Merge pull request #132 from amzn/develop

Merge fixes for 0.2.2 into master

.gitignore

@@ -61,3 +61,4 @@ docs/_autosummary
 docs/_build
 docs/generated
 docs/examples
+venv

.travis.yml

@@ -33,12 +33,6 @@ after_success:
   - codecov
 jobs:
   include:
-  - os: osx
-    language: generic
-    env: PYTHON=3.4.4
-  - os: osx
-    language: generic
-    env: PYTHON=3.5.1
   - os: osx
     language: generic
     env: PYTHON=3.6.6
@@ -48,8 +42,8 @@ jobs:
     deploy:
       provider: pypi
       user: meissnereric
-      password:
-        secure: "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"
       skip_existing: true
+      password:
+        secure: jvE8Tkc2a91bJ9FZLxVUUbt85qjX8VvehrMB5wIScHGXzyZ8EemvNfD+2tCBw6k5QbU4QRGBPM29+4e6UWaI8NKwHqUtIu5T8BdXMCNgGIL4EsvXiHLZbnM3srayprT0qGMCzEoBOMCNzskTj0zJSTPFIwXr8o4jiEq5V7eMNlCMdkCRwfejaTXDX5gW+O26IeAEBYWXbM6c0bvoIUlMGH2mg5e35RRTRlVj6ZSoAHS5UnAj05tgq9S+hNBSjBSYuDNXicCdRh9doexxSyvL8eC1coXj9CabL2yXE6EKK+cIsmHCB7XcTzoIF0CyWdm0MDsGB3BT0csz19kMH5RCXZD413d7Vs6h+fo8bS5otxjkKNowOQLQagor8nimnLCrZVuKU/LEl4/KbR1riEJEZgMe4eH+yTFncMLTfa5IDfw+swkpq0L2RDN1RhagGlOjh3k/7BhCKLxvY5iIteBIkJxpZ8FDEEjZ1o2JFM0LqiOogDGHtrGIMLOSey3DIFMvjjbTwjg8H9X3hBVVLtQ26DUMuwAmzRCIpPW25UKARL3KLmpvTm00WEr5hjmIc4BmEwM6XcbLwceiY49bCBKit0dtIE4lg7xV8rdwjZWADwlmTt+BUYktZBNC9gnaOQNirfE6+SIKYEZkVxaer1TIylokwa2D8wQHYWhSjDdjG10=
       on:
         branch: master

mxfusion/components/distributions/__init__.py

@@ -30,13 +30,14 @@
     random_gen
     univariate
     gp
+    laplace
     wishart
     beta
     dirichlet
 """
 
 __all__ = ['bernoulli', 'categorical', 'distribution', 'normal', 'gamma', 'pointmass', 'random_gen',
-           'univariate', 'gp', 'wishart', 'beta']
+           'univariate', 'gp', 'wishart', 'beta', 'laplace', 'uniform', 'dirichlet']
 
 from .distribution import Distribution
 from .normal import Normal, MultivariateNormal, NormalMeanPrecision, MultivariateNormalMeanPrecision
@@ -48,3 +49,5 @@
 from .wishart import Wishart
 from .beta import Beta
 from .dirichlet import Dirichlet
+from .uniform import Uniform
+from .laplace import Laplace

mxfusion/components/distributions/beta.py

@@ -100,10 +100,10 @@ def draw_samples(self, a, b, rv_shape, num_samples=1, F=None):
         ones = F.ones_like(a)
 
         # Sample X from Gamma(a, 1)
-        x = self._rand_gen.sample_gamma(alpha=a, beta=ones, shape=out_shape, dtype=self.dtype, ctx=self.ctx)
+        x = self._rand_gen.sample_gamma(alpha=a, beta=ones, shape=out_shape, dtype=self.dtype, ctx=self.ctx, F=F)
 
         # Sample Y from Gamma(b, 1)
-        y = self._rand_gen.sample_gamma(alpha=b, beta=ones, shape=out_shape, dtype=self.dtype, ctx=self.ctx)
+        y = self._rand_gen.sample_gamma(alpha=b, beta=ones, shape=out_shape, dtype=self.dtype, ctx=self.ctx, F=F)
 
