make them private

romtools/trial_space/__init__.py

@@ -82,25 +82,22 @@
 
 - `TrialSpaceFromScaledPOD`: POD trial space computed via scaled SVD.
 
-which derive from the abstract class `TrialSpace`. Additionally, we provide two helpers free-functions:
-
-- `tensor_to_matrix`: converts a tensor with shape $[N, M, P]$ to a matrix
-    representation in which the first two dimension are collapsed $[N M, P]$
-
-- `matrix_to_tensor`: inverse operation of `tensor_to_matrix`
+which derive from the abstract class `TrialSpace`.
 
 ---
 ##**API**
 '''
 
 import abc
 import numpy as np
+from romtools.trial_space.utils import tensor_to_matrix, matrix_to_tensor
 from romtools.trial_space.utils.truncater import *
 from romtools.trial_space.utils.shifter import *
 from romtools.trial_space.utils.scaler import *
 from romtools.trial_space.utils.splitter import *
 from romtools.trial_space.utils.orthogonalizer import *
 
+
 class TrialSpace(abc.ABC):
     '''
     Abstract base class for trial space implementations.
@@ -127,7 +124,7 @@ def get_shift_vector(self):
         Retrieves the shift vector of the trial space.
 
         Returns:
-            `np.ndarray`: The shift vector in tensorm form.
+            `np.ndarray`: The shift vector in tensor form.
 
         Concrete subclasses should implement this method to return the shift
         vector specific to their trial space implementation.
@@ -356,23 +353,3 @@ def get_basis(self):
         Concrete implementation of `TrialSpace.get_basis()`
         '''
         return self.__basis
-
-
-def tensor_to_matrix(tensor_input):
-    '''
-    Converts a tensor with shape $[N, M, P]$ to a matrix representation
-    in which the first two dimension are collapsed $[N M, P]$.
-    '''
-    output_tensor = tensor_input.reshape(tensor_input.shape[0]*tensor_input.shape[1],
-                                         tensor_input.shape[2])
-    return output_tensor
-
-
-def matrix_to_tensor(n_var, matrix_input):
-    '''
-    Inverse operation of `tensor_to_matrix`
-    '''
-    d1 = int(matrix_input.shape[0] / n_var)
-    d2 = matrix_input.shape[1]
-    output_matrix = matrix_input.reshape(n_var, d1, d2)
-    return output_matrix

romtools/trial_space/utils/__init__.py

@@ -53,3 +53,24 @@
 from romtools.trial_space.utils.splitter import *
 from romtools.trial_space.utils.truncater import *
 from romtools.trial_space.utils.svd_method_of_snapshots import *
+
+def tensor_to_matrix(tensor_input):
+    '''
+    @private
+    Converts a tensor with shape $[N, M, P]$ to a matrix representation
+    in which the first two dimension are collapsed $[N M, P]$.
+    '''
+    output_tensor = tensor_input.reshape(tensor_input.shape[0]*tensor_input.shape[1],
+                                         tensor_input.shape[2])
+    return output_tensor
+
+
+def matrix_to_tensor(n_var, matrix_input):
+    '''
+    @private
+    Inverse operation of `__tensor_to_matrix`
+    '''
+    d1 = int(matrix_input.shape[0] / n_var)
+    d2 = matrix_input.shape[1]
+    output_matrix = matrix_input.reshape(n_var, d1, d2)
+    return output_matrix

tests/romtools/test_trial_space.py

@@ -14,12 +14,6 @@
 #    assert matrix.shape[0] == 15
 #    assert matrix.shape[1] == 7
 
-
-def tensor_to_matrix(tensor_input):
-    return tensor_input.reshape(tensor_input.shape[0]*tensor_input.shape[1],
-                                tensor_input.shape[2])
-
-
 @pytest.mark.mpi_skip
 def test_dictionary_trial_space():
     snapshots = np.random.normal(size=(3, 8, 6))
@@ -74,7 +68,7 @@ def test_dictionary_trial_space():
                        np.mean(snapshots, axis=2))
     assert np.allclose(my_trial_space.get_dimension(), 12)
     basis = my_trial_space.get_basis()
-    basis = tensor_to_matrix(basis)
+    basis = utils.tensor_to_matrix(basis)
     assert np.allclose(basis.transpose() @ basis, np.eye(12))
 
 
@@ -91,7 +85,7 @@ def test_trial_space_from_pod():
                                           my_splitter,
                                           my_orthogonalizer)
     # truth trial space
-    snapshotMatrix = tensor_to_matrix(snapshots)
+    snapshotMatrix = utils.tensor_to_matrix(snapshots)
     u, s, v = np.linalg.svd(snapshotMatrix, full_matrices=False)
     basis_tensor = my_trial_space.get_basis()
     assert np.allclose(u.reshape(basis_tensor.shape), basis_tensor)
@@ -164,7 +158,7 @@ def test_trial_space_from_scaled_pod():
                                                 my_splitter,
                                                 my_orthogonalizer)
     scaled_snapshots = my_scaler.pre_scaling(snapshots)
-    snapshotMatrix = tensor_to_matrix(scaled_snapshots)
+    snapshotMatrix = utils.tensor_to_matrix(scaled_snapshots)
     u, s, v = np.linalg.svd(snapshotMatrix, full_matrices=False)
     basis_tensor = my_trial_space.get_basis()
     u = u.reshape(basis_tensor.shape)
@@ -189,7 +183,7 @@ def test_trial_space_from_scaled_pod():
     shifted_snapshots, shift_vector = my_shifter(snapshots)
     my_scaler = utils.VariableScaler('max_abs')
     scaled_shifted_snapshots = my_scaler.pre_scaling(shifted_snapshots)
-    snapshot_matrix = tensor_to_matrix(scaled_shifted_snapshots)
+    snapshot_matrix = utils.tensor_to_matrix(scaled_shifted_snapshots)
     u, s, v = np.linalg.svd(snapshot_matrix, full_matrices=False)
     basis_tensor = my_trial_space.get_basis()
     u = u.reshape(basis_tensor.shape)
@@ -214,7 +208,7 @@ def test_trial_space_from_scaled_pod():
     shifted_snapshots, _ = my_shifter(snapshots)
     my_scaler = utils.VariableScaler('max_abs')
     scaled_shifted_snapshots = my_scaler.pre_scaling(shifted_snapshots)
-    snapshot_matrix = tensor_to_matrix(scaled_shifted_snapshots)
+    snapshot_matrix = utils.tensor_to_matrix(scaled_shifted_snapshots)
     snapshot_matrix = my_splitter(snapshot_matrix)
     u, s, v = np.linalg.svd(snapshot_matrix, full_matrices=False)
     basis_tensor = my_trial_space.get_basis()
@@ -241,7 +235,7 @@ def test_trial_space_from_scaled_pod():
     shifted_snapshots, shift_vector = my_shifter(snapshots)
     my_scaler = utils.VariableScaler('max_abs')
     scaled_shifted_snapshots = my_scaler.pre_scaling(shifted_snapshots)
-    snapshot_matrix = tensor_to_matrix(scaled_shifted_snapshots)
+    snapshot_matrix = utils.tensor_to_matrix(scaled_shifted_snapshots)
     snapshot_matrix = my_splitter(snapshot_matrix)
     u, s, v = np.linalg.svd(snapshot_matrix, full_matrices=False)
     ushp = u.shape