Adding NormalFixedMean (#333)

* Implemented NormalFixedMean

* Added NormalFixedVar and NormalFixedMean to test_distns.py

* Match log(sigma) parameterization of Normal class

* Update __init__.py

Linter is demanding a blank line for ending a file (sigh)

ngboost/distns/__init__.py

@@ -7,7 +7,7 @@
 from .laplace import Laplace
 from .lognormal import LogNormal
 from .multivariate_normal import MultivariateNormal
-from .normal import Normal, NormalFixedVar
+from .normal import Normal, NormalFixedMean, NormalFixedVar
 from .poisson import Poisson
 from .t import T, TFixedDf, TFixedDfFixedVar
 
@@ -24,6 +24,7 @@
     "LogNormal",
     "MultivariateNormal",
     "Normal",
+    "NormalFixedMean",
     "NormalFixedVar",
     "Poisson",
     "T",

ngboost/distns/normal.py

@@ -150,3 +150,65 @@ def __init__(self, params):
     def fit(Y):
         m, _ = sp.stats.norm.fit(Y)
         return m
+
+
+# ### Fixed Mean Normal ###
+class NormalFixedMeanLogScore(LogScore):
+    def score(self, Y):
+        return -self.dist.logpdf(Y)
+
+    def d_score(self, Y):
+        D = np.zeros((len(Y), 1))
+        D[:, 0] = 1 - ((self.loc - Y) ** 2) / self.var
+        return D
+
+    def metric(self):
+        FI = np.zeros((self.var.shape[0], 1, 1))
+        FI[:, 0, 0] = 2
+        return FI
+
+
+class NormalFixedMeanCRPScore(CRPScore):
+    def score(self, Y):
+        Z = (Y - self.loc) / self.scale
+        return self.scale * (
+            Z * (2 * sp.stats.norm.cdf(Z) - 1)
+            + 2 * sp.stats.norm.pdf(Z)
+            - 1 / np.sqrt(np.pi)
+        )
+
+    def d_score(self, Y):
+        Z = (Y - self.loc) / self.scale
+        D = np.zeros((len(Y), 1))
+        D[:, 0] = self.score(Y) + (Y - self.loc) * -1 * (2 * sp.stats.norm.cdf(Z) - 1)
+        return D
+
+    def metric(self):
+        I = np.c_[self.var]
+        I = I.reshape((self.var.shape[0], 1, 1))
+        I = 1 / (2 * np.sqrt(np.pi)) * I
+        return I
+
+
+class NormalFixedMean(Normal):
+    """
+    Implements the normal distribution with mean=0 for NGBoost.
+
+    The fixed-mean normal distribution has one parameter, scale which is the standard deviation.
+    This distribution has both LogScore and CRPScore implemented for it.
+    """
+
+    n_params = 1
+    scores = [NormalFixedMeanLogScore, NormalFixedMeanCRPScore]
+
+    # pylint: disable=super-init-not-called
+    def __init__(self, params):
+        self.loc = np.zeros_like(params[0])
+        self.scale = np.exp(params[0])
+        self.var = self.scale**2
+        self.shape = self.loc.shape
+        self.dist = dist(loc=self.loc, scale=self.scale)
+
+    def fit(Y):
+        _, s = sp.stats.norm.fit(Y)
+        return s

tests/test_distns.py

@@ -16,6 +16,8 @@
     LogNormal,
     MultivariateNormal,
     Normal,
+    NormalFixedMean,
+    NormalFixedVar,
     T,
     TFixedDf,
     TFixedDfFixedVar,
@@ -61,7 +63,18 @@ def is_t_distribution(
 @pytest.mark.slow
 @pytest.mark.parametrize(
     "dist",
-    [Normal, LogNormal, Exponential, Gamma, T, TFixedDf, TFixedDfFixedVar, Cauchy],
+    [
+        Normal,
+        NormalFixedVar,
+        NormalFixedMean,
+        LogNormal,
+        Exponential,
+        Gamma,
+        T,
+        TFixedDf,
+        TFixedDfFixedVar,
+        Cauchy,
+    ],
 )
 @pytest.mark.parametrize(
     "learner",