test_likelihoods.py
import unittest
import six
import tensorflow as tf
import numpy as np
import GPflow
from GPflow import settings
float_type = settings.dtypes.float_type
np_float_type = np.float32 if float_type is tf.float32 else np.float64
class TestSetup(object):
def __init__(self, likelihood, Y, tolerance):
self.likelihood, self.Y, self.tolerance = likelihood, Y, tolerance
self.is_analytic = six.get_unbound_function(likelihood.predict_density) is not\
six.get_unbound_function(GPflow.likelihoods.Likelihood.predict_density)
def getTestSetups(includeMultiClass=True, addNonStandardLinks=False):
test_setups = []
rng = np.random.RandomState(1)
for likelihoodClass in GPflow.likelihoods.Likelihood.__subclasses__():
if likelihoodClass == GPflow.likelihoods.Ordinal:
test_setups.append(TestSetup(likelihoodClass(np.array([-1, 1])), rng.randint(0, 3, (10, 2)), 1e-6))
elif likelihoodClass == GPflow.likelihoods.SwitchedLikelihood:
continue # switched likelihood tested separately
elif (likelihoodClass == GPflow.likelihoods.MultiClass):
if includeMultiClass:
sample = rng.randn(10, 2)
# Multiclass needs a less tight tolerance due to presence of clipping.
tolerance = 1e-3
test_setups.append(TestSetup(likelihoodClass(2), np.argmax(sample, 1).reshape(-1, 1), tolerance))
else:
# most likelihoods follow this standard:
test_setups.append(TestSetup(likelihoodClass(), rng.rand(10, 2).astype(np_float_type), 1e-6))
if addNonStandardLinks:
test_setups.append(TestSetup(GPflow.likelihoods.Poisson(invlink=tf.square),
rng.rand(10, 2).astype(np_float_type), 1e-6))
test_setups.append(TestSetup(GPflow.likelihoods.Exponential(invlink=tf.square),
rng.rand(10, 2).astype(np_float_type), 1e-6))
test_setups.append(TestSetup(GPflow.likelihoods.Gamma(invlink=tf.square),
rng.rand(10, 2).astype(np_float_type), 1e-6))
def sigmoid(x):
return 1./(1 + tf.exp(-x))
test_setups.append(TestSetup(GPflow.likelihoods.Bernoulli(invlink=sigmoid),
rng.rand(10, 2).astype(np_float_type), 1e-6))
return test_setups
class TestPredictConditional(unittest.TestCase):
"""
Here we make sure that the conditional_mean and contitional_var functions
give the same result as the predict_mean_and_var function if the prediction
has no uncertainty.
"""
def setUp(self):
tf.reset_default_graph()
self.test_setups = getTestSetups(addNonStandardLinks=True)
self.x = tf.placeholder(float_type)
for test_setup in self.test_setups:
test_setup.likelihood.make_tf_array(self.x)
self.F = tf.placeholder(float_type)
rng = np.random.RandomState(0)
self.F_data = rng.randn(10, 2).astype(np_float_type)
def test_mean(self):
for test_setup in self.test_setups:
l = test_setup.likelihood
with l.tf_mode():
mu1 = tf.Session().run(l.conditional_mean(self.F),
feed_dict={self.x: l.get_free_state(), self.F: self.F_data})
mu2, _ = tf.Session().run(l.predict_mean_and_var(self.F, self.F * 0),
feed_dict={self.x: l.get_free_state(), self.F: self.F_data})
self.assertTrue(np.allclose(mu1, mu2, test_setup.tolerance, test_setup.tolerance))
def test_variance(self):
for test_setup in self.test_setups:
l = test_setup.likelihood
with l.tf_mode():
v1 = tf.Session().run(l.conditional_variance(self.F),
feed_dict={self.x: l.get_free_state(), self.F: self.F_data})
v2 = tf.Session().run(l.predict_mean_and_var(self.F, self.F * 0)[1],
feed_dict={self.x: l.get_free_state(), self.F: self.F_data})
self.assertTrue(np.allclose(v1, v2, atol=test_setup.tolerance))
def test_var_exp(self):
"""
Here we make sure that the variational_expectations gives the same result
as logp if the latent function has no uncertainty.
