn_kernel_exp_sine_squared(self):
x = numpy.array([[1, 2], [3, 4]], dtype=float)
kernel = ExpSineSquared(length_scale=1.2, periodicity=1.1)
def kernel_call_ynone(X, length_scale=1.2, periodicity=1.1, pi=3.141592653589793):
dists = squareform_pdist(X, metric='euclidean')
t_pi = py_make_float_array(pi)
t_periodicity = py_make_float_array(periodicity)
arg = dists / t_periodicity * t_pi
sin_of_arg = numpy.sin(arg)
t_2 = py_make_float_array(2)
t__2 = py_make_float_array(-2)
t_length_scale = py_make_float_array(length_scale)
K = numpy.exp((sin_of_arg / t_length_scale) ** t_2 * t__2)
return K
exp = kernel(x, None)
got = kernel_call_ynone(x)
self.assertEqualArray(exp, got)
context = {'numpy.sin': numpy.sin, 'numpy.exp': numpy.exp,
'numpy_pi': numpy.pi, 'squareform_pdist': 'squareform_pdist',
'py_make_float_array': py_make_float_array}
onnx_code = translate_fct2onnx(kernel_call_ynone, context=context,
output_names=['Z'])
self.assertIn(
"X, length_scale=1.2, periodicity=1.1, pi=3.14159", onnx_code)
self.assertIn("-2", onnx_code)
self.assertIn('metric="euclidean"', onnx_code)
from skl2onnx.algebra.onnx_ops import ( # pylint: disable=E0611,E0401
OnnxAdd, OnnxSin, OnnxMul, OnnxPow, OnnxDiv, OnnxExp
)
from skl2onnx.algebra.complex_functions import onnx_squareform_pdist
ctx = {'OnnxAdd': OnnxAdd, 'OnnxPow': OnnxPow,
'OnnxSin': OnnxSin, 'OnnxDiv': OnnxDiv,
'OnnxMul': OnnxMul, 'OnnxIdentity': OnnxIdentity,
'OnnxExp': OnnxExp, 'numpy': numpy,
'onnx_squareform_pdist': onnx_squareform_pdist,
'py_make_float_array': py_make_float_array}
fct = translate_fct2onnx(kernel_call_ynone, context=context,
cpl=True, context_cpl=ctx,
output_names=['Z'], dtype=numpy.float32)
r = fct('X')
self.assertIsInstance(r, OnnxIdentity)
inputs = {'X': x.astype(numpy.float32)}
onnx_g = r.to_onnx(inputs)
oinf = OnnxInference(onnx_g)
res = oinf.run(inputs)
self.assertEqualArray(exp, res['Z'])
def plot_cloud_top_series():
loc = '/scratchSSD/loh/tracking'
case_name = 'BOMEX'
f_list = sorted(glob.glob(f'{loc}/{case_name}/clouds_*.pq'))
with open('unique_clouds.json', 'r') as f:
c_dict = ujson.load(f)
c_list = [f_list[i] for i in c_dict[f'{cid}']]
with Parallel(n_jobs=16) as Pr:
result = Pr(delayed(get_cloud_top_height)(f) for f in c_list)
#---- Plotting
fig = plt.figure(1, figsize=(6, 3))
fig.clf()
sns.set_context('paper')
sns.set_style('ticks',
{
'axes.grid': False,
'axes.linewidth': '0.75',
'grid.color': '0.75',
'grid.linestyle': u':',
'legend.frameon': True,
})
plt.rc('text', usetex=True)
plt.rc('font', family='Helvetica')
ax = plt.subplot(1, 1, 1)
plt.ylabel(r'$d \hat{\mathcal{H}} / dt$')
plt.xlabel('Time [min]')
# xf, yf = fft_.get_fft_matrix(h_)
# plt.plot(xf, yf, '--o')
plt.plot(t_, h_, '--o', lw=0.75)
# Plot GP regression result
kernel = 1.0 * RBF(length_scale=1e5, length_scale_bounds=(1e3, 1e5)) \
+ 1.0 * WhiteKernel(noise_level=1e-2) \
+ 1.0 * ExpSineSquared(periodicity=15.8, periodicity_bounds=(5, 25))
gp = GaussianProcessRegressor(kernel=kernel, n_restarts_optimizer=0)
gp.fit(t_[:, None], h_[:])
X = np.linspace(t_[0], t_[-1], 180)
y_mean, y_std = gp.predict(X[:, None], return_std=True)
plt.plot(X, y_mean, 'k', lw=1, zorder=9)
plt.fill_between(X, y_mean - y_std, y_mean + y_std,
alpha=0.2, color='k')
print(gp.kernel_)
print(gp.log_marginal_likelihood_value_)
y_samples = gp.sample_y(X[:, None], 10)
