feat: Dask get_fccd_images

.gitignore

@@ -29,6 +29,7 @@ var/
 *.egg
 *.eggs
 doc/_build
+venv/
 
 # PyInstaller
 #  Usually these files are written by a python script from a template

csxtools/fastccd/dask.py

@@ -0,0 +1,53 @@
+from typing import Tuple
+
+import numpy as np
+from dask.array import Array as DaskArray
+from numpy.typing import ArrayLike
+
+GAIN_8 = 0x0000
+GAIN_2 = 0x8000
+GAIN_1 = 0xC000
+BAD_PIXEL = 0x2000
+PIXEL_MASK = 0x1FFF
+
+
+def correct_images(images: DaskArray, dark: ArrayLike, flat: ArrayLike, gain: Tuple[float, float, float]):
+    """Apply intensity corrections to a stack of images.
+
+    Parameters
+    ----------
+    images : DaskArray
+        Input array of images to correct; has shape (N, y, x)  where N is the
+        number of images and x and y are the image size.
+    dark : ArrayLike
+        Input array of dark images. This should be of shape (3, y, x).
+        dark[0] is the GAIN_8 (most sensitive setting) dark image with
+        dark[2] being the GAIN_1 (least sensitive) dark image.
+    flat : ArrayLike
+        Input array for the flatfield correction. This should be of shape
+        (y, x)
+    gain : Tuple[float, float, float]
+        These are the gain multiplication factors for the three different
+        gain settings
+
+    // Note GAIN_1 is the least sensitive setting which means we need to multiply the
+    // measured values by 8. Conversly GAIN_8 is the most sensitive and therefore
+    // does not need a multiplier
+    """
+
+    # Shape checking:
+    if dark.ndim != 3:
+        raise ValueError(f"Expected 3D array, got {dark.ndim}D array for darks")
+    if dark.shape[0] != 3:
+        raise ValueError(f"Expected 3 dark images, got {dark.shape[0]}")
+    if dark.shape[-2:] != images.shape[-2]:
+        raise ValueError(f"Dark images shape {dark.shape[-2:]} does not match images shape {images.shape[-2]}")
+    if flat.shape != images.shape[-2:]:
+        raise ValueError(f"Flatfield shape {flat.shape} does not match images shape {images.shape[-2]}")
+
+    corrected = np.where(images & BAD_PIXEL, np.NaN, images)
+    corrected = np.where(images & GAIN_1, flat * gain[-1] * (corrected - dark[-1, ...]), corrected)
+    corrected = np.where(images & GAIN_2, flat * gain[-2] * (corrected - dark[-2, ...]), corrected)
+    corrected = np.where(images & GAIN_8, flat * gain[-3] * (corrected - dark[-3, ...]), corrected)
+
+    return corrected

csxtools/fastccd/images.py

@@ -1,12 +1,15 @@
+import logging
+import time as ttime
+
 import numpy as np
+
 from ..ext import fastccd
-import time as ttime
+from .dask import correct_images as dask_correct_images
 
-import logging
 logger = logging.getLogger(__name__)
 
 
-def correct_images(images, dark=None, flat=None, gain=(1, 4, 8)):
+def correct_images(images, dark=None, flat=None, gain=(1, 4, 8), *, dask=False):
     """Subtract backgrond and gain correct images
 
     This routine subtrtacts the backgrond and corrects the images
@@ -27,6 +30,11 @@ def correct_images(images, dark=None, flat=None, gain=(1, 4, 8)):
     gain : tuple, optional
         These are the gain multiplication factors for the three different
         gain settings
+    dask : bool, optional
+        Do computation in dask instead of in C extension over full array.
+        This returns a DaskArray or DaskArrayClient with pending execution instead of a numpy array.
+        You can use the .compute() method to get the numpy array.
+        Default is False.
 
     Returns
     -------
@@ -49,8 +57,10 @@ def correct_images(images, dark=None, flat=None, gain=(1, 4, 8)):
     else:
         flat = np.asarray(flat, dtype=np.float32)
 
-    data = fastccd.correct_images(images.astype(np.uint16),
-                                  dark, flat, gain)
+    if dask:
+        data = dask_correct_images(images.astype(np.uint16), dark, flat, gain)
+    else:
+        data = fastccd.correct_images(images.astype(np.uint16), dark, flat, gain)
     t = ttime.time() - t
 
     logger.info("Corrected image stack in %.3f seconds", t)

