pyafv 0.3.6__cp314-cp314-musllinux_1_2_aarch64.whl

pyafv/__init__.py

@@ -0,0 +1,18 @@
+"""
+PyAFV - A Python implementation of the **active-finite-Voronoi (AFV) model** in 2D.
+"""
+
+from .physical_params import PhysicalParams, target_delta
+from .finite_voronoi import FiniteVoronoiSimulator
+
+try:
+    from ._version import __version__
+except ImportError:                          # pragma: no cover
+    __version__ = "unknown"
+
+
+__all__ = [
+    "PhysicalParams",
+    "FiniteVoronoiSimulator",
+    "target_delta",
+]

pyafv/_version.py

@@ -0,0 +1,2 @@
+# pyafv/_version.py
+__version__ = "0.3.6"

pyafv/backend.py

@@ -0,0 +1,20 @@
+# chooses fast vs fallback implementation
+
+try:
+    from . import cell_geom as _impl
+    _BACKEND_NAME = "cython"
+    _IMPORT_ERROR = None
+except Exception as e:             # pragma: no cover
+    from . import cell_geom_fallback as _impl
+    _BACKEND_NAME = "python"
+    _IMPORT_ERROR = e
+
+
+# ---- for explicit API ----
+backend_impl = _impl
+
+__all__ = [
+    "backend_impl",
+    "_BACKEND_NAME",
+    "_IMPORT_ERROR"
+]