plt.plot(X, y_samples, lw=0.5, alpha=0.3)
plt.tight_layout(pad=0.5)
figfile = 'png/{}.png'.format(os.path.splitext(__file__)[0])
print('\t Writing figure to {}...'.format(figfile))
plt.savefig(figfile,bbox_inches='tight', dpi=180, \
facecolor='w', transparent=True)
def knr(X,y):
param_grid = {"alpha": [1e0, 1e-1, 1e-2, 1e-3],
"kernel": [ExpSineSquared(l, p)
for l in np.logspace(-2, 2, 10)
for p in np.logspace(0, 2, 10)]}
kr = GridSearchCV(KernelRidge(), cv=5, param_grid=param_grid)
kr.fit(X,y)
return kr
class GaussianProcessRegressor_Fit(SkLearnNode):
icon = "icons/divide.png"
op_code = GAUSSIANPROCESSREGRESSOR_FIT
op_title = "GaussianProcessRegressor_fit"
content_label = "/"
content_label_objname = "GaussianProcessRegressor_node_fit"
kernel = 1.0 * ExpSineSquared(length_scale=1.0,
periodicity=3.0,
length_scale_bounds=(0.2, 10.0),
periodicity_bounds=(3.0, 6.0))
model = GaussianProcessRegressor(kernel=kernel)
def __init__(self, scene):
super().__init__(scene, inputs=[1, 1, 1], outputs=[2])
def initInnerClasses(self):
self.content = GaussianProcessRegressorContent(self)
self.grNode = SkLearnGraphicsNode(self)
def train(self, x_data, y_data):
if n_neighbors != self.n_neighbors:
self.model = GaussianProcessRegressor(kernel=self.kernel)
self.model.fit(x_data, y_data)
return self.model
def evalImplementation(self, param):
x_Input = self.getInput(0)
y_Input = self.getInput(1)
socket1 = self.getInputSocket(0)
socket2 = self.getInputSocket(1)
# socket3 = self.getInputSocket(2)
# val1=input_node1.eval({"a":socket1})
# val2=input_node2.eval({"a":socket2})
# if n_neighbors_Input is not None:
# self.n_neighbors=n_neighbors_Input.eval({})
# self.content.lbl.setText(str(self.n_neighbors))
if x_Input is None or y_Input is None:
self.markInvalid()
self.markDescendantsDirty()
self.grNode.setToolTip("Connect all inputs")
return None
else:
val = self.train(x_Input.eval({"a": socket1}),
y_Input.eval({"a": socket2}))
self.value = val
self.markDirty(False)
self.markInvalid(False)
self.grNode.setToolTip("")
self.markDescendantsDirty()
self.evalChildren()
return val
def test():
pd.set_option('display.max_columns', None)
pd.set_option('display.max_rows', None)
dataset = read_lake_erie_data()
v = np.array(dataset.index).reshape(-1, 1)
bootstrapped_dataset = gp_bootstrap_noise(np.array(dataset.index).reshape(-1, 1),
dataset["Level"].reshape(-1, 1),
3.27 ** 2 * ExpSineSquared(length_scale=180, periodicity=10) *
ExpSineSquared(length_scale=1.44),
n_samples=50,
alpha=.09)
co2_dataset = read_mauna_loa_co2_data()
samps = gp_bootstrap_noise(co2_dataset['Time'].reshape(-1, 1),
co2_dataset['CO2Concentration'].reshape(-1, 1),
34.4 ** 2 * RBF(length_scale=41.8) +
3.27 ** 2 * RBF(length_scale=180) * ExpSineSquared(length_scale=1.44,
periodicity=1) +
0.446 ** 2 * RationalQuadratic(alpha=17.7, length_scale=0.957) +
0.197 ** 2 * RBF(length_scale=0.138) + WhiteKernel(noise_level=0.0336),
n_samples=50,
alpha=30.0)
f, (ax1, ax2) = plt.subplots(1, 2)
for dat in bootstrapped_dataset:
ax1.plot(dat[1], alpha=.1, color="gray")
ax1.plot(dataset["Level"], color="blue")
ax1.set_xlabel("Month")
ax1.set_ylabel("Level (M)")
for dat in samps:
ax2.plot(dat[1], alpha=.1, color="gray")
ax2.plot(co2_dataset['CO2Concentration'], color="blue")
ax2.set_xlabel("Month")
ax2.set_ylabel("CO2 Concentration (PPM)")
