def test_get_metric_fn():
"""Tests probflow.utils.metrics.get_metric_fn"""
metric_fn = metrics.get_metric_fn("f1")
# Predictive dist
probs = tf.constant([0, 1, 1, 1, 1, 0], dtype=tf.float32)
pred_dist = Bernoulli(probs=probs)
y_true = np.array([1, 0, 1, 1, 0, 0]).astype("float32")
# Compare metric
ppv = 2 / 4
tpr = 2 / 3
f1 = 2 * (ppv * tpr) / (ppv + tpr)
assert is_close(metric_fn(y_true, pred_dist.mean()), f1)
# Should be able to pass a callable
metric_fn = metrics.get_metric_fn(lambda x, y: 3)
assert metric_fn(y_true, pred_dist.mean()) == 3
# Should raise a type error if passed anything else
with pytest.raises(TypeError):
metrics.get_metric_fn(3)
with pytest.raises(TypeError):
metrics.get_metric_fn([1, 2, 3])
with pytest.raises(TypeError):
metrics.get_metric_fn({"apples": 1, "oranges": 2})
# And value error if invalid string
with pytest.raises(ValueError):
metrics.get_metric_fn("asdf")
def test_accuracy():
"""Tests probflow.utils.metrics.accuracy"""
# Predictive dist
probs = tf.constant([1, 1, 1, 1, 1, 1], dtype=tf.float32)
pred_dist = Bernoulli(probs=probs)
y_true = np.array([1, 0, 1, 1, 0, 1]).astype("float32")
# Compare metric
assert is_close(metrics.accuracy(y_true, pred_dist.mean()), 2.0 / 3.0)
def test_precision():
"""Tests probflow.utils.metrics.precision"""
# Predictive dist
probs = tf.constant([1, 1, 1, 1, 1, 0], dtype=tf.float32)
pred_dist = Bernoulli(probs=probs)
y_true = np.array([1, 0, 1, 1, 0, 0]).astype("float32")
# Compare metric
assert is_close(metrics.precision(y_true, pred_dist.mean()), 3.0 / 5.0)
def test_true_positive_rate():
"""Tests probflow.utils.metrics.true_positive_rate"""
# Predictive dist
probs = tf.constant([1, 1, 1, 1, 1, 0], dtype=tf.float32)
pred_dist = Bernoulli(probs=probs)
y_true = np.array([1, 0, 1, 1, 0, 1]).astype("float32")
# Compare metric
assert is_close(metrics.true_positive_rate(y_true, pred_dist.mean()), 0.75)
def test_f1_score():
"""Tests probflow.utils.metrics.f1_score"""
# Predictive dist
probs = tf.constant([0, 1, 1, 1, 1, 0], dtype=tf.float32)
pred_dist = Bernoulli(probs=probs)
y_true = np.array([1, 0, 1, 1, 0, 0]).astype("float32")
# Compare metric
ppv = 2 / 4
tpr = 2 / 3
f1 = 2 * (ppv * tpr) / (ppv + tpr)
assert is_close(metrics.f1_score(y_true, pred_dist.mean()), f1)
def __call__(self, x):
return Bernoulli(x @ self.weight() + self.bias())
def __call__(self, x):
x = to_tensor(x)
return Bernoulli(x @ self.weight() + self.bias())
def test_Bernoulli():
"""Tests Bernoulli distribution"""
# Create the distribution
dist = Bernoulli(0)
# Check default params
assert dist.logits == 0
assert dist.probs is None
# Call should return backend obj
assert isinstance(dist(), tfd.Bernoulli)
# Test methods
assert is_close(dist.prob(0).numpy(), 0.5)
assert is_close(dist.prob(1).numpy(), 0.5)
assert is_close(dist.log_prob(0).numpy(), np.log(0.5))
assert is_close(dist.log_prob(1).numpy(), np.log(0.5))
assert dist.mean().numpy() == 0.5
# Test sampling
samples = dist.sample()
assert isinstance(samples, tf.Tensor)
assert samples.ndim == 0
samples = dist.sample(10)
assert isinstance(samples, tf.Tensor)
assert samples.ndim == 1
assert samples.shape[0] == 10
# Should be able to set params
dist = Bernoulli(probs=0.8)
assert dist.probs == 0.8
assert dist.logits is None
assert is_close(dist.prob(0).numpy(), 0.2)
assert is_close(dist.prob(1).numpy(), 0.8)
# But only with Tensor-like objs
with pytest.raises(TypeError):
dist = Bernoulli("lalala")
with pytest.raises(TypeError):
dist = Bernoulli()
def __call__(self, x):
return Bernoulli(x @ self.w())
def __call__(self, x):
return Bernoulli(torch.tensor(x) @ self.w())
def test_Bernoulli():
"""Tests Bernoulli distribution"""
# Create the distribution
dist = Bernoulli(0)
# Check default params
assert dist.logits == 0
assert dist.probs is None
# Call should return backend obj
assert isinstance(dist(), tod.bernoulli.Bernoulli)
# Test methods
zero = torch.zeros([1])
one = torch.ones([1])
assert is_close(dist.prob(zero).numpy(), 0.5)
assert is_close(dist.prob(one).numpy(), 0.5)
assert is_close(dist.log_prob(zero).numpy(), np.log(0.5))
assert is_close(dist.log_prob(one).numpy(), np.log(0.5))
assert dist.mean().numpy() == 0.5
# Test sampling
samples = dist.sample()
assert isinstance(samples, torch.Tensor)
assert samples.ndim == 0
samples = dist.sample(10)
assert isinstance(samples, torch.Tensor)
assert samples.ndim == 1
assert samples.shape[0] == 10
# Need to pass tensor-like probs
with pytest.raises(TypeError):
dist = Bernoulli("lalala")
# Should be able to set params
dist = Bernoulli(probs=0.8)
assert dist.probs == 0.8
assert dist.logits is None
assert is_close(dist.prob(zero).numpy(), 0.2)
assert is_close(dist.prob(one).numpy(), 0.8)