Added Weibull Distribution

ngboost/distns/__init__.py

@@ -10,6 +10,7 @@
 from .normal import Normal, NormalFixedMean, NormalFixedVar
 from .poisson import Poisson
 from .t import T, TFixedDf, TFixedDfFixedVar
+from .weibull import Weibull
 
 __all__ = [
     "Bernoulli",
@@ -30,4 +31,5 @@
     "T",
     "TFixedDf",
     "TFixedDfFixedVar",
+    "Weibull",
 ]

ngboost/distns/weibull.py

@@ -0,0 +1,65 @@
+"""The NGBoost Weibull distribution and scores"""
+import numpy as np
+from scipy.stats import weibull_min as dist
+
+from ngboost.distns.distn import RegressionDistn
+from ngboost.scores import LogScore
+
+
+class WeibullLogScore(LogScore):
+    def score(self, Y):
+        return -self.dist.logpdf(Y)
+
+    def d_score(self, Y):
+        D = np.zeros((len(Y), 2))
+        shared_term = self.shape * ((Y / self.scale) ** self.shape - 1)
+        D[:, 0] = shared_term * np.log(Y / self.scale) - 1
+        D[:, 1] = -shared_term
+
+        return D
+
+    def metric(self):
+        gamma = 0.5772156649  # Euler's constant
+        FI = np.zeros((self.scale.shape[0], 2, 2))
+        FI[:, 0, 0] = (np.pi**2 / 6) + (1 - gamma) ** 2
+        FI[:, 1, 0] = -self.shape * (1 - gamma)
+        FI[:, 0, 1] = FI[:, 1, 0]
+        FI[:, 1, 1] = self.shape**2
+
+        return FI
+
+
+class Weibull(RegressionDistn):
+    """
+    Implements the Weibull distribution for NGBoost.
+
+    The Weibull distribution has two parameters, shape and scale.
+    The scipy loc parameter is held constant for this implementation.
+    LogScore is supported for the Weibull distribution.
+    """
+
+    n_params = 2
+    scores = [WeibullLogScore]
+
+    # pylint: disable=super-init-not-called
+    def __init__(self, params):
+        self._params = params
+        self.shape = np.exp(params[0])  # shape (c)
+        self.scale = np.exp(params[1])  # scale (labmda)
+        self.dist = dist(c=self.shape, loc=0, scale=self.scale)
+
+    def fit(Y):
+        shape, _loc, scale = dist.fit(Y, floc=0)  # hold loc constant
+        return np.array([np.log(shape), np.log(scale)])
+
+    def sample(self, m):
+        return np.array([self.dist.rvs() for i in range(m)])
+
+    def __getattr__(self, name):
+        if name in dir(self.dist):
+            return getattr(self.dist, name)
+        return None
+
+    @property
+    def params(self):
+        return {"shape": self.shape, "scale": self.scale}

tests/test_distns.py

@@ -21,6 +21,7 @@
     T,
     TFixedDf,
     TFixedDfFixedVar,
+    Weibull,
     k_categorical,
 )
 from ngboost.scores import CRPScore, LogScore, Score
@@ -74,6 +75,7 @@ def is_t_distribution(
         TFixedDf,
         TFixedDfFixedVar,
         Cauchy,
+        Weibull,
     ],
 )
 @pytest.mark.parametrize(

tests/test_score.py

@@ -16,6 +16,7 @@
     T,
     TFixedDf,
     TFixedDfFixedVar,
+    Weibull,
 )
 from ngboost.manifold import manifold
 from ngboost.scores import CRPScore, LogScore, Score
@@ -100,6 +101,7 @@ def idfn(dist_score: DistScore):
     (Laplace, LogScore),
     (Poisson, LogScore),
     (Gamma, LogScore),
+    (Weibull, LogScore),
 ] + [(MultivariateNormal(i), LogScore) for i in range(2, 5)]
 # Fill in the dist, score pair to test the gradient
 # Tests all in TEST_METRIC by default