plt.suptitle("Example of Samples from GP Bootstrap")
plt.show()
def __kernel_ridge(fn,X,y,x):
param_grid = {"alpha": [1e0, 1e-1, 1e-2, 1e-3],
"kernel": [ExpSineSquared(l, p)
for l in np.logspace(-2, 2, 10)
for p in np.logspace(0, 2, 10)]}
kr = GridSearchCV(KernelRidge(), cv=5, param_grid=param_grid)
kr.fit(X, y)
y_kr = kr.predict(x)
return y_kr
def test_bound_check_fixed_hyperparameter():
# Regression test for issue #17943
# Check that having a hyperparameter with fixed bounds doesn't cause an
# error
k1 = 50.0**2 * RBF(length_scale=50.0) # long term smooth rising trend
k2 = ExpSineSquared(length_scale=1.0, periodicity=1.0,
periodicity_bounds="fixed") # seasonal component
kernel = k1 + k2
GaussianProcessRegressor(kernel=kernel).fit(X, y)
def get_gaussian_process():
kernel = ExpSineSquared(length_scale_bounds=(0.5, 100),
periodicity=52,
periodicity_bounds=(10, 100))
gaussian_process = GaussianProcessRegressor(kernel=kernel,
alpha=5,
normalize_y=False,
n_restarts_optimizer=10)
return gaussian_process
def generate_erie_samples(n_samples):
erie_dataset = read_lake_erie_data()
train_data = erie_dataset.iloc[:int(.7 * len(erie_dataset)), ]
samps = gp_bootstrap_noise(np.array(train_data.index).reshape(-1, 1),
train_data['Level'].reshape(-1, 1),
3.27 ** 2 * ExpSineSquared(length_scale=180, periodicity=10) *
ExpSineSquared(length_scale=1.44),
n_samples=n_samples,
alpha=.09)
for osf in glob.glob("./gp_samples/erie/*.pkl"):
os.remove(osf)
for idx, samp in enumerate(samps):
with open("./gp_samples/erie/" + str(idx) + ".pkl", "wb") as sf:
pickle.dump(samp, sf)
def __init__(self,
t,
y,
selected_kernel="RatQuad",
interpolation_factor=None):
super().__init__(t, y)
self.kernels = None
self.selected_kernel = selected_kernel
self.interpolation_factor = interpolation_factor
# TODO: fix this to comply with python standards
self.A_mean = None
self.A_std = None
# Create different kernels that will be explored
self.kernels = dict()
self.kernels["RBF"] = 1.0 * RBF(length_scale=0.5)
self.kernels["RatQuad"] = 1.0 * RationalQuadratic(length_scale=1.0,
alpha=0.2)
self.kernels["ExpSineSquared"] = 1.0 * ExpSineSquared(length_scale=1.0,
periodicity=3)
self.kernels["Matern"] = 1.0 * Matern(length_scale=1.0, nu=1.5)
self.kernels["Matern*ExpSineSquared"] = (
1.0 * Matern(length_scale=1.0, nu=1.5) *
ExpSineSquared(length_scale=1, periodicity=3))
self.kernels["RBF*ExpSineSquared"] = (
1.0 * RBF(length_scale=1.0) *
ExpSineSquared(length_scale=1, periodicity=3))
self.kernels["RatQuad*ExpSineSquared"] = (
1.0 * RationalQuadratic(length_scale=1.0, alpha=0.2) *
ExpSineSquared(length_scale=1, periodicity=3))
self.kernels["Matern*RBF"] = (1.0 * Matern(length_scale=1.0, nu=1.5) *
RBF(length_scale=1))
self.kernels["Matern+ExpSineSquared"] = 1.0 * Matern(
length_scale=1.0, nu=1.5) + ExpSineSquared(length_scale=1,
periodicity=3)
self.kernels["RBF+ExpSineSquared"] = 1.0 * RBF(
length_scale=1.0) + ExpSineSquared(length_scale=1, periodicity=3)
self.kernels["RatQuad+ExpSineSquared"] = 1.0 * RationalQuadratic(
length_scale=1.0) + ExpSineSquared(length_scale=1, periodicity=3)
if selected_kernel not in self.kernels.keys():
raise KeyError(
f"Unknown kernel: {selected_kernel}, available kernels: {self.kernels.keys()}"
)
# Generate the noisy kernels
self.noisy_kernels = dict()
