Add mode argument to maxent (#603)

* Add mode argument to maxent

* disable abstract-method

* raise from exc

.pylintrc

@@ -64,7 +64,8 @@ disable=missing-docstring,
         too-few-public-methods,
         import-error,
         protected-access,
-        no-self-use
+        no-self-use,
+        abstract-method,
 
 
 # Enable the message, report, category or checker with the given id(s). You can

preliz/distributions/bernoulli.py

@@ -119,6 +119,9 @@ def entropy(self):
     def mean(self):
         return self.p
 
+    def mode(self):
+        return self.median()
+
     def median(self):
         return np.where(self.p <= 0.5, 0, 1)
 

preliz/distributions/beta.py

@@ -167,6 +167,11 @@ def entropy(self):
     def mean(self):
         return self.alpha / (self.alpha + self.beta)
 
+    def mode(self):
+        return np.where(
+            (self.alpha > 1) & (self.beta > 1), (self.alpha - 1) / (self.alpha + self.beta - 2), 0.5
+        )
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/cauchy.py

@@ -89,6 +89,9 @@ def entropy(self):
     def mean(self):
         return np.nan
 
+    def mode(self):
+        return self.alpha
+
     def median(self):
         return self.alpha
 

preliz/distributions/distributions.py

@@ -64,7 +64,6 @@ def __repr__(self):
                 else f"{n}=[" + "".join(f"{vi:.3g}, " for vi in v).strip(", ") + "], "
                 for n, v in zip(self.param_names, self.params)
             ).strip(", ")
-
             return f"{bolded_name}({description})"
         else:
             return name
@@ -141,7 +140,7 @@ def rvs(self, size=None, random_state=None):
         random_state : {None, int, numpy.random.Generator, numpy.random.RandomState}
             Defaults to None
         """
-        return self.rvs(size, random_state)
+        raise NotImplementedError
 
     def pdf(self, x):
         """Probability density/mass function.
@@ -151,7 +150,7 @@ def pdf(self, x):
         x : array_like
             Values on which to evaluate the pdf
         """
-        return self.pdf(x)
+        raise NotImplementedError
 
     def cdf(self, x):
         """Cumulative distribution function.
@@ -161,7 +160,7 @@ def cdf(self, x):
         x : array_like
             Values on which to evaluate the cdf
         """
-        return self.cdf(x)
+        raise NotImplementedError
 
     def ppf(self, q):
         """Percent point function (inverse of cdf).
@@ -171,7 +170,7 @@ def ppf(self, q):
         x : array_like
             Values on which to evaluate the inverse of the cdf
         """
-        return self.ppf(q)
+        raise NotImplementedError
 
     def logpdf(self, x):
         """Log probability density/mass function.
@@ -181,35 +180,39 @@ def logpdf(self, x):
         x : array_like
             Values on which to evaluate the logpdf
         """
-        return self.pdf(x)
+        raise NotImplementedError
 
     def entropy(self):
         """Entropy"""
-        return self.entropy()
+        raise NotImplementedError
 
     def mean(self):
         """Mean."""
-        return self.mean()
+        raise NotImplementedError
 
     def median(self):
         """Median."""
-        return self.median()
+        raise NotImplementedError
+
+    def mode(self):
+        """Mode."""
+        raise NotImplementedError
 
     def std(self):
         """Standard deviation."""
-        return self.std()
+        raise NotImplementedError
 
     def var(self):
         """Variance"""
-        return self.var()
+        raise NotImplementedError
 
     def skewness(self):
         """Skewness."""
-        return self.skewness()
+        raise NotImplementedError
 
     def kurtosis(self):
         """Excess kurtosis."""
-        return self.kurtosis()
+        raise NotImplementedError
 
     def moments(self, types="mvsk"):
         """

preliz/distributions/exponential.py

@@ -110,6 +110,9 @@ def median(self):
     def mean(self):
         return self.beta
 
+    def mode(self):
+        return np.zeros_like(self.beta)
+
     def std(self):
         return self.beta
 

preliz/distributions/gamma.py

@@ -137,6 +137,9 @@ def entropy(self):
     def mean(self):
         return self.alpha / self.beta
 
+    def mode(self):
+        return np.where(self.alpha <= 1, 0, (self.alpha - 1) / self.beta)
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/geometric.py

@@ -82,6 +82,9 @@ def entropy(self):
     def mean(self):
         return 1 / self.p
 
+    def mode(self):
+        return np.ones_like(self.p)
+
     def median(self):
         return np.ceil(-1 / np.log(1 - self.p))
 

preliz/distributions/gumbel.py

@@ -96,6 +96,9 @@ def entropy(self):
     def mean(self):
         return self.mu + self.beta * np.euler_gamma
 
+    def mode(self):
+        return self.mu
+
     def median(self):
         return self.mu - self.beta * np.log(np.log(2))
 

preliz/distributions/halfcauchy.py

@@ -83,6 +83,9 @@ def entropy(self):
     def mean(self):
         return np.inf
 
+    def mode(self):
+        return np.zeros_like(self.beta)
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/halfnormal.py

