https://github.com/GPflow/GPflow
Tip revision: 4f1faf11ce53037efcdd3ed60be52646c3206222 authored by Sergio Diaz on 10 September 2019, 15:11:24 UTC
SGPR and FITC updated
SGPR and FITC updated
Tip revision: 4f1faf1
test_predict.py
# Copyright 2016 the GPflow authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
import pytest
import gpflow
from gpflow.inducing_variables import InducingPoints
from gpflow.kernels import Matern32
rng = np.random.RandomState(0)
class ModelSetup:
def __init__(self,
model_class,
kernel=Matern32(),
likelihood=gpflow.likelihoods.Gaussian(),
whiten=None,
q_diag=None,
requires_Z_as_input=True,
requires_data_as_input=False
):
self.model_class = model_class
self.kernel = kernel
self.likelihood = likelihood
self.whiten = whiten
self.q_diag = q_diag
self.requires_Z_as_input = requires_Z_as_input
self.requires_data_as_input = requires_data_as_input
def get_model(self, Z, num_latent, data=None):
if self.whiten is not None and self.q_diag is not None:
return self.model_class(inducing_variables=Z,
kernel=self.kernel,
likelihood=self.likelihood,
num_latent=num_latent,
whiten=self.whiten,
q_diag=self.q_diag)
elif self.requires_data_as_input:
return self.model_class(data,
inducing_variables=Z,
kernel=self.kernel,
num_latent=num_latent)
else:
return self.model_class(inducing_variables=Z,
kernel=self.kernel,
likelihood=self.likelihood,
num_latent=num_latent)
def __repr__(self):
return f"ModelSetup({self.model_class.__name__}, {self.whiten}, {self.q_diag})"
model_setups = [
ModelSetup(model_class=gpflow.models.SVGP, whiten=False, q_diag=True),
ModelSetup(model_class=gpflow.models.SVGP, whiten=True, q_diag=False),
ModelSetup(model_class=gpflow.models.SVGP, whiten=True, q_diag=True),
ModelSetup(model_class=gpflow.models.SVGP, whiten=False, q_diag=False),
ModelSetup(model_class=gpflow.models.SGPR, requires_data_as_input=True),
# ModelSetup(model_class=gpflow.models.GPRF),
# ModelSetup(model_class=gpflow.models.VGP, requires_Z_as_input = False),
# ModelSetup(model_class=gpflow.models.GPMC, requires_Z_as_input = False ),
# ModelSetup(model_class=gpflow.models.SGPMC)
]
@pytest.mark.parametrize('Ntrain, Ntest, D', [[100, 10, 2]])
def test_gaussian_mean_and_variance(Ntrain, Ntest, D):
data = rng.randn(Ntrain, D), rng.randn(Ntrain, 1)
Xtest, _ = rng.randn(Ntest, D), rng.randn(Ntest, 1)
kernel = Matern32() + gpflow.kernels.White()
model_gp = gpflow.models.GPR(data, kernel=kernel)
mu_f, var_f = model_gp.predict_f(Xtest)
mu_y, var_y = model_gp.predict_y(Xtest)
assert np.allclose(mu_f, mu_y)
assert np.allclose(var_f, var_y - 1.)
