Merge pull request #165 from Jammy2211/feature/jax_and_numba

Feature/jax and numba

Author: Jammy2211

autoarray/dataset/imaging/dataset.py

@@ -162,7 +162,7 @@ def __init__(
 
         if psf is not None and use_normalized_psf:
             psf = Kernel2D.no_mask(
-                values=psf.native, pixel_scales=psf.pixel_scales, normalize=True
+                values=psf.native._array, pixel_scales=psf.pixel_scales, normalize=True
             )
 
         self.psf = psf
@@ -193,7 +193,7 @@ def convolver(self):
             The convolver given the masked imaging data's mask and PSF.
         """
 
-        return Convolver(mask=self.mask, kernel=self.psf)
+        return Convolver(mask=self.mask, kernel=Kernel2D(values=self.psf._array, mask=self.psf.mask, header=self.psf.header))
 
     @cached_property
     def w_tilde(self):

autoarray/fit/fit_util.py

@@ -1,6 +1,6 @@
 from functools import wraps
+import jax.numpy as np
 
-from autoarray.numpy_wrapper import np
 from autoarray.mask.abstract_mask import Mask
 
 from autoarray import type as ty
@@ -83,7 +83,7 @@ def chi_squared_from(*, chi_squared_map: ty.DataLike) -> float:
     chi_squared_map
         The chi-squared-map of values of the model-data fit to the dataset.
     """
-    return np.sum(chi_squared_map)
+    return np.sum(chi_squared_map._array)
 
 
 def noise_normalization_from(*, noise_map: ty.DataLike) -> float:
@@ -97,7 +97,7 @@ def noise_normalization_from(*, noise_map: ty.DataLike) -> float:
     noise_map
         The masked noise-map of the dataset.
     """
-    return np.sum(np.log(2 * np.pi * noise_map**2.0))
+    return np.sum(np.log(2 * np.pi * noise_map._array**2.0))
 
 
 def normalized_residual_map_complex_from(

autoarray/geometry/geometry_2d.py

@@ -184,8 +184,9 @@ def scaled_coordinates_2d_from(
         -------
         A 2D (y,x) pixel-value coordinate.
         """
+
         return geometry_util.scaled_coordinates_2d_from(
-            pixel_coordinates_2d=pixel_coordinates_2d,
+            pixel_coordinates_2d=np.array(pixel_coordinates_2d),
             shape_native=self.shape_native,
             pixel_scales=self.pixel_scales,
             origins=self.origin,

autoarray/geometry/geometry_util.py

@@ -1,5 +1,7 @@
+import jax.numpy as jnp
+import numpy as np
 from typing import Tuple, Union
-from autoarray.numpy_wrapper import np, use_jax
+
 
 from autoarray import numba_util
 from autoarray import type as ty
@@ -179,8 +181,69 @@ def convert_pixel_scales_2d(pixel_scales: ty.PixelScales) -> Tuple[float, float]
 
     return pixel_scales
 
+@numba_util.jit()
+def central_pixel_coordinates_2d_numba_from(
+    shape_native: Tuple[int, int],
+) -> Tuple[float, float]:
+    """
+    Returns the central pixel coordinates of a 2D geometry (and therefore a 2D data structure like an ``Array2D``)
+    from the shape of that data structure.
+
+    Examples of the central pixels are as follows:
+
+    - For a 3x3 image, the central pixel is pixel [1, 1].
+    - For a 4x4 image, the central pixel is [1.5, 1.5].
+
+    Parameters
+    ----------
+    shape_native
+        The dimensions of the data structure, which can be in 1D, 2D or higher dimensions.
+
+    Returns
+    -------
+    The central pixel coordinates of the data structure.
+    """
+    return (float(shape_native[0] - 1) / 2, float(shape_native[1] - 1) / 2)
 