csxtools/image/dask.py

@@ -0,0 +1,35 @@
+import dask.array as da
+from dask.array import Array as DaskArray
+
+
+def rotate90(images: DaskArray, sense: bool) -> DaskArray:
+    """
+    Rotate images by 90 degrees using Dask.
+    This whole function is a moot wrapper around `da.rot90` from Dask, but written
+    explicitly to match the old C code.
+
+    Parameters
+    ----------
+    images : da.Array
+        Input Dask array of images to rotate of shape (N, y, x),
+        where N is the number of images and y, x are the image dimensions.
+    sense : bool
+        False to rotate clockwise, True to rotate anticlockwise.
+
+    Returns
+    -------
+    da.Array
+        The rotated images as a Dask array.
+    """
+    # Rotate images. The axes (1, 2) specify the plane of rotation (y-x plane for each image).
+    # k controls the direction and repetitions of the rotation.
+    if sense != 0:
+        # Counter-clockwise
+        k = 1
+    elif sense == 0:
+        # Clockwise
+        k = -1
+    else:
+        raise ValueError(f"{sense = } is not a valid integer")
+    rotated_images = da.rot90(images, k=k, axes=(-2, -1))
+    return rotated_images

csxtools/image/transform.py

@@ -1,7 +1,8 @@
 from ..ext import image as extimage
+from .dask import rotate90 as dask_rotate90
 
 
-def rotate90(a, sense='ccw'):
+def rotate90(a, sense="ccw", *, dask=False):
     """Rotate a stack of images by 90 degrees
 
     This routine rotates a stack of images by 90. The rotation is performed
@@ -14,6 +15,11 @@ def rotate90(a, sense='ccw'):
             Input array to be rotated. This should be of shape (N, y, x).
         sense : string
             'cw' to rotate clockwise, 'ccw' to rotate anitclockwise
+        dask : bool, optional
+            Do computation in dask instead of in C extension over full array.
+            This returns a DaskArray or DaskArrayClient with pending execution instead of a numpy array.
+            You can use the .compute() method to get the numpy array.
+            Default is False.
 
     Returns
     -------
@@ -22,11 +28,14 @@ def rotate90(a, sense='ccw'):
 
     """
 
-    if sense == 'ccw':
+    if sense == "ccw":
         sense = 1
-    elif sense == 'cw':
+    elif sense == "cw":
         sense = 0
     else:
         raise ValueError("sense must be 'cw' or 'ccw'")
 
-    return extimage.rotate90(a, sense)
+    if dask:
+        return dask_rotate90(a, sense)
+    else:
+        return extimage.rotate90(a, sense)

csxtools/utils.py

@@ -1,17 +1,17 @@
-import numpy as np
+import logging
 import time as ttime
 
+import numpy as np
+from databroker.assets.handlers import AreaDetectorHDF5TimestampHandler
+
 from .fastccd import correct_images
 from .image import rotate90, stackmean
 from .settings import detectors
-from databroker.assets.handlers import AreaDetectorHDF5TimestampHandler
 
-import logging
 logger = logging.getLogger(__name__)
 
 
-def get_fastccd_images(light_header, dark_headers=None,
-                       flat=None, gain=(1, 4, 8), tag=None, roi=None):
+def get_fastccd_images(light_header, dark_headers=None, flat=None, gain=(1, 4, 8), tag=None, roi=None, *, dask=False):
     """Retreive and correct FastCCD Images from associated headers
 
     Retrieve FastCCD Images from databroker and correct for:
@@ -50,14 +50,17 @@ def get_fastccd_images(light_header, dark_headers=None,
         coordinates of the upper-left corner and width and height of
         the ROI: e.g., (x, y, w, h)
 
+    dask : bool, optional
+        Use dask for computation. Default is False.
+
     Returns
     -------
     dask.array : corrected images
 
     """
 
     if tag is None:
-        tag = detectors['fccd']
+        tag = detectors["fccd"]
 
     # Now lets sort out the ROI
     if roi is not None:
@@ -72,8 +75,7 @@ def get_fastccd_images(light_header, dark_headers=None,
         logger.warning("Processing without dark images")
     else:
         if dark_headers[0] is None:
-            raise NotImplementedError("Use of header metadata to find dark"
-                                      " images is not implemented yet.")
+            raise NotImplementedError("Use of header metadata to find dark" " images is not implemented yet.")
 
         # Read the images for the dark headers
         t = ttime.time()
@@ -91,25 +93,20 @@ def get_fastccd_images(light_header, dark_headers=None,
 
                 tt = ttime.time()
                 b = bgnd_events.astype(dtype=np.uint16)
-                logger.info("Image conversion took %.3f seconds",
-                            ttime.time() - tt)
+                logger.info("Image conversion took %.3f seconds", ttime.time() - tt)
 
                 b = correct_images(b, gain=(1, 1, 1))
                 tt = ttime.time()
                 b = stackmean(b)
-                logger.info("Mean of image stack took %.3f seconds",
-                            ttime.time() - tt)
+                logger.info("Mean of image stack took %.3f seconds", ttime.time() - tt)
 
             else:
-                if (i == 0):
-                    logger.warning("Missing dark image"
-                                   " for gain setting 8")
-                elif (i == 1):
-                    logger.warning("Missing dark image"
-                                   " for gain setting 2")
-                elif (i == 2):
-                    logger.warning("Missing dark image"
-                                   " for gain setting 1")
+                if i == 0:
+                    logger.warning("Missing dark image" " for gain setting 8")
+                elif i == 1:
+                    logger.warning("Missing dark image" " for gain setting 2")
+                elif i == 2:
+                    logger.warning("Missing dark image" " for gain setting 1")
 
             dark.append(b)
 