for key, kernel in self.kernels.items():
self.noisy_kernels[key] = kernel + WhiteKernel(
noise_level=1, noise_level_bounds=(1e-7, 1e7))
def build_gp(arguments):
"""
Build Gaussian Process using scikit-learn, print hyperparams and return model
:return: gp model
"""
kernel_dict = {
'c_rbf':
C(arguments.const_val,
(arguments.const_val_min, arguments.const_val_max)) *
RBF(length_scale=arguments.length_scale,
length_scale_bounds=(arguments.length_scale_min,
arguments.length_scale_max)),
'rbf':
RBF(length_scale=arguments.length_scale,
length_scale_bounds=(arguments.length_scale_min,
arguments.length_scale_max)),
'matern':
Matern(length_scale=arguments.length_scale,
length_scale_bounds=(arguments.length_scale_min,
arguments.length_scale_max),
nu=arguments.nu),
'expsinesquared':
ExpSineSquared(length_scale=arguments.length_scale,
periodicity=arguments.periodicity,
length_scale_bounds=(arguments.length_scale_min,
arguments.length_scale_max),
periodicity_bounds=(arguments.periodicity_min,
arguments.periodicity_max))
}
kernel = kernel_dict[arguments.kernel]
if arguments.verbosity:
print(
"\n================================================================"
)
print("\nKernel: {}".format(kernel))
print("\nHyperparameters:")
for hyperparameter in kernel.hyperparameters:
print(hyperparameter)
print("\nParameters:")
params = kernel.get_params()
for key in sorted(params):
print("%s : %s" % (key, params[key]))
print(
"\n================================================================"
)
gp = GaussianProcessRegressor(kernel=kernel,
n_restarts_optimizer=20,
normalize_y=True)
return gp
def define_model():
k0 = WhiteKernel(noise_level=0.3**2, noise_level_bounds=(0.1**2, 0.5**2))
k1 = ConstantKernel(constant_value=2)* \
ExpSineSquared(length_scale=1.0, periodicity=40, periodicity_bounds=(35,45))
k2 = ConstantKernel(constant_value=10, constant_value_bounds=(1e-2, 1e3))* \
RBF(length_scale=100.0, length_scale_bounds=(1, 1e4))
kernel_1 = k0 + k1 + k2
linear_model = gaussian_process.GaussianProcessRegressor(
kernel=kernel_1, n_restarts_optimizer=10, normalize_y=True, alpha=0.0)
return linear_model
def gaussian_regressor_param_selection(X, y, nfolds):
kernel_rbf = ConstantKernel(1.0, constant_value_bounds="fixed") * RBF(1.0, length_scale_bounds="fixed")
kernel_rq = ConstantKernel(1.0, constant_value_bounds="fixed") * RationalQuadratic(alpha=0.1, length_scale=1)
kernel_expsine = ConstantKernel(1.0, constant_value_bounds="fixed") * ExpSineSquared(1.0, 5.0, periodicity_bounds=(1e-2, 1e1))
Kernels = [kernel_rbf, kernel_rq]
param_grid = {'kernel': Kernels}
grid_search = GridSearchCV(GaussianProcessRegressor(random_state=0), param_grid, cv=nfolds, n_jobs=-1)
grid_search.fit(X, y)
print('GaussianRegressor Lowest MSE Score:'+str(grid_search.best_score_))
print('GaussianRegressor With Parameters:'+str(grid_search.best_params_))
return grid_search.best_params_
def test_duck_typing_nested_estimator():
# Test duck typing metaestimators with random search
kernel_ridge = KernelRidge(kernel=ExpSineSquared())
param_distributions = {"alpha": [1, 2]}
kernel_ridge_tuned = RandomizedSearchCV(
kernel_ridge,
param_distributions=param_distributions,
)
html_output = estimator_html_repr(kernel_ridge_tuned)
assert "estimator: KernelRidge</label>" in html_output
def test_kernel_exp_sine_squared(self):