pyafv/cell_geom_fallback.py

@@ -0,0 +1,249 @@
+"""
+These are pure-Python fallback implementations of key functions for finite Voronoi
+cell geometry and derivative calculations, used when Cython extensions are not
+available.
+"""
+
+import numpy as np
+from collections import defaultdict
+
+
+def build_vertexpair_and_vertexpoints(ridge_vertices_all, ridge_points, num_vertices, N):
+    # Build ridge incidence per vertex, and a lookup for (v1,v2) -> ridge id
+    vertex_incident_ridges = defaultdict(list)
+    # Cheaper key: store both directions as tuple -> int
+    vertexpair2ridge: dict[tuple[int, int], int] = {}
+
+    rv_full = np.asarray(ridge_vertices_all, dtype=int)
+    R = rv_full.shape[0]
+    for k in range(R):
+        v1, v2 = int(rv_full[k, 0]), int(rv_full[k, 1])
+        vertex_incident_ridges[v1].append(k)
+        vertex_incident_ridges[v2].append(k)
+        vertexpair2ridge[(v1, v2)] = k
+        vertexpair2ridge[(v2, v1)] = k
+
+    # For each finite vertex, record which input points (cells) meet there
+    vertex_points = {}
+    if N > 2:
+        for v_id in range(num_vertices):
+            s = set()
+            for ridge_id in vertex_incident_ridges[v_id]:
+                i, j = ridge_points[ridge_id]
+                s.add(i), s.add(j)
+            vertex_points[v_id] = list(s)
+
+    return vertexpair2ridge, vertex_points
+
+
+def pad_regions(regions):
+    # fake function to match cython backend interface
+    return regions
+
+
+def build_point_edges(vor_regions, point_region,
+                      vertices_all, pts,
+                      num_vertices, vertexpair2ridge,
+                      p1, p1_edges_pack, p1_verts_pack,
+                      p2, p2_edges_pack, p2_verts_pack):
+
+    N = pts.shape[0]
+
+    point_edges_type = []
+    point_vertices_f_idx = []
+
+    # --- fast vectorized per-cell processing (no inner edge loop) ---
+    for idx in range(N):
+        region_id = point_region[idx]
+        v_ids = np.asarray(vor_regions[region_id], dtype=int)
+
+        if v_ids.size == 0:
+            point_edges_type.append([])
+            point_vertices_f_idx.append([])
+            continue
+
+        # sort vertices clockwise around cell center
+        rel = vertices_all[v_ids] - pts[idx]
+        angles = np.arctan2(rel[:, 1], rel[:, 0])
+        order = np.argsort(angles)[::-1]
+        v_ids = v_ids[order]
+
+        # consecutive pairs (wrap) -> candidate edges around this cell
+        v1_ids = v_ids
+        v2_ids = np.roll(v_ids, -1)
+
+        # skip ray-ray (both >= num_vertices)
+        valid = ~((v1_ids >= num_vertices) & (v2_ids >= num_vertices))
+        if not np.any(valid):                # pragma: no cover
+            point_edges_type.append([])
+            point_vertices_f_idx.append([])
+            continue
+
+        v1_ids = v1_ids[valid]
+        v2_ids = v2_ids[valid]
+
+        # ---- vectorized ridge id lookup for all edges of this cell ----
+        # use the dict you already built with both orientations:
+        #   vertexpair2ridge[(v1, v2)] = ridge_id
+        # convert all edge pairs in one go via list comprehension (still fast, no Python loop per-edge body)
+        # NB: we keep it simple & reliable; if needed, switch to a sorted-structured-array map later.
+        keys = list(zip(v1_ids.tolist(), v2_ids.tolist()))
+        ridge_ids = np.fromiter(
+            (vertexpair2ridge[k] for k in keys), dtype=int, count=len(keys))
+
+        # decide which endpoint pack to use (p1 vs p2) for each edge
+        use_p1 = (p1[ridge_ids] == idx)
+        use_p2 = ~use_p1
+
+        # gather packs (shape (E,3)), then mask out the -1 slots
+        pack_e = np.empty((len(ridge_ids), 3), dtype=int)
+        pack_v = np.empty((len(ridge_ids), 3), dtype=int)
+
+        if np.any(use_p1):
+            pack_e[use_p1] = p1_edges_pack[ridge_ids[use_p1]]
+            pack_v[use_p1] = p1_verts_pack[ridge_ids[use_p1]]
+        if np.any(use_p2):
+            pack_e[use_p2] = p2_edges_pack[ridge_ids[use_p2]]
+            pack_v[use_p2] = p2_verts_pack[ridge_ids[use_p2]]
+
+        # flatten valid entries in pack order (keeps your exact edge ordering)
+        mask = (pack_e >= 0)
+        edges_type = pack_e[mask].tolist()
+        vertices_f_idx = pack_v[mask].tolist()
+
+        if len(vertices_f_idx) != len(edges_type):
+            raise ValueError("Vertex and edge number not equal!")
+
+        point_edges_type.append(edges_type)
+        point_vertices_f_idx.append(vertices_f_idx)
+
+    return point_edges_type, point_vertices_f_idx
+
+
+def compute_vertex_derivatives(
+            point_edges_type,            # list-of-lists / arrays of edge types