         # Return X / (X + Y)
         return F.broadcast_div(x, F.broadcast_add(x, y))

mxfusion/components/distributions/laplace.py

@@ -0,0 +1,102 @@
+from ...common.config import get_default_MXNet_mode
+from ..variables import Variable
+from .univariate import UnivariateDistribution, UnivariateLogPDFDecorator, UnivariateDrawSamplesDecorator
+
+
+class Laplace(UnivariateDistribution):
+    """
+    The one-dimensional Laplace distribution. The Laplace distribution can be defined over a scalar random variable
+    or an array of random variables. In case of an array of random variables, the location and scale are broadcasted
+    to the shape of the output random variable (array).
+
+    :param location: Location of the Laplace distribution.
+    :type location: Variable
+    :param scale: Scale of the Laplace distribution.
+    :type scale: Variable
+    :param rand_gen: the random generator (default: MXNetRandomGenerator).
+    :type rand_gen: RandomGenerator
+    :param dtype: the data type for float point numbers.
+    :type dtype: numpy.float32 or numpy.float64
+    :param ctx: the mxnet context (default: None/current context).
+    :type ctx: None or mxnet.cpu or mxnet.gpu
+    """
+    def __init__(self, location, scale, rand_gen=None, dtype=None, ctx=None):
+        if not isinstance(location, Variable):
+            location = Variable(value=location)
+        if not isinstance(scale, Variable):
+            scale = Variable(value=scale)
+
+        inputs = [('location', location), ('scale', scale)]
+        input_names = [k for k, _ in inputs]
+        output_names = ['random_variable']
+        super(Laplace, self).__init__(inputs=inputs, outputs=None,
+                                      input_names=input_names,
+                                      output_names=output_names,
+                                      rand_gen=rand_gen, dtype=dtype, ctx=ctx)
+
+    @UnivariateLogPDFDecorator()
+    def log_pdf(self, location, scale, random_variable, F=None):
+        """
+        Computes the logarithm of the probability density function (PDF) of the Laplace distribution.
+
+        :param location: the location of the Laplace distribution.
+        :type location: MXNet NDArray or MXNet Symbol
+        :param scale: the scale of the Laplace distributions.
+        :type scale: MXNet NDArray or MXNet Symbol
+        :param random_variable: the random variable of the Laplace distribution.
+        :type random_variable: MXNet NDArray or MXNet Symbol
+        :param F: the MXNet computation mode (mxnet.symbol or mxnet.ndarray).
+        :returns: log pdf of the distribution.
+        :rtypes: MXNet NDArray or MXNet Symbol
+        """
+        F = get_default_MXNet_mode() if F is None else F
+        logvar = -F.log(2 * scale)
+        logL = F.broadcast_minus(logvar, F.broadcast_div(
+            F.abs(F.broadcast_minus(random_variable, location)), scale)) * self.log_pdf_scaling
+        return logL
+
+    @UnivariateDrawSamplesDecorator()
+    def draw_samples(self, location, scale, rv_shape, num_samples=1, F=None):
+        """
+        Draw samples from the Laplace distribution.
+
+        :param location: the location of the Laplace distribution.
+        :type location: MXNet NDArray or MXNet Symbol
+        :param scale: the scale of the Laplace distributions.
+        :type scale: MXNet NDArray or MXNet Symbol
+        :param rv_shape: the shape of each sample.
+        :type rv_shape: tuple
+        :param num_samples: the number of drawn samples (default: one).
+        :int num_samples: int
+        :param F: the MXNet computation mode (mxnet.symbol or mxnet.ndarray).
+        :returns: a set samples of the Laplace distribution.
+        :rtypes: MXNet NDArray or MXNet Symbol
+        """
+        F = get_default_MXNet_mode() if F is None else F
+        out_shape = (num_samples,) + rv_shape
+        return F.broadcast_add(F.broadcast_mul(self._rand_gen.sample_laplace(
+            shape=out_shape, dtype=self.dtype, ctx=self.ctx),
+            scale), location)
+
+    @staticmethod
+    def define_variable(location=0., scale=1., shape=None, rand_gen=None,
+                        dtype=None, ctx=None):
+        """
+        Creates and returns a random variable drawn from a Laplace distribution.
+
+        :param location: Location of the distribution.
+        :param scale: Scale of the distribution.
+        :param shape: the shape of the random variable(s).
+        :type shape: tuple or [tuple]
+        :param rand_gen: the random generator (default: MXNetRandomGenerator).
+        :type rand_gen: RandomGenerator
+        :param dtype: the data type for float point numbers.
+        :type dtype: numpy.float32 or numpy.float64
+        :param ctx: the mxnet context (default: None/current context).
+        :type ctx: None or mxnet.cpu or mxnet.gpu
+        :returns: the random variables drawn from the Laplace distribution.
+        :rtypes: Variable
+        """
+        var = Laplace(location=location, scale=scale, rand_gen=rand_gen, dtype=dtype, ctx=ctx)
+        var._generate_outputs(shape=shape)
+        return var.random_variable

mxfusion/components/distributions/random_gen.py

@@ -40,6 +40,14 @@ def sample_bernoulli(prob_true=0.5, dtype='bool', F=None):
         pass
 
