Update package structure for pypi release

ProxNest/tests/test_optimisations.py

@@ -2,6 +2,7 @@
 import numpy as np
 import ProxNest.utils as utils
 import ProxNest.optimisations as opts
+from ProxNest.operators import sensing_operators as sense_ops
 
 
 @pytest.mark.parametrize("tight", [True, False])
@@ -20,7 +21,8 @@ def test_l2_ball_projection(tight: bool, pos: bool):
     tau = -LogLikeliL(x0) * 1e-1
 
     # Create a parameters structure.
-    params = utils.create_parameters_dict(y=data, tight=tight, pos=pos, reality=True)
+    id = sense_ops.Identity()
+    params = utils.create_parameters_dict(y=data, Phi=id, Psi=id, tight=tight, pos=pos, reality=True)
 
     # Evaluate the projection algorithm
     z = opts.l2_ball_proj.sopt_fast_proj_B2(x0, tau, params)
@@ -34,7 +36,8 @@ def test_l1_norm_projection_extremes(tight: bool, pos: bool):
     x = np.random.randn(64, 64)
 
     # Create a parameters structure.
-    params = utils.create_parameters_dict(tight=tight, pos=pos, reality=True)
+    id = sense_ops.Identity()
+    params = utils.create_parameters_dict(tight=tight, Phi=id, Psi=id, pos=pos, reality=True)
 
     if tight:
         # Evaluate the l1-norm sub-iterations lambda=0
@@ -56,7 +59,8 @@ def test_l1_norm_projection_specific(tight: bool, pos: bool):
     obj_pred = (3 / 8) * len(x.flatten("C"))
 
     # Create a parameters structure.
-    params = utils.create_parameters_dict(tight=tight, pos=pos, reality=True)
+    id = sense_ops.Identity()
+    params = utils.create_parameters_dict(tight=tight, Phi=id, Psi=id, pos=pos, reality=True)
 
     # Evaluate the l1-norm sub-iterations
     z = opts.l1_norm_prox.l1_norm_prox(x, lamb, params)

ProxNest/tests/test_sampling.py

@@ -4,14 +4,13 @@
 import ProxNest.utils as utils
 import ProxNest.sampling as sampling
 import ProxNest.optimisations as optimisations
-import ProxNest.operators as operators
+from ProxNest.operators import sensing_operators as sense_ops
 
 def test_against_analytic_gaussian():
     """ Tests ProxNest against analytic Gaussian """
 
     # A simple identity forward model and redundant dictionary
-    phi = operators.sensing_operators.Identity()
-    psi = operators.sensing_operators.Identity()  
+    id = sense_ops.Identity()
     sigma = 1
     iterations = 20
     delta = 1/2
@@ -22,7 +21,7 @@ def test_against_analytic_gaussian():
     # Parameter dictionary associated with optimisation problem of resampling from the prior subject to the likelihood iso-ball
     params = utils.create_parameters_dict(
               y = image,                # Measurements i.e. data
-            Phi = phi,                  # Forward model
+            Phi = id,                   # Forward model
         epsilon = 1e-3,                 # Radius of L2-ball of likelihood 
           tight = True,                 # Is Phi a tight frame or not?
              nu = 1,                    # Bound on the squared-norm of Phi
@@ -55,16 +54,16 @@ def test_against_analytic_gaussian():
         params["y"] = image
 
         # Lambda functions to evaluate cost function
-        LogLikeliL = lambda sol : - np.linalg.norm(image-phi.dir_op(sol))**2/(2*sigma**2)
+        LogLikeliL = lambda sol : - np.linalg.norm(image-id.dir_op(sol))**2/(2*sigma**2)
 
         # Lambda function for L2-norm identity prior backprojection steps
-        proxH = lambda x, T : x - 2*T*psi.adj_op(psi.dir_op(x))*2*delta
+        proxH = lambda x, T : x - 2*T*id.adj_op(id.dir_op(x))*2*delta
 
         # Lambda function for L2-ball likelihood projection during resampling
         proxB = lambda x, tau: optimisations.l2_ball_proj.sopt_fast_proj_B2(x, tau, params)
 
         # Select a starting position
-        X0 = np.abs(phi.adj_op(image))
+        X0 = np.abs(id.adj_op(image))
 
         # Perform proximal nested sampling
         NS_BayEvi, NS_Trace = sampling.proximal_nested.ProxNestedSampling(X0, LogLikeliL, proxH, proxB, params, options)

ProxNest/utils.py

@@ -1,11 +1,7 @@
-import numpy as np
-from ProxNest.operators import sensing_operators as sense
-
-
 def create_parameters_dict(
     y=0,
-    Phi=sense.Identity(),
-    Psi=sense.Identity(),
+    Phi=None,
+    Psi=None,
     epsilon=1e-3,
     tight=True,
     nu=1,
@@ -23,9 +19,9 @@ def create_parameters_dict(
     Args:
         y (np.ndarray): Measurements (default = 0).
 
-        Phi (linear operator): Sensing operator (default = sense.Identity).
+        Phi (linear operator): Sensing operator (default = None).
 
-        Psi (linear operator): Redundant dictionary (default = sense.Identity).
+        Psi (linear operator): Redundant dictionary (default = None).
 
         epsilon (float): Radius of the :math:`\ell_2` ball (default = 1e-3).
 

setup.py

@@ -52,8 +52,8 @@ def read_file(file):
     author_email="[email protected]",
     license="GNU General Public License v3 (GPLv3)",
     install_requires=required,
-    description="Proximal nested sampling for high-dimensional Bayesian inference",
+    description="Proximal nested sampling for high-dimensional Bayesian model selection",
     long_description_content_type="text/x-rst",
     long_description=long_description,
-    packages=["ProxNest"],
+    packages=["ProxNest", "ProxNest.operators", "ProxNest.optimisations", "ProxNest.sampling"],
 )