@@ -108,6 +108,9 @@ def entropy(self):
     def mean(self):
         return self.sigma * 0.7978845608028655
 
+    def mode(self):
+        return np.zeros_like(self.sigma)
+
     def median(self):
         return self.sigma * 0.6744897501960818
 

preliz/distributions/halfstudentt.py

@@ -144,6 +144,9 @@ def mean(self):
                 mean = self.sigma * (2 / np.pi) ** 0.5
         return mean
 
+    def mode(self):
+        return np.zeros_like(self.sigma)
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/inversegamma.py

@@ -125,6 +125,9 @@ def entropy(self):
     def mean(self):
         return np.where(self.alpha > 1, self.beta / (self.alpha - 1), np.inf)
 
+    def mode(self):
+        return self.beta / (self.alpha + 1)
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/laplace.py

@@ -91,6 +91,9 @@ def median(self):
     def mean(self):
         return self.mu
 
+    def mode(self):
+        return self.mu
+
     def std(self):
         return self.var() ** 0.5
 

preliz/distributions/logistic.py

@@ -88,6 +88,9 @@ def entropy(self):
     def mean(self):
         return self.mu
 
+    def mode(self):
+        return self.mu
+
     def median(self):
         return self.mu
 

preliz/distributions/lognormal.py

@@ -95,6 +95,9 @@ def entropy(self):
     def mean(self):
         return np.exp(self.mu + self.sigma**2 / 2)
 
+    def mode(self):
+        return np.exp(self.mu - self.sigma**2)
+
     def median(self):
         return np.exp(self.mu)
 

preliz/distributions/moyal.py

@@ -99,6 +99,9 @@ def entropy(self):
     def mean(self):
         return self.mu + self.sigma * (np.euler_gamma + np.log(2))
 
+    def mode(self):
+        return self.mu
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/negativebinomial.py

@@ -141,6 +141,9 @@ def entropy(self):
     def mean(self):
         return self.mu
 
+    def mode(self):
+        return np.where(self.n < 1, 0, np.floor((self.n - 1) * (1 - self.p) / self.p))
+
     def median(self):
         return self.ppf(0.5)
 

preliz/distributions/normal.py

@@ -116,6 +116,9 @@ def entropy(self):
     def mean(self):
         return self.mu
 
+    def mode(self):
+        return self.mu
+
     def median(self):
         return self.mu
 

preliz/distributions/poisson.py

@@ -88,6 +88,9 @@ def entropy(self):
     def mean(self):
         return self.mu
 
+    def mode(self):
+        return np.floor(self.mu)
+
     def median(self):
         return np.floor(self.mu + 1 / 3 - 0.02 / self.mu)
 

preliz/distributions/triangular.py

@@ -101,6 +101,9 @@ def entropy(self):
     def mean(self):
         return (self.lower + self.c + self.upper) / 3
 
+    def mode(self):
+        return self.c
+
     def median(self):
         return np.where(
             self.c >= (self.lower + self.upper) / 2,

preliz/internal/optimization.py

@@ -10,7 +10,7 @@
 from .distribution_helper import init_vals as default_vals
 
 
-def optimize_max_ent(dist, lower, upper, mass, none_idx, fixed):
+def optimize_max_ent(dist, lower, upper, mass, none_idx, fixed, mode=None):
     def prob_bound(params, dist, lower, upper, mass):
         params = get_params(dist, params, none_idx, fixed)
         dist._parametrization(**params)
@@ -19,6 +19,8 @@ def prob_bound(params, dist, lower, upper, mass):
         cdf0 = dist.cdf(lower)
         cdf1 = dist.cdf(upper)
         loss = (cdf1 - cdf0) - mass
+        if mode is not None:
+            loss = loss - abs(dist.mode() - mode)
         return loss
 
     def entropy_loss(params, dist):

preliz/tests/test_maxent.py

@@ -180,5 +180,17 @@ def test_maxent(dist, lower, upper, mass, support, result):
     assert_allclose(dist.opt.x, result, atol=0.001)
 
 
+def test_maxent_mode():
+    dist = Beta()
+    maxent(dist, 0.1, 0.7, 0.94, mode=0.3)
+    assert_almost_equal(dist.mode(), 0.3)
+    assert_almost_equal(dist.params, (2.7, 5.1), 1)
+
+    dist = Gamma()
+    maxent(dist, 0, 3, 0.8, mode=2)
+    assert_almost_equal(dist.mode(), 2)
+    assert_almost_equal(dist.params, (7.2, 3.1), 1)
+
+
 def test_maxent_plot():
     maxent(Normal(), plot_kwargs={"support": "restricted", "pointinterval": True})

preliz/unidimensional/__init__.py

@@ -1,4 +1,3 @@
-from .beta_mode import beta_mode
 from .combine import combine
 from .combine_roulette import combine_roulette
 from .maxent import maxent
@@ -8,7 +7,6 @@
 from .roulette import Roulette
 
 __all__ = [
-    "beta_mode",
     "combine",
     "combine_roulette",
     "maxent",