 
+    @staticmethod
+    def sample_uniform(low=0., high=1., shape=None, dtype=None, out=None, ctx=None, F=None):
+        pass
+
+    @staticmethod
+    def sample_laplace(location=0., scale=1., shape=None, dtype=None, out=None, ctx=None, F=None):
+        pass
+
 class MXNetRandomGenerator(RandomGenerator):
     """
     The MXNet pseudo-random number generator.
@@ -141,6 +149,66 @@ def sample_gamma(alpha=1, beta=1, shape=None, dtype=None, out=None, ctx=None, F=
         :return: Array of samples
         """
         F = get_default_MXNet_mode() if F is None else F
+
         return MXNetRandomGenerator._sample_univariate(
             func=F.random.gamma, alpha=alpha, beta=beta,
             shape=shape, dtype=dtype, out=out, ctx=ctx, F=F)
+
+    @staticmethod
+    def sample_uniform(low=0., high=1., shape=None, dtype=None, out=None, ctx=None, F=None):
+        """
+        Sample uniformly distributed variables
+        Samples are uniformly distributed over the half-open interval [low, high) (includes low, but excludes high).
+
+        :param low: lower boundary of output interval
+        :param high: upper boundary of output interval
+        :param shape: Array shape of samples
+        :param dtype: Data type
+        :param out: output variable
+        :param ctx: execution context
+        :param F: MXNet node
+        :return: Array of samples
+        """
+        F = get_default_MXNet_mode() if F is None else F
+
+        samples = MXNetRandomGenerator._sample_univariate(
+            func=F.random.uniform, shape=shape, dtype=dtype, out=out, ctx=ctx, F=F)
+        # samples = F.broadcast_add(F.broadcast_mul(samples, F.broadcast_sub(high, low)), low)
+        samples = samples * (high - low) + low
+        return samples
+
+    @staticmethod
+    def sample_laplace(location=0., scale=1., shape=None, dtype=None, out=None, ctx=None, F=None):
+        """
+        Sample Laplace distributed variables
+
+        :param location: Location parameter (=mean)
+        :param scale: (>0) Also known as diversity
+        :param shape: Array shape of samples
+        :param dtype: Data type
+        :param out: output variable
+        :param ctx: execution context
+        :param F: MXNet node
+        :return: Array of samples
+        """
+        F = get_default_MXNet_mode() if F is None else F
+
+        # Given a random variable U drawn from the uniform distribution in the interval (-1/2,1/2], the random variable
+        # X =\mu - b\, \sgn(U)\, \ln(1 - 2 | U |)
+        # has a Laplace distribution with parameters \mu and b
+
+        U = MXNetRandomGenerator.sample_uniform(low=-0.5, high=0.5, shape=shape, dtype=dtype, out=out, ctx=ctx, F=F)
+
+        if isinstance(scale, F.NDArray):
+            b_sgn_U = F.broadcast_mul(scale, F.sign(U))
+        else:
+            b_sgn_U = scale * F.sign(U)
+
+        ln_1_2_U = F.log(1 - 2 * F.abs(U))
+
+        if isinstance(location, F.NDArray):
+            samples = F.broadcast_minus(location, F.broadcast_mul(b_sgn_U, ln_1_2_U))
+        else:
+            samples = location - F.broadcast_mul(b_sgn_U, ln_1_2_U)
+
+        return samples

