pytme 0.2.0b0__cp311-cp311-macosx_14_0_arm64.whl → 0.2.2__cp311-cp311-macosx_14_0_arm64.whl

scripts/extract_candidates.py

@@ -1,5 +1,5 @@
 #!python3
-""" CLI to simplify analysing the output of match_template.py.
+""" Prepare orientations stack for refinement.
 
     Copyright (c) 2023 European Molecular Biology Laboratory
 
@@ -12,13 +12,33 @@ import numpy as np
 
 from tme import Density, Orientations
 from tme.matching_utils import (
-    create_mask,
+    load_pickle,
     generate_tempfile_name,
     rotation_aligning_vectors,
     euler_from_rotationmatrix,
+    euler_to_rotationmatrix,
 )
 
 
+class ProgressBar:
+    """
+    ASCII progress bar.
+    """
+
+    def __init__(self, message : str, nchars : int, total : int):
+        self._size = nchars - len(message)
+        self._message = message
+        self._total = total
+
+    def update(self, cur):
+        x = int(cur * self._size / self._total)
+        print(
+            "%s[%s%s] %i/%i\r"
+            % (self._message, "#" * x, "." * (self._size - x), cur, self._total),
+            end=''
+        )
+
+
 def parse_args():
     parser = argparse.ArgumentParser(
         description="Extract matching candidates for further refinement."
@@ -52,57 +72,67 @@ def parse_args():
     io_group.add_argument(
         "-o",
         "--output_file",
-        required=False,
+        required=True,
         type=str,
         help="Path to output HDF5 file.",
     )
 
-    extraction_group = parser.add_argument_group("Extraction")
-    extraction_group.add_argument(
-        "--box_size",
+    alignment_group = parser.add_argument_group("Alignment")
+    alignment_group.add_argument(
+        "--align_orientations",
+        action="store_true",
         required=False,
-        type=int,
-        help="Box size for extraction, defaults to two times the template box.",
+        help="Whether to align extracted orientations based on their angles. Allows "
+        "for efficient subsequent sampling of cone angles.",
     )
-    extraction_group.add_argument(
-        "--alignment_vector",
+    alignment_group.add_argument(
+        "--angles_are_vector",
+        action="store_true",
         required=False,
-        type=str,
-        help="Vector to align extracted coordinate to e.g. '1,0,0'. If provided, euler_z"
-        "euler_y and euler_x are assumed to be a vector that will be rotated to match "
-        "alignment_vector.",
+        help="Considers euler_z euler_y, euler_x as vector that will be rotated to align "
+        "with the z-axis (1,0,0). Only considered when --align_orientations is set."
     )
-    extraction_group.add_argument(
+    alignment_group.add_argument(
         "--interpolation_order",
         dest="interpolation_order",
         required=False,
         type=int,
         default=1,
-        help="Spline interpolation used for template rotations. If less than zero "
-        "no interpolation is performed.",
+        help="Interpolation order for alignment, less than zero is no interpolation."
+    )
+
+    extraction_group = parser.add_argument_group("Extraction")
+    extraction_group.add_argument(
+        "--box_size",
+        required=False,
+        type=int,
+        help="Box size for extraction, defaults to two times the template.",
     )
     extraction_group.add_argument(
         "--translation_uncertainty",
-        dest="translation_uncertainty",
         required=False,
         type=int,
-        default=None,
-        help="Creates a cented spherical target mask with given radius in voxel.",
+        help="Sets box size for extraction to template box plus this value.",
     )
     extraction_group.add_argument(
-        "--mask_path",
-        dest="mask_path",
+        "--keep_out_of_box",
+        action="store_true",
         required=False,
-        type=str,
-        default="target_mask.h5",
-        help="Path to write spherical mask to, defaults to target_mask.h5.",
+        help="Whether to keep orientations that fall outside the box. If the "
+        "orientations are sensible, it is safe to pass this flag.",
     )
+
     args = parser.parse_args()
 
-    if args.alignment_vector is not None:
-        args.alignment_vector = [float(x) for x in args.alignment_vector.split(",")]
-        if len(args.alignment_vector) != 3:
-            raise ValueError("Only three-dimensional vectors are supported")
+    data_present = args.target is not None and args.template is not None
+    if args.input_file is None and not data_present:
+        raise ValueError(
+            "Either --input_file or --target and --template need to be specified."
+        )
+    elif args.input_file is not None and data_present:
+        raise ValueError(
+            "Please specific either --input_file or --target and --template."
+        )
 
     return args
 
@@ -111,57 +141,62 @@ def main():
     args = parse_args()
     orientations = Orientations.from_file(args.orientations)
 
+    if args.input_file is not None:
+        data = load_pickle(args.input_file)
+        target_origin, _, sampling_rate, cli_args = data[-1]
+        args.target, args.template  = cli_args.target, cli_args.template
+
     target = Density.from_file(args.target, use_memmap=True)
-    template = Density.from_file(args.template)
 
-    box_size = template.shape
-    if args.box_size is not None:
-        args.box_size = args.box_size // 2
-        box_size = np.full(template.data.ndim, fill_value=args.box_size, dtype=int)
+    try:
+        template = Density.from_file(args.template)
+    except Exception:
+        template = Density.from_structure(args.template, sampling_rate = target.sampling_rate)
+
+    box_size = args.box_size
+    if box_size is None:
+        box_size = np.multiply(template.shape, 2)
+    if args.translation_uncertainty is not None:
+        box_size = np.add(template.shape, args.translation_uncertainty)
+
+    box_size = np.array(box_size)
+    box_size = np.repeat(box_size, template.data.ndim // box_size.size).astype(int)
+
+    extraction_shape = np.copy(box_size)
+    if args.align_orientations:
+        extraction_shape[:] = int(np.linalg.norm(box_size) + 1)
 
