Added PPF tests

enterprise/signals/parameter.py

@@ -266,6 +266,9 @@ def NormalPPF(value, mu, sigma):
     Handles scalar mu and sigma, compatible vector value/mu/sigma,
     vector value/mu and compatible covariance matrix sigma."""
 
+    if np.ndim(sigma) == 2:
+        raise NotImplementedError("PPF not implemented when sigma is 2D")
+
     return sstats.norm.ppf(value, loc=mu, scale=sigma)
 
 
@@ -285,7 +288,7 @@ def Normal(mu=0, sigma=1, size=None):
     class Normal(Parameter):
         _size = size
         _prior = Function(NormalPrior, mu=mu, sigma=sigma)
-        _ppf = Function(NormaPPF, mu=mu, sigma=sigma)
+        _ppf = Function(NormalPPF, mu=mu, sigma=sigma)
         _sampler = staticmethod(NormalSampler)
         _typename = _argrepr("Normal", mu=mu, sigma=sigma)
 

tests/test_parameter.py

@@ -13,10 +13,10 @@
 import numpy as np
 import scipy.stats
 
-from enterprise.signals.parameter import UniformPrior, UniformSampler, Uniform
-from enterprise.signals.parameter import NormalPrior, NormalSampler, Normal
+from enterprise.signals.parameter import UniformPrior, UniformSampler, Uniform, UniformPPF
+from enterprise.signals.parameter import NormalPrior, NormalSampler, Normal, NormalPPF
 from enterprise.signals.parameter import TruncNormalPrior, TruncNormalSampler, TruncNormal
-from enterprise.signals.parameter import LinearExpPrior, LinearExpSampler
+from enterprise.signals.parameter import LinearExpPrior, LinearExpSampler, LinearExpPPF
 
 
 class TestParameter(unittest.TestCase):
@@ -35,6 +35,9 @@ def test_uniform(self):
         assert p_min < x1 < p_max, msg2
         assert type(x1) == float, msg2
 
+        msg3 = "Enterprise and scipy PPF do not match"
+        assert np.allclose(UniformPPF(x, p_min, p_max), scipy.stats.uniform.ppf(x, p_min, p_max - p_min)), msg3
+
         # vector argument
         x = np.array([0.5, 0.1])
         assert np.allclose(UniformPrior(x, p_min, p_max), scipy.stats.uniform.pdf(x, p_min, p_max - p_min)), msg1
@@ -43,9 +46,13 @@ def test_uniform(self):
         assert np.all((p_min < x1) & (x1 < p_max)), msg2
         assert x1.shape == (3,), msg2
 
+        # vector argument
+        assert np.allclose(UniformPPF(x, p_min, p_max), scipy.stats.uniform.ppf(x, p_min, p_max - p_min)), msg3
+
         # vector bounds
         p_min, p_max = np.array([0.2, 0.3]), np.array([1.1, 1.2])
         assert np.allclose(UniformPrior(x, p_min, p_max), scipy.stats.uniform.pdf(x, p_min, p_max - p_min)), msg1
+        assert np.allclose(UniformPPF(x, p_min, p_max), scipy.stats.uniform.ppf(x, p_min, p_max - p_min)), msg3
 
         x1 = UniformSampler(p_min, p_max)
         assert np.all((p_min < x1) & (x1 < p_max)), msg2
@@ -68,6 +75,10 @@ def test_linearexp(self):
         msg1b = "Scalar sampler out of range"
         assert p_min <= x <= p_max, msg1b
 
+        msg1c = "Scalar PPF does not match"
+        x = 0.5
+        assert np.allclose(LinearExpPPF(x, p_min, p_max), np.log10(10**p_min + x*(10**p_max-10**p_min))), msg1c
+
         # vector argument
         x = np.array([0, 1.5, 2.5])
         msg2 = "Vector-argument prior does not match"
@@ -79,6 +90,13 @@ def test_linearexp(self):
         msg2b = "Vector-argument sampler out of range"
         assert np.all((p_min < x) & (x < p_max)), msg2b
 
+        x = np.array([0.5, 0.75])
+        msg2c = "Vector-argument PPF does not match"
+        assert np.allclose(
+            LinearExpPPF(x, p_min, p_max),
+            np.log10(10**p_min + x * (10**p_max-10**p_min))
+        ), msg2c
+
         # vector bounds
         p_min, p_max = np.array([0, 1]), np.array([2, 3])
         x = np.array([1, 2])
@@ -88,6 +106,15 @@ def test_linearexp(self):
             np.array([10**1 / (10**2 - 10**0), 10**2 / (10**3 - 10**1)]) * np.log(10),
         ), msg3
 
+        # Vector PPF
+        x = np.array([0.5, 0.75])
+        p_min, p_max = np.array([0, 1]), np.array([2, 3])
+        msg3c = "Vector-argument PPF+bounds does not match"
+        assert np.allclose(
+            LinearExpPPF(x, p_min, p_max),
+            np.log10(10**p_min + x * (10**p_max-10**p_min))
+        ), msg3c
+
     def test_normal(self):
         """Test Normal parameter prior and sampler for various combinations of scalar and vector arguments."""
 
@@ -105,6 +132,11 @@ def test_normal(self):
         # this should almost never fail
         assert -5 < (x1 - mu) / sigma < 5, msg2
 
+        msg3 = "Enterprise and scipy PPF do not match"
+        assert np.allclose(
+            NormalPPF(x, mu, sigma), scipy.stats.norm.ppf(x, loc=mu, scale=sigma)
+        ), msg3
+
         # vector argument
         x = np.array([-0.2, 0.1, 0.5])
 
@@ -118,6 +150,11 @@ def test_normal(self):
         )
         assert x1.shape == x2.shape, msg2
 
+        x = np.array([0.1, 0.25, 0.65])
+        assert np.allclose(
+            NormalPPF(x, mu, sigma), scipy.stats.norm.ppf(x, loc=mu, scale=sigma)
+        ), msg3
+
         # vector bounds; note the different semantics from `NormalPrior`,
         # which returns a vector consistently with `UniformPrior`
         mu, sigma = np.array([0.1, 0.15, 0.2]), np.array([2, 1, 2])