from skl2onnx.operator_converters.gaussian_process import convert_kernel
ker = ExpSineSquared()
onx = convert_kernel(ker, 'X', output_names=['Y'], dtype=numpy.float32)
model_onnx = onx.to_onnx(inputs=[('X', FloatTensorType([None, None]))])
sess = OnnxInference(model_onnx)
Xtest_ = numpy.arange(6).reshape((3, 2))
res = sess.run({'X': Xtest_.astype(numpy.float32)})
m1 = res['Y']
m2 = ker(Xtest_)
self.assertEqualArray(m1, m2, decimal=5)
def set_estimators(self):
#gauss radial kernel
params = {
'kernel': [
C(1.0, (1e-3, 1e3)),
C(1.0, (1e-3, 1e3)) * RBF(10, (1e-2, 1e2)),
ExpSineSquared()
],
'n_restarts_optimizer': [9]
}
self.estimators['gaussian_process'] = (GaussianProcessRegressor,
params)
def _test_scikit_gaussian_process_one(self):
problem = MyProblemSin()
problem.surrogate = SurrogateModelScikit(problem)
problem.surrogate.sigma_threshold = 0.1
kernel = 1.0 * ExpSineSquared(length_scale=1.0,
periodicity=3.0,
length_scale_bounds=(0.1, 10.0),
periodicity_bounds=(1.0, 10.0))
problem.surrogate.regressor = GaussianProcessRegressor(kernel=kernel)
self._check_scikit_one(problem, 5.0)
def special_conversions(self, params):
"""
TODO: replace this logic with something better
"""
# create list parameters
lists = defaultdict(list)
element_regex = re.compile(r'(.*)\[(\d)\]')
for name, param in list(params.items()):
# look for variables of the form "param_name[1]"
match = element_regex.match(name)
if match:
# name of the list parameter
lname = match.groups()[0]
# index of the list item
index = int(match.groups()[1])
lists[lname].append((index, param))
# drop the element parameter from our list
del params[name]
for lname, items in list(lists.items()):
# drop the list size parameter
del params['len(%s)' % lname]
# sort the list by index
params[lname] = [val for idx, val in sorted(items)]
# Gaussian process classifier
if self.method == "gp":
if params["kernel"] == "constant":
params["kernel"] = ConstantKernel()
elif params["kernel"] == "rbf":
params["kernel"] = RBF()
elif params["kernel"] == "matern":
params["kernel"] = Matern(nu=params["nu"])
del params["nu"]
elif params["kernel"] == "rational_quadratic":
params["kernel"] = RationalQuadratic(
length_scale=params["length_scale"], alpha=params["alpha"])
del params["length_scale"]
del params["alpha"]
elif params["kernel"] == "exp_sine_squared":
params["kernel"] = ExpSineSquared(
length_scale=params["length_scale"],
periodicity=params["periodicity"])
del params["length_scale"]
del params["periodicity"]
# return the updated parameter vector
return params
def str2ker(str):
k1 = C(1.0) * RBF(length_scale=1)
k2 = C(1.0) * RationalQuadratic(length_scale=1)
k4 = DotProduct(sigma_0=1)
k3 = C(1.0) * ExpSineSquared(length_scale=1, periodicity=1)
k5 = WhiteKernel(1.0)
map = {"s": k1, "r": k2, "p": k3, "l": k4}
# if basic kernel
if len(str) == 1:
ker = map[str]
else:
# if composite kernel
ker = []
factor = map[str[0]]
op = str[1]
for i in range(2, len(str), 2):
# if the operator is *, use @ covProd to continue costructing the
# factor
if op == '*':
factor = factor * map[str[i]]
# the end?
if i == len(str) - 1:
if not ker:
ker = factor
else:
ker = ker + factor
else:
op = str[i + 1]
# if the oprator is +, combine current factor with ker then form a
# new factor
else:
if not ker:
ker = factor
else:
ker = ker + factor
factor = map[str[i]]