 @numba_util.jit()
+def central_scaled_coordinate_2d_numba_from(
+    shape_native: Tuple[int, int],
+    pixel_scales: ty.PixelScales,
+    origin: Tuple[float, float] = (0.0, 0.0),
+) -> Tuple[float, float]:
+    """
+    Returns the central scaled coordinates of a 2D geometry (and therefore a 2D data structure like an ``Array2D``)
+    from the shape of that data structure.
+
+    This is computed by using the data structure's shape and converting it to scaled units using an input
+    pixel-coordinates to scaled-coordinate conversion factor `pixel_scales`.
+
+    The origin of the scaled grid can also be input and moved from (0.0, 0.0).
+
+    Parameters
+    ----------
+    shape_native
+        The 2D shape of the data structure whose central scaled coordinates are computed.
+    pixel_scales
+        The (y,x) scaled units to pixel units conversion factor of the 2D data structure.
+    origin
+        The (y,x) scaled units origin of the coordinate system the central scaled coordinate is computed on.
+
+    Returns
+    -------
+    The central coordinates of the 2D data structure in scaled units.
+    """
+
+    central_pixel_coordinates = central_pixel_coordinates_2d_numba_from(
+        shape_native=shape_native
+    )
+
+    y_pixel = central_pixel_coordinates[0] + (origin[0] / pixel_scales[0])
+    x_pixel = central_pixel_coordinates[1] - (origin[1] / pixel_scales[1])
+
+    return (y_pixel, x_pixel)
+
+
 def central_pixel_coordinates_2d_from(
     shape_native: Tuple[int, int],
 ) -> Tuple[float, float]:
@@ -205,7 +268,6 @@ def central_pixel_coordinates_2d_from(
     return (float(shape_native[0] - 1) / 2, float(shape_native[1] - 1) / 2)
 
 
-@numba_util.jit()
 def central_scaled_coordinate_2d_from(
     shape_native: Tuple[int, int],
     pixel_scales: ty.PixelScales,
@@ -234,7 +296,7 @@ def central_scaled_coordinate_2d_from(
     The central coordinates of the 2D data structure in scaled units.
     """
 
-    central_pixel_coordinates = central_pixel_coordinates_2d_from(
+    central_pixel_coordinates = central_pixel_coordinates_2d_numba_from(
         shape_native=shape_native
     )
 
@@ -243,8 +305,6 @@ def central_scaled_coordinate_2d_from(
 
     return (y_pixel, x_pixel)
 
-
-@numba_util.jit()
 def pixel_coordinates_2d_from(
     scaled_coordinates_2d: Tuple[float, float],
     shape_native: Tuple[int, int],
@@ -352,7 +412,7 @@ def scaled_coordinates_2d_from(
         origin=(0.0, 0.0)
     )
     """
-    central_scaled_coordinates = central_scaled_coordinate_2d_from(
+    central_scaled_coordinates = central_scaled_coordinate_2d_numba_from(
         shape_native=shape_native, pixel_scales=pixel_scales, origin=origins
     )
 
@@ -382,18 +442,16 @@ def transform_grid_2d_to_reference_frame(
     grid
         The 2d grid of (y, x) coordinates which are transformed to a new reference frame.
     """
-    if use_jax:
-        shifted_grid_2d = grid_2d.array - np.array(centre)
-    else:
-        shifted_grid_2d = grid_2d - np.array(centre)
-    radius = np.sqrt(np.sum(shifted_grid_2d**2.0, axis=1))
-    theta_coordinate_to_profile = np.arctan2(
+    shifted_grid_2d = np.array(grid_2d) - jnp.array(centre)
+
+    radius = jnp.sqrt(jnp.sum(shifted_grid_2d**2.0, axis=1))
+    theta_coordinate_to_profile = jnp.arctan2(
         shifted_grid_2d[:, 0], shifted_grid_2d[:, 1]
-    ) - np.radians(angle)
-    return np.vstack(
+    ) - jnp.radians(angle)
+    return jnp.vstack(
         [
-            radius * np.sin(theta_coordinate_to_profile),
-            radius * np.cos(theta_coordinate_to_profile),
+            radius * jnp.sin(theta_coordinate_to_profile),
+            radius * jnp.cos(theta_coordinate_to_profile),
         ]
     ).T
 
@@ -435,7 +493,6 @@ def transform_grid_2d_from_reference_frame(
     return np.vstack((y, x)).T
 
