python2 compat

Author: slinderman

aux/demo_ll.png

@@ -44,4 +44,4 @@ cdef inline void copy_upper_lower(int n, floating *x) nogil:
     for i in range(n):
         for j in range(i):
             x[n*i+j] = x[n*j+i]
-
+            

aux/demo_smooth.png

@@ -13,7 +13,7 @@ from numpy.math cimport INFINITY, PI
 from scipy.linalg.cython_blas cimport dsymm, dcopy, dgemm, dgemv, daxpy, dsyrk, \
     dtrmv, dger, dnrm2, ddot
 from scipy.linalg.cython_lapack cimport dpotrf, dpotrs, dpotri, dtrtrs
-from util cimport copy_transpose, copy_upper_lower
+from cyutil cimport copy_transpose, copy_upper_lower
 
 
 # TODO instead of specializing last step in info filtering and rts, we could

pylds/util.pxd pylds/cyutil.pxd

@@ -13,7 +13,7 @@ from numpy.math cimport INFINITY, PI
 from scipy.linalg.cython_blas cimport dsymm, dcopy, dgemm, dgemv, daxpy, dsyrk, \
     dtrmv, dger, dnrm2, ddot
 from scipy.linalg.cython_lapack cimport dpotrf, dpotrs, dpotri, dtrtrs
-from util cimport copy_transpose, copy_upper_lower
+from cyutil cimport copy_transpose, copy_upper_lower
 
 # NOTE: because the matrix operations are done in Fortran order but the code
 # expects C ordered arrays as input, the BLAS and LAPACK function calls mark

pylds/lds_info_messages.pyx

@@ -5,21 +5,15 @@
 from pybasicbayes.abstractions import Model, ModelGibbsSampling, \
     ModelEM, ModelMeanField, ModelMeanFieldSVI
 
-from pybasicbayes.distributions import DiagonalRegression
-
+from pybasicbayes.distributions import DiagonalRegression, Gaussian, Regression
+from pylds.distributions import PoissonRegression
 from pylds.states import LDSStates, LDSStatesZeroInflatedCountData, LaplaceApproxPoissonLDSStates
+from pylds.util import random_rotation
 
-# TODO make separate versions for stationary, nonstationary,
-# nonstationary-and-distinct-for-each-sequence
-
-# NOTE: dynamics_distn should probably be an instance of AutoRegression,
-# emission_distn should probably be an instance of Regression, and
+# NOTE: dynamics_distn should be an instance of Regression,
+# emission_distn should be an instance of Regression, and
 # init_dynamics_distn should probably be an instance of Gaussian
 
-######################
-#  algorithm mixins  #
-######################
-
 class _LDSBase(Model):
 
     _states_class = LDSStates
@@ -381,7 +375,7 @@ def sigma_init(self, sigma_init):
 class ZeroInflatedCountLDS(_LDSGibbsSampling, _LDSBase):
     _states_class = LDSStatesZeroInflatedCountData
 
-    def __init__(self, *args, rho=1.0, **kwargs):
+    def __init__(self, rho, *args, **kwargs):
         """
         :param rho: Probability of count drawn from model
                     With pr 1-rho, the emission is deterministically zero
@@ -401,7 +395,6 @@ def _generate_obs(self,s, inputs):
                 x=np.hstack((s.gaussian_states, inputs)), return_xy=False)
 
             # Zero out data
-            from scipy.sparse import csr_matrix
             zeros = np.random.rand(s.T, self.D_obs) > self.rho
             data[zeros] = 0
             s.data = csr_matrix(data)
@@ -509,25 +502,6 @@ def expected_log_likelihood(self):
 ##############################
 
 # TODO make data-dependent default constructors
-
-from pybasicbayes.distributions import Regression
-
-
-def random_rotation(n, theta=None):
-    if theta is None:
-        # Sample a random, slow rotation
-        theta = 0.5 * np.pi * np.random.rand()
-
-    if n == 1:
-        return np.random.rand() * np.eye(1)
-
-    rot = np.array([[np.cos(theta), -np.sin(theta)],
-                    [np.sin(theta), np.cos(theta)]])
-    out = np.zeros((n, n))
-    out[:2, :2] = rot
-    q = np.linalg.qr(np.random.randn(n, n))[0]
-    return q.dot(out).dot(q.T)
-
 def DefaultLDS(D_obs, D_latent, D_input=0,
                mu_init=None, sigma_init=None,
                A=None, B=None, sigma_states=None,
@@ -561,10 +535,6 @@ def DefaultLDS(D_obs, D_latent, D_input=0,
 
     return model
 
-
-from pybasicbayes.distributions import Gaussian
-from pylds.distributions import PoissonRegression
-
 def DefaultPoissonLDS(D_obs, D_latent, D_input=0,
                       mu_init=None, sigma_init=None,
                       A=None, B=None, sigma_states=None,