# % the end?
if i == len(str) - 1:
ker = ker + factor
else:
op = str[i + 1]
ker = ker + k5
return ker
def fig_sin_kernel(only_trace: bool = True):
from sklearn.gaussian_process.kernels import ExpSineSquared
kernel = ExpSineSquared()
x = np.linspace(-1, 1, 100)
x_exp = np.expand_dims(x, axis=1)
surf_data = kernel(x_exp, x_exp)
trace = go.Surface(x=x, y=x, z=surf_data, showscale=False)
if only_trace:
return trace
else:
fig.update_layout(scene=dict(xaxis_title='xi',
yaxis_title='xj',
zaxis_title='Linear Kernel Value'),
margin=dict(r=10, b=10, l=10, t=10))
return fig
def run(self):
import numpy as np
from sklearn.gaussian_process import GaussianProcessClassifier
from sklearn.gaussian_process.kernels \
import RBF, WhiteKernel, RationalQuadratic, ExpSineSquared
X = np.array(self.train.data)
Y = np.array(self.train.occupancy).flatten()
kernel = RBF() + RBF() * ExpSineSquared() + RationalQuadratic(
) + WhiteKernel()
gp = GaussianProcessClassifier(kernel=kernel,
optimizer='fmin_l_bfgs_b').fit(X, Y)
predict_occupancy = gp.predict(np.array(self.test.data))
return np.reshape(predict_occupancy, (-1, 1))
def test_onnxt_gpr_iris(self):
iris = load_iris()
X, y = iris.data, iris.target
X_train, _, y_train, __ = train_test_split(X, y, random_state=11)
clr = GaussianProcessRegressor(ExpSineSquared(), alpha=20.)
clr.fit(X_train, y_train)
model_def = to_onnx(clr, X_train)
oinf = OnnxInference(model_def)
res1 = oinf.run({'X': X_train})
new_model = onnx_optimisations(model_def)
oinf = OnnxInference(new_model)
res2 = oinf.run({'X': X_train})
self.assertEqualArray(res1['GPmean'], res2['GPmean'])
self.assertNotIn('op_type: "CDist"', str(new_model))
def __init__(self,
f,
domain=None,
kernel=None,
alpha=1e-10,
n_restarts_optimizer=2,
random_inits=2,
**kwargs):
if kernel is None:
kernel = kernel = C(1.0, (1e-3, 1e3)) * RBF(10, (1e-2, 1e2)) + C(
1e-2, (1e-3, 1e3)) * ExpSineSquared(1.0, 1.0, (1e-5, 1e5),
(1e-5, 1e5))
# set default kernel
# cov_amplitude = C(1.0, (1e-3, 1e3))
# matern = Matern(length_scale=)
if domain is None:
domain = np.atleast_2d(np.linspace(0.01, 10, 1000)).T
self.domain = domain
self.f = f
self.n_restarts_optimizer = n_restarts_optimizer
nrs = n_restarts_optimizer
self.gp = GaussianProcessRegressor(kernel=kernel,
alpha=alpha,
n_restarts_optimizer=nrs,
**kwargs)
self.X_ind = np.random.randint(0, len(self.domain), random_inits)
self.X = np.atleast_2d(self.domain[self.X_ind])
# self.y = f(self.X).ravel()
self.y = np.asarray([f(x) for x in self.X])
self.fig_stocker = []
self.fig_info_stocker = []
self.current_argmax = self.domain[np.argmax(self.y)]
self.current_max = max(self.y)
self.current_step = random_inits
self.other_info_to_stock = {
'theta_after_fit_log': [],
'kernel_hyperparams_after_fit': []
}
self.gp.fit(self.X, self.y)
self.other_info_to_stock['theta_after_fit_log'].append(
self.gp.kernel_.theta)
self.other_info_to_stock['kernel_hyperparams_after_fit'].append(
self.gp.kernel_.get_params())
self.y_pred, self.sigma = self.gp.predict(self.domain, return_std=True)
self.current_argmax = self.domain[self.X_ind[np.argmax(
self.y_pred[self.X_ind])]]
self.current_max = max(self.y_pred[self.X_ind])
def fit(self, long_term_length_scale=None,
pre_periodic_term_length_scale=None,
periodic_term_length_scale=None, periodicity=None,
noise_level=None, do_plot=False, fig=None):
data = self.data[['mjd', 'mag', 'err']]
data = np.atleast_2d(data)
time = data[:, 0] - data[0, 0]
time = np.atleast_2d(time).T
if self._gp is None:
time_scale = data[-1, 0] - data[0, 0]
data_scale = np.max(data[:, 1]) - np.min(data[:, 1])
noise_std = np.median(data[:, 2])
if long_term_length_scale is None:
long_term_length_scale = 0.5 * time_scale
if pre_periodic_term_length_scale is None:
pre_periodic_term_length_scale = 0.5 * time_scale
if periodic_term_length_scale is None:
periodic_term_length_scale = 0.1 * time_scale
if periodicity is None:
periodicity = 0.1 * time_scale
if noise_level is None:
noise_level = noise_std
k1 = data_scale ** 2 * RBF(length_scale=long_term_length_scale)
k2 = 0.1 * data_scale *\
RBF(length_scale=pre_periodic_term_length_scale) *\
ExpSineSquared(length_scale=periodic_term_length_scale,
periodicity=periodicity)
k3 = WhiteKernel(noise_level=noise_level ** 2,
noise_level_bounds=(1e-3, 1.))
