def test_tanh_normal_log_prob_of_clipped_action(self):
"""Verify that clipped actions still have a valid log probability."""
mean = torch.zeros(2)
std = torch.ones(2)
dist = TanhNormal(mean, std)
action = torch.Tensor([[1., -1.]])
log_prob_approx = dist.log_prob(action)
assert torch.isfinite(log_prob_approx)
assert log_prob_approx < -20
del dist
def test_tanh_normal_rsample(self):
"""Test the bounds of the tanh_normal rsample function."""
mean = torch.zeros(1)
std = torch.ones(1)
dist = TanhNormal(mean, std)
sample = dist.rsample()
pre_tanh_action, action = dist.rsample_with_pre_tanh_value()
assert (pre_tanh_action.tanh() == action).all()
assert -1 <= action <= 1.
assert -1 <= sample <= 1.
del dist
def test_tanh_normal_bounds(self):
"""Test to make sure the tanh_normal dist obeys the bounds (-1,1)."""
mean = torch.ones(1) * 100
std = torch.ones(1) * 100
dist = TanhNormal(mean, std)
assert dist.mean <= 1.
del dist
mean = torch.ones(1) * -100
std = torch.ones(1) * 100
dist = TanhNormal(mean, std)
assert dist.mean >= -1.
def test_tanh_normal_log_prob(self):
"""Verify the correctnes of the tanh_normal log likelihood function."""
mean = torch.zeros(1)
std = torch.ones(1)
dist = TanhNormal(mean, std)
pre_tanh_action = torch.Tensor([[2.0960]])
action = pre_tanh_action.tanh()
log_prob = dist.log_prob(action, pre_tanh_action)
log_prob_approx = dist.log_prob(action)
assert torch.allclose(log_prob, torch.Tensor([-0.2798519]))
assert torch.allclose(log_prob_approx, torch.Tensor([-0.2798185]))
del dist
def test_tanh_normal_expand(self):
"""Test for expand function.
Checks whether expand returns a distribution that has potentially a
different batch size from the already existing distribution.
"""
mean = torch.zeros(1)
std = torch.ones(1)
dist = TanhNormal(mean, std)
new_dist = dist.expand((2, ))
sample = new_dist.sample()
assert sample.shape == torch.Size((2, 1))
def test_new_tanh_normal(self):
"""Tests the tanh_normal constructor."""
mean = torch.ones(1)
std = torch.ones(1)
dist = TanhNormal(mean, std)
del dist
def test_tanh_normal_repr(self):
"""Test that the repr function outputs the class name."""
mean = torch.zeros(1)
std = torch.ones(1)
dist = TanhNormal(mean, std)
assert repr(dist) == 'TanhNormal'