[TEST] Cleaning pre-conditioner code and tests

Author: Leguark

gempy_engine/core/data/internal_structs.py

@@ -22,8 +22,8 @@ def __init__(self, sp_internal: SurfacePointsInternals, ori_internal: Orientatio
                  xyz_to_interpolate: np.ndarray=None, fault_internal=None):
         self.sp_internal = sp_internal
         self.ori_internal = ori_internal
-    
-        self.xyz_to_interpolate = xyz_to_interpolate
+        if xyz_to_interpolate is not None:
+            self.xyz_to_interpolate = xyz_to_interpolate.astype(BackendTensor.dtype)
         self._fault_internal = fault_internal
 
     def __hash__(self):

gempy_engine/modules/solver/_numpy_solvers.py

@@ -1,4 +1,6 @@
 import numpy as np
+
+from gempy_engine import optional_dependencies
 from gempy_engine.core.backend_tensor import BackendTensor
 
 bt = BackendTensor
@@ -14,31 +16,30 @@ def pykeops_numpy_cg(b, cov, dtype):
         backend="CPU"
     )
 
-
     return w
 
 
 def numpy_solve(b, cov, dtype):
     w = BackendTensor.tfnp.linalg.solve(cov.astype(dtype), b[:, 0])
     return w
 
-def numpy_cg(b, cov, ):
-    global i
+
+def numpy_cg(b, cov):
     if bt.use_gpu is False and BackendTensor.pykeops_enabled is True:
         cov.backend = 'CPU'
 
     from ._pykeops_solvers.incomplete_cholesky import ichol
     from ._pykeops_solvers.cg import cg
-    from scipy.sparse.linalg import aslinearoperator
-    import scipy.sparse as sps
+    
+    scipy = optional_dependencies.require_scipy()
 
     sparse_cov = cov.copy()
     sparse_cov[np.abs(cov) < 1e-10] = 0
 
-    sparse_cov = sps.csc_matrix(sparse_cov,)
+    sparse_cov = scipy.sparse.csc_matrix(sparse_cov, )
     conditioner = ichol(sparse_cov)
 
-    A = aslinearoperator(cov)
+    A = scipy.sparse.linalg.aslinearoperator(cov)
     print(f'A size: {A.shape}')
 
     w = cg(
@@ -65,7 +66,3 @@ def numpy_gmres(b, cov):
     )
     w = np.atleast_2d(w).T
     return w
-
-
-
-

gempy_engine/modules/solver/_pykeops_solvers/incomplete_cholesky.py

@@ -186,7 +186,7 @@ def __call__(self, b):
 def ichol(
     A,
     discard_threshold=1e-4,
-    shifts=[0.0, 1e-4, 1e-3, 1e-2, 0.1, 0.5, 1.0, 10.0, 100, 1e3, 1e4, 1e5],
+        shifts=None,
     max_nnz=int(4e9 / 16),
     relative_discard_threshold=0.0,
     diag_keep_discarded=True,
@@ -238,7 +238,13 @@ def ichol(
     >>> cholesky_decomposition(np.array([1.0, 2.0]))
     array([-1.,  1.])
     """
+    
+    # Check A is float64 and raise and NotImplementedError if not
+    if A.dtype != np.float64:
+        raise NotImplementedError( "Matrix A must be of dtype np.float64.") 
 
+    if shifts is None:
+        shifts = [0.0, 1e-4, 1e-3, 1e-2, 0.1, 0.5, 1.0, 10.0, 100, 1e3, 1e4, 1e5]
     if isinstance(A, scipy.sparse.csr_matrix):
         A = A.T