mxfusion/components/distributions/uniform.py

@@ -0,0 +1,109 @@
+from ...common.config import get_default_MXNet_mode
+from ..variables import Variable
+from .univariate import UnivariateDistribution, UnivariateLogPDFDecorator, UnivariateDrawSamplesDecorator
+
+
+class Uniform(UnivariateDistribution):
+    """
+    The one-dimensional Uniform distribution over the half-open interval [low, high) (includes low, but excludes high).
+    The Uniform distribution can be defined over a scalar random variable
+    or an array of random variables. In case of an array of random variables, the low and high are broadcasted
+    to the shape of the output random variable (array).
+
+    :param low: Low boundary of the Uniform distribution.
+    :type low: Variable
+    :param high: High boundary of the Uniform distribution.
+    :type high: Variable
+    :param rand_gen: the random generator (default: MXNetRandomGenerator).
+    :type rand_gen: RandomGenerator
+    :param dtype: the data type for float point numbers.
+    :type dtype: numpy.float32 or numpy.float64
+    :param ctx: the mxnet context (default: None/current context).
+    :type ctx: None or mxnet.cpu or mxnet.gpu
+    """
+    def __init__(self, low, high, rand_gen=None, dtype=None, ctx=None):
+        if not isinstance(low, Variable):
+            low = Variable(value=low)
+        if not isinstance(high, Variable):
+            high = Variable(value=high)
+
+        inputs = [('low', low), ('high', high)]
+        input_names = [k for k, _ in inputs]
+        output_names = ['random_variable']
+        super(Uniform, self).__init__(inputs=inputs, outputs=None,
+                                      input_names=input_names,
+                                      output_names=output_names,
+                                      rand_gen=rand_gen, dtype=dtype, ctx=ctx)
+
+    @UnivariateLogPDFDecorator()
+    def log_pdf(self, low, high, random_variable, F=None):
+        """
+        Computes the logarithm of the probability density function (PDF) of the Uniform distribution.
+
+        :param low: the low boundary of the Uniform distribution.
+        :type low: MXNet NDArray or MXNet Symbol
+        :param high: the high boundary of the Uniform distributions.
+        :type high: MXNet NDArray or MXNet Symbol
+        :param random_variable: the random variable of the Uniform distribution.
+        :type random_variable: MXNet NDArray or MXNet Symbol
+        :param F: the MXNet computation mode (mxnet.symbol or mxnet.ndarray).
+        :returns: log pdf of the distribution.
+        :rtypes: MXNet NDArray or MXNet Symbol
+        """
+        F = get_default_MXNet_mode() if F is None else F
+
+        # next 3 lines are the broadcasting equivalent of clip(random_variable, low, high)
+        lower_safe = (low - random_variable) <= 0
+        upper_safe = (high - random_variable) > 0
+        in_bounds = F.broadcast_mul(lower_safe, upper_safe)
+        log_likelihood = F.where(
+            in_bounds,
+            -F.log(F.broadcast_minus(high, low)),
+            F.log(F.zeros_like(random_variable))) * self.log_pdf_scaling
+        return log_likelihood
+
+    @UnivariateDrawSamplesDecorator()
+    def draw_samples(self, low, high, rv_shape, num_samples=1, F=None):
+        """
+        Draw samples from the Uniform distribution over the half-open interval [low, high)
+        (includes low, but excludes high).
+
+        :param low: the low boundary of the Uniform distribution.
+        :type low: MXNet NDArray or MXNet Symbol
+        :param high: the high boundary of the Uniform distributions.
+        :type high: MXNet NDArray or MXNet Symbol
+        :param rv_shape: the shape of each sample.
+        :type rv_shape: tuple
+        :param num_samples: the number of drawn samples (default: one).
+        :int num_samples: int
+        :param F: the MXNet computation mode (mxnet.symbol or mxnet.ndarray).
+        :returns: a set samples of the Uniform distribution.
+        :rtypes: MXNet NDArray or MXNet Symbol
+        """
+        F = get_default_MXNet_mode() if F is None else F
+        out_shape = (num_samples,) + rv_shape
+        return self._rand_gen.sample_uniform(low=low, high=high,
+                                             shape=out_shape, dtype=self.dtype, ctx=self.ctx, F=F)
+
+    @staticmethod
+    def define_variable(low=0, high=1, shape=None, rand_gen=None,
+                        dtype=None, ctx=None):
+        """
+        Creates and returns a random variable drawn from a Uniform distribution.
+
+        :param low: Low boundary of the distribution.
+        :param high: High boundary of the distribution.
+        :param shape: the shape of the random variable(s).
+        :type shape: tuple or [tuple]
+        :param rand_gen: the random generator (default: MXNetRandomGenerator).
+        :type rand_gen: RandomGenerator
+        :param dtype: the data type for float point numbers.
+        :type dtype: numpy.float32 or numpy.float64
+        :param ctx: the mxnet context (default: None/current context).
+        :type ctx: None or mxnet.cpu or mxnet.gpu
+        :returns: the random variables drawn from the Uniform distribution.
+        :rtypes: Variable
+        """
+        var = Uniform(low=low, high=high, rand_gen=rand_gen, dtype=dtype, ctx=ctx)
+        var._generate_outputs(shape=shape)
+        return var.random_variable