 
-@numba_util.jit()
 def grid_pixels_2d_slim_from(
     grid_scaled_2d_slim: np.ndarray,
     shape_native: Tuple[int, int],
@@ -476,33 +533,15 @@ def grid_pixels_2d_slim_from(
     grid_pixels_2d_slim = grid_scaled_2d_slim_from(grid_scaled_2d_slim=grid_scaled_2d_slim, shape=(2,2),
                                                            pixel_scales=(0.5, 0.5), origin=(0.0, 0.0))
     """
-
     centres_scaled = central_scaled_coordinate_2d_from(
         shape_native=shape_native, pixel_scales=pixel_scales, origin=origin
     )
-    if use_jax:
-        centres_scaled = np.array(centres_scaled)
-        pixel_scales = np.array(pixel_scales)
-        sign = np.array([-1, 1])
-        return (sign * grid_scaled_2d_slim / pixel_scales) + centres_scaled + 0.5
-    else:
-        grid_pixels_2d_slim = np.zeros((grid_scaled_2d_slim.shape[0], 2))
-        for slim_index in range(grid_scaled_2d_slim.shape[0]):
-            grid_pixels_2d_slim[slim_index, 0] = (
-                (-grid_scaled_2d_slim[slim_index, 0] / pixel_scales[0])
-                + centres_scaled[0]
-                + 0.5
-            )
-            grid_pixels_2d_slim[slim_index, 1] = (
-                (grid_scaled_2d_slim[slim_index, 1] / pixel_scales[1])
-                + centres_scaled[1]
-                + 0.5
-            )
-
-    return grid_pixels_2d_slim
+    centres_scaled = np.array(centres_scaled)
+    pixel_scales = np.array(pixel_scales)
+    sign = np.array([-1, 1])
+    return (sign * grid_scaled_2d_slim / pixel_scales) + centres_scaled + 0.5
 
 
-@numba_util.jit()
 def grid_pixel_centres_2d_slim_from(
     grid_scaled_2d_slim: np.ndarray,
     shape_native: Tuple[int, int],
@@ -547,32 +586,14 @@ def grid_pixel_centres_2d_slim_from(
         shape_native=shape_native, pixel_scales=pixel_scales, origin=origin
     )
 
-    if use_jax:
-        centres_scaled = np.array(centres_scaled)
-        pixel_scales = np.array(pixel_scales)
-        sign = np.array([-1.0, 1.0])
-        grid_pixels_2d_slim = (
-            (sign * grid_scaled_2d_slim / pixel_scales) + centres_scaled + 0.5
-        ).astype(int)
-    else:
-        grid_pixels_2d_slim = np.zeros((grid_scaled_2d_slim.shape[0], 2))
-
-        for slim_index in range(grid_scaled_2d_slim.shape[0]):
-            grid_pixels_2d_slim[slim_index, 0] = int(
-                (-grid_scaled_2d_slim[slim_index, 0] / pixel_scales[0])
-                + centres_scaled[0]
-                + 0.5
-            )
-            grid_pixels_2d_slim[slim_index, 1] = int(
-                (grid_scaled_2d_slim[slim_index, 1] / pixel_scales[1])
-                + centres_scaled[1]
-                + 0.5
-            )
-
-    return grid_pixels_2d_slim
+    centres_scaled = np.array(centres_scaled)
+    pixel_scales = np.array(pixel_scales)
+    sign = np.array([-1.0, 1.0])
+    return (
+        (sign * grid_scaled_2d_slim / pixel_scales) + centres_scaled + 0.5
+    ).astype(int)
 
 
-@numba_util.jit()
 def grid_pixel_indexes_2d_slim_from(
     grid_scaled_2d_slim: np.ndarray,
     shape_native: Tuple[int, int],
@@ -625,25 +646,13 @@ def grid_pixel_indexes_2d_slim_from(
         origin=origin,
     )
 
