Merge pull request #333 from alan-turing-institute/development

MILESTONE: Merging `development` into `main` 🎉

examples/show_data.py

@@ -1,52 +1,94 @@
+import click
 import napari
 
 import pandas as pd
 import numpy as np
 
-from grace.base import GraphAttrs
-from grace.io.image_dataset import mrc_reader
 from pathlib import Path
 
+from grace.base import GraphAttrs
+from grace.io.image_dataset import FILETYPES
 
-# Expects the image data & H5 node positions in the same folder.
-# Use identical naming convention for files & specify whole path to mrc file:
-# e.g. /Users/kulicna/Desktop/dataset/shape_squares/MRC_Synthetic_File_000.mrc
 
-IMAGE_PATH = Path(
-    input(
-        "Enter absolute path to your file "
-        "(e.g. /Users/path/to/your/data/image.mrc, omit ''): "
-    )
+# Define a click command to input the file name directly:
+@click.command(name="Napari Annotator")
+@click.option(
+    "--image_path",
+    type=click.Path(exists=True),
+    help="Path to the image to open in napari annotator",
 )
-NODES_PATH = Path(str(IMAGE_PATH).replace(".mrc", ".h5"))
+@click.option(
+    "--show_features",
+    type=bool,
+    default=True,
+    help="Show feature representation of individual nodes",
+)
+def run_napari_annotator(
+    image_path: str | Path,
+    show_features: bool = False,
+) -> None:
+    """Function to open an image & annotate it in napari.
 
-image_data = mrc_reader(IMAGE_PATH)
-nodes_data = pd.read_hdf(NODES_PATH)
+    Parameters
+    ----------
+    image_path : str | Path
+        Absolute filename of the image to be opened.
+        Expects the image data & H5 node positions in the same folder.
+        Use identical naming convention for these files to pair them up.
+    show_features : bool
+        Whether to display node features stored in the `h5` file.
+        Defaults to False.
 
-points = np.asarray(nodes_data.loc[:, [GraphAttrs.NODE_Y, GraphAttrs.NODE_X]])
-# features = {
-#     GraphAttrs.NODE_FEATURES:
-#     [np.squeeze(f.numpy()) for f in nodes_data.loc[:, "features"]]
-# }
-features = None
-data_name = f"{IMAGE_PATH.stem}"
+    Notes
+    -----
+    - expected file organisation:
+        /path/to/your/image/MRC_Synthetic_File_000.mrc
+        ...identical to...
+        /path/to/your/nodes/MRC_Synthetic_File_000.h5
+    """
+    # Process the image data + load nodes:
+    suffix = str(image_path).split(".")[-1]
+    assert suffix in FILETYPES, f"Choose these filetypes: {FILETYPES.keys()}"
 
-mn, mx = np.min(image_data), np.max(image_data)
+    image_reader = FILETYPES[suffix]
+    image_data = image_reader(Path(image_path))
 
-viewer = napari.Viewer()
-img_layer = viewer.add_image(
-    image_data, name=data_name, contrast_limits=(mn, mx)
-)
-pts_layer = viewer.add_points(
-    points, features=features, size=32, name=f"nodes_{data_name}"
-)
+    nodes_path = image_path.replace(".mrc", ".h5")
+    nodes_data = pd.read_hdf(Path(nodes_path))
 
+    data_name = f"{Path(image_path).stem}"
+
+    # Start a napari window:
+    viewer = napari.Viewer()
+    mn, mx = np.min(image_data), np.max(image_data)
+    viewer.add_image(image_data, name=data_name, contrast_limits=(mn, mx))
+
+    # Locate the nodes as points:
+    points = np.asarray(
+        nodes_data.loc[:, [GraphAttrs.NODE_Y, GraphAttrs.NODE_X]]
+    )
+
+    # Process the node features:
+    if show_features is True:
+        features = {
+            "node_central_pixel_value": [
+                image_data[int(point[0]), int(point[1])] for point in points
+            ]
+        }
+    else:
+        features = None
+
+    viewer.add_points(
+        points, features=features, size=32, name=f"nodes_{data_name}"
+    )
+
+    viewer.window.add_plugin_dock_widget(
+        plugin_name="grace", widget_name="GRACE"
+    )
+    napari.run()
 
-_, widget = viewer.window.add_plugin_dock_widget(
-    plugin_name="grace", widget_name="GRACE"
-)
 
 if __name__ == "__main__":
     # The napari event loop needs to be run under here to allow the window
     # to be spawned from a Python script
-    napari.run()
+    run_napari_annotator()

grace/base.py

@@ -1,13 +1,13 @@
 from __future__ import annotations
 
-import dataclasses
 import enum
 import networkx as nx
 import numpy as np
 import numpy.typing as npt
 import pandas as pd
 
 from typing import Any
+from dataclasses import dataclass
 from scipy.spatial import Delaunay
 