+            point_vertices_f_idx,        # list-of-lists / arrays of vertex ids
+            vertices_all,
+            pts, r, A0_list, P0, num_vertices_ext, num_ridges,
+            vertex_out_points):
+
+    N = pts.shape[0]
+
+    # --- helpers ---
+    def _row_cross(a, b):
+        # z-component of 2D cross, row-wise
+        return a[:, 0] * b[:, 1] - a[:, 1] * b[:, 0]
+
+    def _perp(u):
+        # rotate 90° CW: (ux,uy) -> (uy,-ux)
+        return np.column_stack((u[:, 1], -u[:, 0]))
+
+    vertex_out_da_dtheta = np.zeros((2 * num_ridges, 2))
+    vertex_out_dl_dtheta = np.zeros((2 * num_ridges, 2))
+
+    dA_poly_dh = np.zeros((num_vertices_ext + 2 * num_ridges, 2))
+    dP_poly_dh = np.zeros((num_vertices_ext + 2 * num_ridges, 2))
+
+    area_list = np.zeros(N)
+    perimeter_list = np.zeros(N)
+
+    for idx in range(N):
+        edges_type = np.asarray(point_edges_type[idx], dtype=int)
+        vertices_f_idx = np.asarray(point_vertices_f_idx[idx], dtype=int)
+        E = edges_type.size
+
+        if E < 2:
+            area_list[idx] = np.pi * (r ** 2)
+            perimeter_list[idx] = 2.0 * np.pi * r
+            continue
+
+        # ring indices
+        v1_idx = vertices_f_idx
+        v2_idx = np.roll(vertices_f_idx, -1)
+        v0_idx = np.roll(vertices_f_idx,  1)
+
+        ri = pts[idx]
+        V1 = vertices_all[v1_idx]
+        V2 = vertices_all[v2_idx]
+        V0 = vertices_all[v0_idx]
+        V1mR = V1 - ri
+        V2mR = V2 - ri
+        V0mR = V0 - ri
+
+        mask_str = (edges_type == 1)
+        mask_arc = ~mask_str
+
+        # ----- perimeter & area -----
+        seg12 = V1 - V2
+        l12 = np.linalg.norm(seg12, axis=1)
+        Pi_straight = l12[mask_str].sum()
+        Ai_straight = (-0.5 * _row_cross(V1mR[mask_str], V2mR[mask_str])).sum()
+
+        if np.any(mask_arc):
+            a1_full = np.arctan2(V1mR[:, 1], V1mR[:, 0])
+            a2_full = np.arctan2(V2mR[:, 1], V2mR[:, 0])
+            dangle_full = (a1_full - a2_full) % (2.0 * np.pi)
+            dangle_arc = dangle_full[mask_arc]
+            Pi_arc = (r * dangle_arc).sum()
+            Ai_arc = (0.5 * (r ** 2) * dangle_arc).sum()
+        else:
+            Pi_arc = 0.0
+            Ai_arc = 0.0
+
+        Pi = Pi_straight + Pi_arc
+        Ai = Ai_straight + Ai_arc
+        perimeter_list[idx] = Pi
+        area_list[idx] = Ai
+
+        # ----- dA_poly/dh, dP_poly/dh for v1 -----
+        dAi_v1 = -0.5 * _perp(V2mR) + 0.5 * _perp(V0mR)       # (E,2)
+
+        dPi_v1 = np.zeros((E, 2))
+        if np.any(mask_str):
+            dPi_v1[mask_str] += seg12[mask_str] / l12[mask_str][:, None]
+
+        mask_prev_str = np.roll(mask_str, 1)
+        seg10 = V1 - V0
+        l10 = np.linalg.norm(seg10, axis=1)
+        if np.any(mask_prev_str):
+            dPi_v1[mask_prev_str] += seg10[mask_prev_str] / l10[mask_prev_str][:, None]
+
+        np.add.at(dA_poly_dh, v1_idx, (Ai - A0_list[idx]) * dAi_v1)
+        np.add.at(dP_poly_dh, v1_idx, (Pi - P0) * dPi_v1)
+
+        # ----- arc endpoint sensitivities at outer vertices -----
+        if np.any(mask_arc):
+            # endpoint rows in vertex_out_* are (outer_id - num_vertices_ext)
+            v1_arc_idx = v1_idx[mask_arc]
+            v2_arc_idx = v2_idx[mask_arc]
+            k1 = v1_arc_idx - num_vertices_ext
+            k2 = v2_arc_idx - num_vertices_ext
+            valid1 = (k1 >= 0)
+            valid2 = (k2 >= 0)
+
+            if np.any(valid1) or np.any(valid2):
+                # da/dtheta for endpoints (sector - triangle)
+                da1_full = 0.5 * (r ** 2) * (1.0 - np.cos(dangle_full))   # v1 endpoint
+                da2_full = -da1_full                                      # v2 endpoint
+                da1_arc = da1_full[mask_arc]
+                da2_arc = da2_full[mask_arc]
+
+                # dl/dtheta is ±r
+                dl1 = r
+                dl2 = -r
+
+                vop = vertex_out_points  # rows are sorted [i,j]; column 1 is max(i,j)
+                if np.any(valid1):
+                    k1v = k1[valid1]
+                    # CORRECT which_point: 0 if max(i,j) > idx else 1
+                    which1 = (vop[k1v, 1] <= idx).astype(int)
+                    vertex_out_da_dtheta[k1v, which1] = da1_arc[valid1]
+                    vertex_out_dl_dtheta[k1v, which1] = dl1
+
+                if np.any(valid2):
+                    k2v = k2[valid2]
+                    which2 = (vop[k2v, 1] <= idx).astype(int)
+                    vertex_out_da_dtheta[k2v, which2] = da2_arc[valid2]
+                    vertex_out_dl_dtheta[k2v, which2] = dl2
+
+    return vertex_out_da_dtheta, vertex_out_dl_dtheta, dA_poly_dh, dP_poly_dh, area_list, perimeter_list