@@ -125,7 +122,7 @@ def get_fastccd_images(light_header, dark_headers=None,
     if flat is not None and roi is not None:
         flat = _crop(flat, roi)
 
-    return _correct_fccd_images(events, bgnd, flat, gain)
+    return _correct_fccd_images(events, bgnd, flat, gain, dask=dask)
 
 
 def get_images_to_4D(images, dtype=None):
@@ -147,8 +144,7 @@ def get_images_to_4D(images, dtype=None):
     >>> a = get_images_to_4D(images, dtype=np.float32)
 
     """
-    im = np.array([np.asarray(im, dtype=dtype) for im in images],
-                  dtype=dtype)
+    im = np.array([np.asarray(im, dtype=dtype) for im in images], dtype=dtype)
     return im
 
 
@@ -183,9 +179,9 @@ def _get_images(header, tag, roi=None):
     return images
 
 
-def _correct_fccd_images(image, bgnd, flat, gain):
-    image = correct_images(image, bgnd, flat, gain)
-    image = rotate90(image, 'cw')
+def _correct_fccd_images(image, bgnd, flat, gain, *, dask=False):
+    image = correct_images(image, bgnd, flat, gain, dask=dask)
+    image = rotate90(image, "cw", dask=dask)
     return image
 
 
@@ -196,11 +192,11 @@ def _crop_images(image, roi):
 def _crop(image, roi):
     image_shape = image.shape
     # Assuming ROI is specified in the "rotated" (correct) orientation
-    roi = [image_shape[-2]-roi[3], roi[0], image_shape[-1]-roi[1], roi[2]]
-    return image.T[roi[1]:roi[3], roi[0]:roi[2]].T
+    roi = [image_shape[-2] - roi[3], roi[0], image_shape[-1] - roi[1], roi[2]]
+    return image.T[roi[1] : roi[3], roi[0] : roi[2]].T
 
 
-def get_fastccd_timestamps(header, tag='fccd_image'):
+def get_fastccd_timestamps(header, tag="fccd_image"):
     """Return the FastCCD timestamps from the Areadetector Data File
 
     Return a list of numpy arrays of the timestamps for the images as
@@ -218,8 +214,7 @@ def get_fastccd_timestamps(header, tag='fccd_image'):
         list of arrays of the timestamps
 
     """
-    with header.db.reg.handler_context(
-            {'AD_HDF5': AreaDetectorHDF5TimestampHandler}):
+    with header.db.reg.handler_context({"AD_HDF5": AreaDetectorHDF5TimestampHandler}):
         timestamps = list(header.data(tag))
 
     return timestamps
@@ -259,9 +254,8 @@ def calculate_flatfield(image, limits=(0.6, 1.4)):
     return flat
 
 
-
 def get_fastccd_flatfield(light, dark, flat=None, limits=(0.6, 1.4), half_interval=False):
-    """Calculate a flatfield from two headers 
+    """Calculate a flatfield from two headers
 
     This routine calculates the flatfield using the
     :func:calculate_flatfield() function after obtaining the images from
@@ -278,7 +272,7 @@ def get_fastccd_flatfield(light, dark, flat=None, limits=(0.6, 1.4), half_interv
     limits : tuple limits used for returning corrected pixel flatfield
         The tuple setting lower and upper bound. np.nan returned value is outside bounds
     half_interval : boolean or tuple to perform calculation for only half of the FastCCD
-        Default is False. If True, then the hard-code portion is retained.  Customize image 
+        Default is False. If True, then the hard-code portion is retained.  Customize image
         manipulation using a tuple of length 2 for (row_start, row_stop).
 
 
@@ -291,16 +285,15 @@ def get_fastccd_flatfield(light, dark, flat=None, limits=(0.6, 1.4), half_interv
     images = stackmean(images)
     if half_interval:
         if isinstance(half_interval, bool):
-            row_start, row_stop = (7, 486) #hard coded for the broken half of the fccd
+            row_start, row_stop = (7, 486)  # hard coded for the broken half of the fccd
         else:
             row_start, row_stop = half_interval
             print(row_start, row_stop)
         images[:, row_start:row_stop] = np.nan
     flat = calculate_flatfield(images, limits)
     removed = np.sum(np.isnan(flat))
     if removed != 0:
-        logger.warning("Flatfield correction removed %d pixels (%.2f %%)" %
-                       (removed, removed * 100 / flat.size))
+        logger.warning("Flatfield correction removed %d pixels (%.2f %%)" % (removed, removed * 100 / flat.size))
     return flat