kernel = k1 + k2 + k3
gp = GaussianProcessRegressor(kernel=kernel,
alpha=(data[:, 2] / data[:, 1]) ** 2,
normalize_y=True,
n_restarts_optimizer=10)
gp.fit(time, data[:, 1])
self._gp = gp
if do_plot:
self.plot_fitted(fig=fig)
def _cast(self, n, X, y):
"""
Evaluates and optimizes all legitimate combinations of length `n`
:param n: The length of pipelines
:param X: Training data
:param y: Observed values
:return: None
"""
from .structsearch import SurrogateRandomCV, BoxSample, CompactSample
from importlib import import_module
if self.couldBfirst == []:
from sklearn.pipeline import Pipeline
else:
from imblearn.pipeline import Pipeline
from sklearn.model_selection import RandomizedSearchCV
if self.surrogates is None:
from numpy import logspace
from sklearn.gaussian_process import GaussianProcessRegressor
from sklearn.kernel_ridge import KernelRidge
from sklearn.gaussian_process.kernels import Matern, Sum, ExpSineSquared, WhiteKernel
param_grid_gpr = {"alpha": logspace(-8, 1, 20),
"kernel": [Sum(Matern(length_scale=l, nu=p), WhiteKernel(noise_level=q))
for l in logspace(-3, 3, 20)
for p in [0.5, 1.5, 2.5]
for q in logspace(-3, 1.5, 20)]}
GPR = RandomizedSearchCV(GaussianProcessRegressor(), param_distributions=param_grid_gpr, n_iter=20, cv=2)
param_grid_krr = {"alpha": logspace(-4, 0, 10),
"kernel": [Sum(Matern(), ExpSineSquared(l, p))
for l in logspace(-2, 2, 20)
for p in logspace(0, 2, 20)]}
KRR = RandomizedSearchCV(KernelRidge(), param_distributions=param_grid_krr, n_iter=30, cv=2)
self.surrogates = [(KRR, 35, CompactSample, 'L-BFGS-B'), (GPR, 50, BoxSample, 'L-BFGS-B')]
self.min_random_evals = 10
Pop = []
candidates = self.words.Generate(n)
for cnddt in candidates:
if self._validate_sequence(cnddt):
Pop.append(cnddt)
for seq in Pop:
if not self._validate_sequence(seq):
continue
best_mdl, best_scr = self.optimize_pipeline(seq, X, y)
self.models[seq] = (best_mdl, best_scr)
if self.verbose > 0:
print("score:%f" % best_scr)
print(best_mdl)
def GPR(X, Y, Z):
k1 = 50.0**2 * RBF(length_scale=50.0) # long term smooth rising trend
k2 = 2.0**2 * RBF(length_scale=100.0) \
* ExpSineSquared(length_scale=1.0, periodicity=1.0,
periodicity_bounds="fixed") # seasonal component
# medium term irregularities
k3 = 0.5**2 * RationalQuadratic(length_scale=1.0, alpha=1.0)
k4 = 0.1**2 * RBF(length_scale=0.1) \
+ WhiteKernel(noise_level=0.1**2,
noise_level_bounds=(1e-3, np.inf)) # noise terms
kernel = k4
gp = GaussianProcessRegressor(kernel=kernel, alpha=0, normalize_y=True)
gp.fit(X, Y)
mu, var = gp.predict(Z, return_std=True)
return mu, var