"""
for test_setup in self.test_setups:
l = test_setup.likelihood
y = test_setup.Y
with l.tf_mode():
r1 = tf.Session().run(l.logp(self.F, y), feed_dict={self.x: l.get_free_state(), self.F: self.F_data})
r2 = tf.Session().run(l.variational_expectations(self.F, self.F * 0, test_setup.Y),
feed_dict={self.x: l.get_free_state(), self.F: self.F_data})
self.assertTrue(np.allclose(r1, r2, test_setup.tolerance, test_setup.tolerance))
class TestQuadrature(unittest.TestCase):
"""
Where quadratre methods have been overwritten, make sure the new code
does something close to the quadrature
"""
def setUp(self):
tf.reset_default_graph()
self.rng = np.random.RandomState()
self.Fmu, self.Fvar, self.Y = self.rng.randn(3, 10, 2).astype(np_float_type)
self.Fvar = 0.01 * self.Fvar ** 2
self.test_setups = getTestSetups(includeMultiClass=False)
def test_var_exp(self):
# get all the likelihoods where variational expectations has been overwritten
for test_setup in self.test_setups:
if not test_setup.is_analytic:
continue
l = test_setup.likelihood
y = test_setup.Y
x_data = l.get_free_state()
x = tf.placeholder(float_type)
l.make_tf_array(x)
# 'build' the functions
with l.tf_mode():
F1 = l.variational_expectations(self.Fmu, self.Fvar, y)
F2 = GPflow.likelihoods.Likelihood.variational_expectations(l, self.Fmu, self.Fvar, y)
# compile and run the functions:
F1 = tf.Session().run(F1, feed_dict={x: x_data})
F2 = tf.Session().run(F2, feed_dict={x: x_data})
self.assertTrue(np.allclose(F1, F2, test_setup.tolerance, test_setup.tolerance))
def test_pred_density(self):
# get all the likelihoods where predict_density has been overwritten.
for test_setup in self.test_setups:
if not test_setup.is_analytic:
continue
l = test_setup.likelihood
y = test_setup.Y
x_data = l.get_free_state()
# make parameters if needed
x = tf.placeholder(float_type)
l.make_tf_array(x)
# 'build' the functions
with l.tf_mode():
F1 = l.predict_density(self.Fmu, self.Fvar, y)
F2 = GPflow.likelihoods.Likelihood.predict_density(l, self.Fmu, self.Fvar, y)
# compile and run the functions:
F1 = tf.Session().run(F1, feed_dict={x: x_data})
F2 = tf.Session().run(F2, feed_dict={x: x_data})
self.assertTrue(np.allclose(F1, F2, test_setup.tolerance, test_setup.tolerance))
class TestRobustMaxMulticlass(unittest.TestCase):
"""
Some specialized tests to the multiclass likelihood with RobustMax inverse link function.
"""
def setUp(self):
tf.reset_default_graph()
def testSymmetric(self):
"""
This test is based on the observation that for
symmetric inputs the class predictions must have equal probability.
"""
nClasses = 5
nPoints = 10
tolerance = 1e-4
epsilon = 1e-3
F = tf.placeholder(float_type)
x = tf.placeholder(float_type)
F_data = np.ones((nPoints, nClasses))
l = GPflow.likelihoods.MultiClass(nClasses)
l.invlink.epsilon = epsilon
rng = np.random.RandomState(1)
Y = rng.randint(nClasses, size=(nPoints, 1))
with l.tf_mode():
mu, _ = tf.Session().run(l.predict_mean_and_var(F, F), feed_dict={x: l.get_free_state(), F: F_data})
pred = tf.Session().run(l.predict_density(F, F, Y), feed_dict={x: l.get_free_state(), F: F_data})
variational_expectations = tf.Session().run(l.variational_expectations(F, F, Y),
feed_dict={x: l.get_free_state(), F: F_data})
expected_mu = (1./nClasses * (1. - epsilon) + (1. - 1./nClasses) * epsilon / (nClasses - 1)) * np.ones((nPoints, 1))
self.assertTrue(np.allclose(mu, expected_mu, tolerance, tolerance))
expected_log_denisty = np.log(expected_mu)
self.assertTrue(np.allclose(pred, expected_log_denisty, 1e-3, 1e-3))
validation_variational_expectation = 1./nClasses * np.log(1. - epsilon) + \
(1. - 1./nClasses) * np.log(epsilon / (nClasses - 1))
self.assertTrue(np.allclose(variational_expectations,
np.ones((nPoints, 1)) * validation_variational_expectation,
tolerance, tolerance))
def testPredictDensity(self):
tol = 1e-4
num_points = 100
mock_prob = 0.73
class MockRobustMax(GPflow.likelihoods.RobustMax):
def prob_is_largest(self, Y, Fmu, Fvar, gh_x, gh_w):
return tf.ones((num_points, 1)) * mock_prob
epsilon = 0.231
num_classes = 5
l = GPflow.likelihoods.MultiClass(num_classes, invlink=MockRobustMax(num_classes, epsilon))
F = tf.placeholder(float_type)
y = tf.placeholder(float_type)
F_data = np.ones((num_points, num_classes))
rng = np.random.RandomState(1)
Y_data = rng.randint(num_classes, size=(num_points, 1))
with l.tf_mode():
pred = tf.Session().run(l.predict_density(F, F, y), feed_dict={F: F_data, y: Y_data})
expected_prediction = -0.5499780059
#^ evaluated on calculator: log( (1-\epsilon) * 0.73 + (1-0.73) * \epsilon/(num_classes -1))
self.assertTrue(np.allclose(pred, expected_prediction, tol, tol))
class TestMulticlassIndexFix(unittest.TestCase):
"""
A regression test for a bug in multiclass likelihood.