-    if use_jax:
-        grid_pixel_indexes_2d_slim = (
-            (grid_pixels_2d_slim * np.array([shape_native[1], 1]))
-            .sum(axis=1)
-            .astype(int)
-        )
-    else:
-        grid_pixel_indexes_2d_slim = np.zeros(grid_pixels_2d_slim.shape[0])
-
-        for slim_index in range(grid_pixels_2d_slim.shape[0]):
-            grid_pixel_indexes_2d_slim[slim_index] = int(
-                grid_pixels_2d_slim[slim_index, 0] * shape_native[1]
-                + grid_pixels_2d_slim[slim_index, 1]
-            )
-
-    return grid_pixel_indexes_2d_slim
+    return (
+        (grid_pixels_2d_slim * np.array([shape_native[1], 1]))
+        .sum(axis=1)
+        .astype(int)
+    )
 
 
-@numba_util.jit()
 def grid_scaled_2d_slim_from(
     grid_pixels_2d_slim: np.ndarray,
     shape_native: Tuple[int, int],
@@ -682,33 +691,18 @@ def grid_scaled_2d_slim_from(
     grid_pixels_2d_slim = grid_scaled_2d_slim_from(grid_pixels_2d_slim=grid_pixels_2d_slim, shape=(2,2),
                                                            pixel_scales=(0.5, 0.5), origin=(0.0, 0.0))
     """
-
     centres_scaled = central_scaled_coordinate_2d_from(
         shape_native=shape_native, pixel_scales=pixel_scales, origin=origin
     )
-    if use_jax:
-        centres_scaled = np.array(centres_scaled)
-        pixel_scales = np.array(pixel_scales)
-        sign = np.array([-1, 1])
-        grid_scaled_2d_slim = (
-            (grid_pixels_2d_slim - centres_scaled - 0.5) * pixel_scales * sign
-        )
-    else:
-        grid_scaled_2d_slim = np.zeros((grid_pixels_2d_slim.shape[0], 2))
-
-        for slim_index in range(grid_scaled_2d_slim.shape[0]):
-            grid_scaled_2d_slim[slim_index, 0] = (
-                -(grid_pixels_2d_slim[slim_index, 0] - centres_scaled[0] - 0.5)
-                * pixel_scales[0]
-            )
-            grid_scaled_2d_slim[slim_index, 1] = (
-                grid_pixels_2d_slim[slim_index, 1] - centres_scaled[1] - 0.5
-            ) * pixel_scales[1]
-
-    return grid_scaled_2d_slim
+
+    centres_scaled = np.array(centres_scaled)
+    pixel_scales = np.array(pixel_scales)
+    sign = np.array([-1, 1])
+    return (
+        (grid_pixels_2d_slim - centres_scaled - 0.5) * pixel_scales * sign
+    )
 
 
-@numba_util.jit()
 def grid_pixel_centres_2d_from(
     grid_scaled_2d: np.ndarray,
     shape_native: Tuple[int, int],
@@ -753,30 +747,12 @@ def grid_pixel_centres_2d_from(
         shape_native=shape_native, pixel_scales=pixel_scales, origin=origin
     )
 
-    if use_jax:
-        centres_scaled = np.array(centres_scaled)
-        pixel_scales = np.array(pixel_scales)
-        sign = np.array([-1.0, 1.0])
-        grid_pixels_2d = (
-            (sign * grid_scaled_2d / pixel_scales) + centres_scaled + 0.5
-        ).astype(int)
-    else:
-        grid_pixels_2d = np.zeros((grid_scaled_2d.shape[0], grid_scaled_2d.shape[1], 2))
-
-        for y in range(grid_scaled_2d.shape[0]):
-            for x in range(grid_scaled_2d.shape[1]):
-                grid_pixels_2d[y, x, 0] = int(
-                    (-grid_scaled_2d[y, x, 0] / pixel_scales[0])
-                    + centres_scaled[0]
-                    + 0.5
-                )
-                grid_pixels_2d[y, x, 1] = int(
-                    (grid_scaled_2d[y, x, 1] / pixel_scales[1])
-                    + centres_scaled[1]
-                    + 0.5
-                )
-
-    return grid_pixels_2d
+    centres_scaled = np.array(centres_scaled)
+    pixel_scales = np.array(pixel_scales)
+    sign = np.array([-1.0, 1.0])
+    return (
+        (sign * grid_scaled_2d / pixel_scales) + centres_scaled + 0.5
+    ).astype(int)
 
 
 def extent_symmetric_from(