@pytest.mark.parametrize('Ntrain, Ntest, D', [[100, 10, 2]])
def test_gaussian_log_density(Ntrain, Ntest, D):
data = rng.randn(Ntrain, D), rng.randn(Ntrain, 1)
Xtest, Ytest = rng.randn(Ntest, D), rng.randn(Ntest, 1)
kernel = Matern32() + gpflow.kernels.White()
model_gp = gpflow.models.GPR(data, kernel=kernel)
mu_y, var_y = model_gp.predict_y(Xtest)
data = Xtest, Ytest
log_density = model_gp.predict_log_density(data)
log_density_hand = (-0.5 * np.log(2 * np.pi) - 0.5 * np.log(var_y) - 0.5 * np.square(mu_y - Ytest) / var_y)
assert np.allclose(log_density_hand, log_density)
@pytest.mark.parametrize('input_dim, output_dim, N, Ntest, M', [[3, 2, 20, 30, 5]])
def test_gaussian_full_cov(input_dim, output_dim, N, Ntest, M):
covar_shape = (output_dim, Ntest, Ntest)
X, Y, Z = rng.randn(N, input_dim), rng.randn(N, output_dim), rng.randn(M, input_dim)
Xtest = rng.randn(Ntest, input_dim)
kernel = Matern32()
model_gp = gpflow.models.GPR([X, Y], kernel=kernel)
mu1, var = model_gp.predict_f(Xtest, full_cov=False)
mu2, covar = model_gp.predict_f(Xtest, full_cov=True)
assert np.allclose(mu1, mu2, atol=1.e-10)
assert covar.shape == covar_shape
assert var.shape == (Ntest, output_dim)
for i in range(output_dim):
assert np.allclose(var[:, i], np.diag(covar[i, :, :]))
@pytest.mark.skip(reason='GPR model is not ready')
@pytest.mark.parametrize('input_dim, output_dim, N, Ntest, M, num_samples', [[3, 2, 20, 30, 5, 5]])
def test_gaussian_full_cov_samples(input_dim, output_dim, N, Ntest, M, num_samples):
samples_shape = (num_samples, Ntest, output_dim)
X, Y, _ = rng.randn(N, input_dim), rng.randn(N, output_dim), rng.randn(M, input_dim)
Xtest = rng.randn(Ntest, input_dim)
kernel = Matern32()
model_gp = gpflow.models.GPR([X, Y], kernel=kernel)
samples = model_gp.predict_f_samples(Xtest, num_samples)
assert samples.shape == samples_shape
@pytest.mark.parametrize('model_setup', model_setups)
@pytest.mark.parametrize('input_dim', [3])
@pytest.mark.parametrize('output_dim', [2])
@pytest.mark.parametrize('N', [20])
@pytest.mark.parametrize('Ntest', [30])
@pytest.mark.parametrize('M', [5])
def test_other_models_full_cov(model_setup, input_dim, output_dim, N, Ntest, M):
covar_shape = (output_dim, Ntest, Ntest)
X, Y = rng.randn(N, input_dim), rng.randn(N, output_dim)
Z = InducingPoints(rng.randn(M, input_dim))
Xtest = rng.randn(Ntest, input_dim)
model_gp = model_setup.get_model(Z, num_latent=output_dim, data=(X, Y))
mu1, var = model_gp.predict_f(Xtest, full_cov=False)
mu2, covar = model_gp.predict_f(Xtest, full_cov=True)
assert np.allclose(mu1, mu2, atol=1.e-10)
assert covar.shape == covar_shape
assert var.shape == (Ntest, output_dim)
for i in range(output_dim):
assert np.allclose(var[:, i], np.diag(covar[i, :, :]))
@pytest.mark.parametrize('model_setup', model_setups)
@pytest.mark.parametrize('input_dim', [3])
@pytest.mark.parametrize('output_dim', [2])
@pytest.mark.parametrize('N', [20])
@pytest.mark.parametrize('Ntest', [30])
@pytest.mark.parametrize('M', [5])
@pytest.mark.parametrize('num_samples', [5])
def test_other_models_full_cov_samples(model_setup, input_dim, output_dim, N, Ntest, M, num_samples):
samples_shape = (num_samples, Ntest, output_dim)
X, Y, Z = rng.randn(N, input_dim), rng.randn(N, output_dim), rng.randn(M, input_dim)
Xtest = rng.randn(Ntest, input_dim)
model_gp = model_setup.get_model(Z, num_latent=output_dim, data=(X, Y))
samples = model_gp.predict_f_samples(Xtest, num_samples)
assert samples.shape == samples_shape