Replace generate_splines by a more flexible class

Also allow to calculate the derivative of the radial integral
numerically.

docs/src/references/api/python/misc.rst

@@ -13,5 +13,3 @@ Miscelaneous
 .. autoclass:: rascaline.Profiler
     :members:
     :undoc-members:
-
-.. autofunction:: rascaline.generate_splines

docs/src/references/api/python/utils/index.rst

@@ -8,3 +8,4 @@ Utility functions and classes that extend the core usage of rascaline.
     :maxdepth: 1
 
     power-spectrum
+    splines

docs/src/references/api/python/utils/splines.rst

@@ -0,0 +1 @@
+.. automodule:: rascaline.utils.splines

python/rascaline-torch/rascaline/torch/__init__.py

@@ -37,7 +37,6 @@
 
 __all__ = [
     "AtomicComposition",
-    "CalculatorBase",
     "LodeSphericalExpansion",
     "NeighborList",
     "SoapPowerSpectrum",

python/rascaline/examples/splined-radial-integral.py

@@ -1,4 +1,6 @@
 """
+.. _example-splines:
+
 Splined radial integrals
 ========================
 
@@ -20,7 +22,8 @@
 from scipy.special import jv
 from scipy.special import spherical_jn as j_l
 
-from rascaline import SphericalExpansion, generate_splines
+from rascaline import SphericalExpansion
+from rascaline.utils import RadialIntegralFromFunction
 
 
 # %%
@@ -114,16 +117,16 @@ def laplacian_eigenstate_basis_derivative(n, el, r):
 # %%
 #
 # The radial basis functions and their derivatives can be input into a spline
-# generator. This will output the positions of the spline points, the
+# generator class. This will output the positions of the spline points, the
 # values of the basis functions evaluated at the spline points, and the
 # corresponding derivatives.
-spline_points = generate_splines(
-    laplacian_eigenstate_basis,
-    laplacian_eigenstate_basis_derivative,
-    max_radial,
-    max_angular,
-    cutoff,
-    requested_accuracy=1e-5,
+spliner = RadialIntegralFromFunction(
+    radial_integral=laplacian_eigenstate_basis,
+    radial_integral_derivative=laplacian_eigenstate_basis_derivative,
+    spline_cutoff=cutoff,
+    max_radial=max_radial,
+    max_angular=max_angular,
+    accuracy=1e-5,
 )
 
 # %%
@@ -136,11 +139,7 @@ def laplacian_eigenstate_basis_derivative(n, el, r):
     "max_radial": max_radial,
     "max_angular": max_angular,
     "center_atom_weight": 0.0,
-    "radial_basis": {
-        "TabulatedRadialIntegral": {
-            "points": spline_points,
-        }
-    },
+    "radial_basis": spliner.compute(),
     "atomic_gaussian_width": 1.0,  # ignored
     "cutoff_function": {"Step": {}},
 }

python/rascaline/rascaline/__init__.py

@@ -15,15 +15,13 @@
 )
 from .log import set_logging_callback  # noqa
 from .profiling import Profiler  # noqa
-from .splines import generate_splines  # noqa
 from .status import RascalError  # noqa
 from .systems import IntoSystem, SystemBase  # noqa
 from .version import __version__  # noqa
 
 
 __all__ = [
     "AtomicComposition",
-    "CalculatorBase",
     "LodeSphericalExpansion",
     "NeighborList",
     "SoapPowerSpectrum",

python/rascaline/rascaline/utils/__init__.py

@@ -2,10 +2,8 @@
 
 import metatensor
 
-from .power_spectrum import PowerSpectrum
-
-
-__all__ = ["PowerSpectrum"]
+from .power_spectrum import PowerSpectrum  # noqa
+from .splines import RadialIntegralFromFunction, RadialIntegralSplinerBase  # noqa
 
 
 # path that can be used with cmake to access the rascaline library and headers