-    subtomo_shape = np.multiply(box_size, 2).astype(int)
-    extraction_scaling = 3 if args.alignment_vector is not None else 2
-    extraction_shape = np.multiply(box_size, extraction_scaling).astype(int)
     orientations, cand_slices, obs_slices = orientations.get_extraction_slices(
         target_shape=target.shape,
         extraction_shape=extraction_shape,
-        drop_out_of_box=True,
+        drop_out_of_box=not args.keep_out_of_box,
         return_orientations=True,
     )
 
+    if args.align_orientations:
+        orientations.rotations = orientations.rotations.astype(np.float32)
+        for index in range(orientations.rotations.shape[0]):
+            rotation_matrix = euler_to_rotationmatrix(orientations.rotations[index])
+            rotation_matrix = np.linalg.inv(rotation_matrix)
+            if args.angles_are_vector:
+                rotation_matrix = rotation_aligning_vectors(
+                    orientations.rotations[index], target_vector=(1,0,0)
+                )
+            orientations.rotations[index] = euler_from_rotationmatrix(rotation_matrix)
+
+
     filename = generate_tempfile_name()
     output_dtype = target.data.dtype
-    if args.alignment_vector is not None:
+    if args.align_orientations is not None:
         output_dtype = np.float32
 
-    if args.translation_uncertainty is not None:
-        dens = Density(
-            np.memmap(
-                args.mask_path,
-                mode="w+",
-                shape=(len(obs_slices), *subtomo_shape),
-                dtype=output_dtype,
-            ),
-            sampling_rate=(1, *target.sampling_rate),
-            origin=(0, *target.origin),
-        )
-        dens.data[:] = 0
-        mask = create_mask(
-            mask_type="ellipse",
-            center=template.shape,
-            radius=args.translation_uncertainty,
-            shape=subtomo_shape,
-        )
-        dens.data[:] = mask
-        dens.to_file(args.mask_path)
-
     target.data = target.data.astype(output_dtype)
 
     dens = Density(
         np.memmap(
             filename,
             mode="w+",
-            shape=(len(obs_slices), *subtomo_shape),
+            shape=(len(obs_slices), *box_size),
             dtype=output_dtype,
         ),
         sampling_rate=(1, *target.sampling_rate),
@@ -169,50 +204,54 @@ def main():
     )
     dens.data[:] = target.metadata["mean"]
 
-    target_meta = {
-        k: v for k, v in target.metadata.items() if k in ("mean", "max", "min", "std")
-    }
-
-    total_slices = len(obs_slices)
-    magnitude = len(str(total_slices))
-    for index, obs_slice in enumerate(obs_slices):
-        print(f"Processing {index:{magnitude}} / {total_slices:{magnitude}}", end="\r")
+    data_subset = np.zeros(extraction_shape, dtype = target.data.dtype)
+    pbar = ProgressBar(message = "Orientation ", nchars = 80, total = len(obs_slices))
+    for index, (obs_slice, cand_slice) in enumerate(zip(obs_slices, cand_slices)):
+        pbar.update(index + 1)
 
+        data_subset.fill(0)
+        data_subset[cand_slice] = target.data[obs_slice]
         target_subset = Density(
-            target.data[obs_slice],
+            data_subset,
             sampling_rate=target.sampling_rate,
             origin=target.origin,
         )
 
-        rotation_matrix = np.eye(target_subset.data.ndim)
-        if args.alignment_vector is not None:
-            normal = orientations.rotations[index]
-            # To visualize only normals were picked we would need to do the opposite
-            # However, this allows us to align everything to the template axis
-            rotation_matrix = rotation_aligning_vectors(
-                normal, target_vector=args.alignment_vector
-            )
-
+        if args.align_orientations:
+            rotation_matrix = euler_to_rotationmatrix(orientations.rotations[index])
             target_subset = target_subset.rigid_transform(
                 rotation_matrix=rotation_matrix,
                 use_geometric_center=True,
                 order=args.interpolation_order,
             )
-        target_subset.pad(subtomo_shape, center=True)
+        target_subset.pad(box_size, center=True)
+
+        # target_value = target.data[tuple(orientations.translations[index].astype(int))]
+        # center = np.divide(target_subset.data.shape, 2).astype(int ) + np.mod(target_subset.shape, 2)
+        # print(np.where(target_subset.data == target_value), center)
+        # print(target_subset.data[tuple(center.astype(int))],
+            # target_value,
+            # target_subset.data[tuple(center.astype(int))] == target_value
+        # )
+
         dens.data[index] = target_subset.data
-        orientations.rotations[index] = euler_from_rotationmatrix(rotation_matrix)
+    print("")
 
+    target_meta = {
+        k: v for k, v in target.metadata.items() if k in ("mean", "max", "min", "std")
+    }
     dens.metadata.update(target_meta)
+    dens.metadata["batch_dimension"] = (0, )
+
     dens.to_file(args.output_file)
     orientations.to_file(
-        f"{splitext(args.output_file)}_aligned.tsv",
+        f"{splitext(args.output_file)[0]}_aligned.tsv",
         file_format="text"
     )
-    orientations.to_file(
-        f"{splitext(args.output_file)}_aligned.star",
-        file_format="relion"
-    )
-
+    # orientations.to_file(
+    #     f"{splitext(args.output_file)[0]}_aligned.star",
+    #     file_format="relion"
+    # )
 
 if __name__ == "__main__":
     main()