 
@@ -19,12 +19,14 @@ class GraphAttrs(str, enum.Enum):
     NODE_Y = "y"
     NODE_GROUND_TRUTH = "node_ground_truth"
     NODE_PREDICTION = "node_prediction"
-    NODE_FEATURES = "features"
+    NODE_IMG_EMBEDDING = "node_image_patch_latent_embedding"
+    NODE_ENV_EMBEDDING = "laplacian_matrix_multiplied_embedding"
     NODE_CONFIDENCE = "confidence"
     EDGE_SOURCE = "source"
     EDGE_TARGET = "target"
     EDGE_GROUND_TRUTH = "edge_ground_truth"
     EDGE_PREDICTION = "edge_prediction"
+    EDGE_PROPERTIES = "edge_properties"
 
 
 class Annotation(enum.IntEnum):
@@ -35,7 +37,7 @@ class Annotation(enum.IntEnum):
     UNKNOWN = 2
 
 
-@dataclasses.dataclass
+@dataclass
 class Prediction:
     """Prediction dataclass all normalised softmax class probabilities.
 
@@ -89,6 +91,75 @@ def prob_UNKNOWN(self) -> float:
         return self.softmax_probs[Annotation.UNKNOWN]
 
 
+@enum.unique
+class EdgeProps(str, enum.Enum):
+    """Ordered list of edge properties to be used for classifier training."""
+
+    EDGE_LENGTH = "edge_length_nrm"
+    EDGE_ORIENT = "edge_orient_rad"
+    EAST_NEIGHBOUR_LENGTH = "east_to_mid_length_nrm"
+    WEST_NEIGHBOUR_LENGTH = "west_to_mid_length_nrm"
+    EAST_NEIGHBOUR_ORIENT = "east_to_mid_orient_rad"
+    WEST_NEIGHBOUR_ORIENT = "west_to_mid_orient_rad"
+    EAST_TRIANGLE_AREA = "east_triangle_area_nrm"
+    WEST_TRIANGLE_AREA = "west_triangle_area_nrm"
+    EAST_DISTANCE = "east_to_mid_length_rel"
+    WEST_DISTANCE = "west_to_mid_length_rel"
+
+
+@enum.unique
+class PointCoords(str, enum.Enum):
+    """Ordered list of relative point positions (x, y coords) in the graph."""
+
+    SOUTH_POS_X_REL = "south_pos_x_rel"
+    SOUTH_POS_Y_REL = "south_pos_y_rel"
+    NORTH_POS_X_REL = "north_pos_x_rel"
+    NORTH_POS_Y_REL = "north_pos_y_rel"
+    MID_POS_X_REL = "mid_pos_x_rel"
+    MID_POS_Y_REL = "mid_pos_y_rel"
+    EAST_POS_X_REL = "east_pos_x_rel"
+    EAST_POS_Y_REL = "east_pos_y_rel"
+    WEST_POS_X_REL = "west_pos_x_rel"
+    WEST_POS_Y_REL = "west_pos_y_rel"
+
+
+@dataclass
+class Properties:
+    """Structure to organise the key: value pairs of EdgeProps properties."""
+
+    properties_dict: dict[str, float] = None
+
+    @property
+    def property_keys(self) -> list[str]:
+        return list(self.properties_dict.keys())
+
+    @property
+    def property_vals(self) -> list[float]:
+        return list(self.properties_dict.values())
+
+    @property
+    def property_training_data(
+        self, include_relative_coords: bool = False
+    ) -> npt.NDArray:
+        if include_relative_coords is False:
+            return np.stack(
+                [self.properties_dict[prop] for prop in EdgeProps], axis=0
+            )
+        else:
+            keys = list(EdgeProps) + list(PointCoords)
+            return np.stack(
+                [self.properties_dict[prop] for prop in keys], axis=0
+            )
+
+    def from_keys_and_values(
+        self, keys: list[str], values: list[float]
+    ) -> None:
+        assert len(keys) == len(values)
+        assert all(isinstance(k, str) for k in keys)
+        assert all(isinstance(v, (float, np.floating)) for v in values)
+        self.properties_dict = {k: v for k, v in zip(keys, values)}
+
+
 def _map_annotation(annotation: int | Annotation) -> Annotation:
     if isinstance(annotation, Annotation):
         return Annotation

grace/evaluation/__init__.py

@@ -0,0 +1 @@
+from grace import styling  # noqa: F401