"""
def setUp(self):
tf.reset_default_graph()
def testA(self):
mu, var = tf.placeholder(float_type), tf.placeholder(float_type)
Y = tf.placeholder(tf.int32)
lik = GPflow.likelihoods.MultiClass(3)
ve = lik.variational_expectations(mu, var, Y)
tf.gradients(tf.reduce_sum(ve), mu)
class TestSwitchedLikelihood(unittest.TestCase):
"""
SwitchedLikelihood is saparately tested here.
Here, we make sure the partition-stictch works fine.
"""
def setUp(self):
rng = np.random.RandomState(1)
self.Y_list = [rng.randn(3, 2), rng.randn(4, 2), rng.randn(5, 2)]
self.F_list = [rng.randn(3, 2), rng.randn(4, 2), rng.randn(5, 2)]
self.Fvar_list = [np.exp(rng.randn(3, 2)), np.exp(rng.randn(4, 2)), np.exp(rng.randn(5, 2))]
self.Y_label = [np.ones((3, 1))*0, np.ones((4, 1))*1, np.ones((5, 1))*2]
self.Y_perm = list(range(3+4+5))
rng.shuffle(self.Y_perm)
# shuffle the original data
self.Y_sw = np.hstack([np.concatenate(self.Y_list), np.concatenate(self.Y_label)])[self.Y_perm, :]
self.F_sw = np.concatenate(self.F_list)[self.Y_perm, :]
self.Fvar_sw = np.concatenate(self.Fvar_list)[self.Y_perm, :]
# likelihoods
self.likelihoods = [GPflow.likelihoods.Gaussian(),
GPflow.likelihoods.Gaussian(),
GPflow.likelihoods.Gaussian()]
for lik in self.likelihoods:
lik.variance = np.exp(rng.randn(1))
self.switched_likelihood = GPflow.likelihoods.SwitchedLikelihood(self.likelihoods)
# initialize switched likelihood
self.sess = tf.Session()
free_array = self.switched_likelihood.get_free_state()
self.switched_likelihood.make_tf_array(free_array)
self.sess.run(tf.global_variables_initializer())
def test_logp(self):
# switchedlikelihood
with self.switched_likelihood.tf_mode():
switched_rslt = self.sess.run(self.switched_likelihood.logp(self.F_sw, self.Y_sw))
# likelihood
rslts = []
for lik, y, f in zip(self.likelihoods, self.Y_list, self.F_list):
with lik.tf_mode():
rslts.append(self.sess.run(lik.logp(f, y)))
self.assertTrue(np.allclose(switched_rslt, np.concatenate(rslts)[self.Y_perm, :]))
self.sess.close()
def test_predict_density(self):
# switchedlikelihood
with self.switched_likelihood.tf_mode():
switched_rslt = self.sess.run(
self.switched_likelihood.predict_density(self.F_sw, self.Fvar_sw, self.Y_sw))
# likelihood
rslts = []
for lik, y, f, fvar in zip(self.likelihoods, self.Y_list, self.F_list, self.Fvar_list):
self.sess.run(tf.global_variables_initializer())
with lik.tf_mode():
rslts.append(self.sess.run(lik.predict_density(f, fvar, y)))
self.assertTrue(np.allclose(switched_rslt, np.concatenate(rslts)[self.Y_perm, :]))
self.sess.close()
def test_variational_expectations(self):
# switchedlikelihood
with self.switched_likelihood.tf_mode():
switched_rslt = self.sess.run(
self.switched_likelihood.variational_expectations(self.F_sw, self.Fvar_sw, self.Y_sw))
# likelihood
rslts = []
for lik, y, f, fvar in zip(self.likelihoods, self.Y_list, self.F_list, self.Fvar_list):
self.sess.run(tf.global_variables_initializer())
with lik.tf_mode():
rslts.append(self.sess.run(lik.variational_expectations(f, fvar, y)))
self.assertTrue(np.allclose(switched_rslt, np.concatenate(rslts)[self.Y_perm, :]))
self.sess.close()
if __name__ == "__main__":
unittest.main()
