diskpack 0.7.0__tar.gz → 0.8.0__tar.gz

{diskpack-0.7.0/src/diskpack.egg-info → diskpack-0.8.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: diskpack
-Version: 0.7.0
+Version: 0.8.0
 Summary: A high-performance vectorized circle packer with spatial hashing.
 Author-email: James Kelly <mrkellyjam@gmail.com>
 Requires-Python: >=3.8

{diskpack-0.7.0 → diskpack-0.8.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "diskpack"
-version = "0.7.0"
+version = "0.8.0"
 authors = [{ name="James Kelly", email="mrkellyjam@gmail.com" }]
 description = "A high-performance vectorized circle packer with spatial hashing."
 readme = "README.md"

diskpack-0.8.0/src/diskpack/packer.py

@@ -0,0 +1,782 @@
+import numpy as np
+import heapq
+from dataclasses import dataclass, field
+from typing import List, Optional, Iterator, Tuple, Dict, Set
+from enum import Enum
+
+# Type aliases
+Polygon = np.ndarray
+Point = np.ndarray
+GridKey = Tuple[int, int]
+Circle = Tuple[float, float, float]
+
+# Threshold for switching between vectorized and spatial index approaches
+VECTORIZED_THRESHOLD = 750
+
+
+class PackingMode(Enum):
+    """Available packing strategies."""
+    RANDOM = "random"           # Random sampling (original)
+    HEX_GRID = "hex_grid"       # Hexagonal grid (fixed radius only)
+    FRONT = "front"             # Front-based packing (highest density)
+
+
+@dataclass
+class PackingConfig:
+    """Configuration parameters for the circle packing algorithm."""
+    padding: float = 1.5
+    min_radius: float = 1.0
+    grid_resolution_divisor: float = 25
+    max_failed_attempts: int = 200
+    mega_circle_threshold: float = 0.5
+    ray_cast_epsilon: float = 1e-10
+    sample_batch_size: int = 50
+    fixed_radius: Optional[float] = None
+    use_hex_grid: bool = True
+    use_front_packing: bool = False  # New option for front-based packing
+    verbose: bool = False
+
+
+@dataclass
+class PackingProgress:
+    """Tracks the current state of the packing algorithm."""
+    circles_placed: int = 0
+    failed_attempts: int = 0
+    max_failed_attempts: int = 200
+
+    @property
+    def progress_ratio(self) -> float:
+        """How close we are to stopping (0.0 = just started, 1.0 = about to stop)."""
+        return self.failed_attempts / self.max_failed_attempts
+
+    def __str__(self) -> str:
+        return f"Placed: {self.circles_placed} | Failed attempts: {self.failed_attempts}/{self.max_failed_attempts} ({self.progress_ratio:.0%})"
+
+
+class PolygonGeometry:
+    """Handles geometric calculations for polygon boundaries."""
+
+    def __init__(self, polygons: List[Polygon], epsilon: float = 1e-10):
+        self.polygons = [np.array(p, dtype=float) for p in polygons]
+        self.epsilon = epsilon
+        self._compute_bounds()
+        self._precompute_edges()
+
+    def _compute_bounds(self) -> None:
+        all_vertices = np.vstack(self.polygons)
+        self.min_coords = np.min(all_vertices, axis=0)
+        self.max_coords = np.max(all_vertices, axis=0)
+
+    def _precompute_edges(self) -> None:
+        """Precompute edge data for vectorized distance calculations."""
+        all_p1 = []
+        all_p2 = []
+        for poly in self.polygons:
+            n = len(poly)
+            for i in range(n):
+                all_p1.append(poly[i])
+                all_p2.append(poly[(i + 1) % n])
+
+        self.edge_starts = np.array(all_p1)
+        self.edge_ends = np.array(all_p2)
+        self.edge_vecs = self.edge_ends - self.edge_starts
+        self.edge_lengths_sq = np.sum(self.edge_vecs ** 2, axis=1)
+        self.edge_lengths = np.sqrt(self.edge_lengths_sq)
+        
+        # Compute inward normals for each edge
+        self.edge_normals = np.zeros_like(self.edge_vecs)
+        for i, (start, vec) in enumerate(zip(self.edge_starts, self.edge_vecs)):
+            # Perpendicular (rotate 90 degrees)
+            normal = np.array([-vec[1], vec[0]])
+            if self.edge_lengths[i] > 0:
+                normal = normal / self.edge_lengths[i]
+            # Check if normal points inward (toward polygon center)
+            midpoint = start + vec / 2
+            test_point = midpoint + normal * 0.001
+            if not self._point_in_polygon_single(test_point):
+                normal = -normal
+            self.edge_normals[i] = normal
+
+    def _point_in_polygon_single(self, point: Point) -> bool:
+        """Check if a single point is inside the polygon."""
+        x, y = point[0], point[1]
+        inside = False
+        for poly in self.polygons:
+            n = len(poly)
+            for i in range(n):
+                p1, p2 = poly[i], poly[(i + 1) % n]
+                if ((p1[1] > y) != (p2[1] > y)) and \
+                   (x < (p2[0] - p1[0]) * (y - p1[1]) / (p2[1] - p1[1] + self.epsilon) + p1[0]):
+                    inside = not inside
+        return inside
+
+    def contains_points(self, points: np.ndarray) -> np.ndarray:
+        """Even-Odd Rule for interior detection, supports holes."""
+        x, y = points[:, 0], points[:, 1]
+        inside = np.zeros(len(points), dtype=bool)
+
+        for poly in self.polygons:
+            n = len(poly)
+            for i in range(n):
+                p1, p2 = poly[i], poly[(i + 1) % n]
+                crosses_edge = (p1[1] > y) != (p2[1] > y)
+                dy = p2[1] - p1[1] + self.epsilon
+                x_intercept = (p2[0] - p1[0]) * (y - p1[1]) / dy + p1[0]
+                inside ^= crosses_edge & (x < x_intercept)
+
+        return inside
+
+    def distance_to_boundary(self, point: Point) -> float:
+        """Vectorized distance to nearest polygon edge."""
+        to_point = point - self.edge_starts
+        dots = np.sum(to_point * self.edge_vecs, axis=1)
+
+        with np.errstate(divide='ignore', invalid='ignore'):
+            t = np.clip(dots / self.edge_lengths_sq, 0, 1)
+            t = np.where(self.edge_lengths_sq == 0, 0, t)
+
+        projections = self.edge_starts + t[:, np.newaxis] * self.edge_vecs
+        distances = np.linalg.norm(point - projections, axis=1)
+
+        return float(np.min(distances))
+
+    def distances_to_boundary_batch(self, points: np.ndarray) -> np.ndarray:
+        """Vectorized distance calculation for multiple points at once."""
+        to_point = points[:, np.newaxis, :] - self.edge_starts[np.newaxis, :, :]
+        dots = np.sum(to_point * self.edge_vecs[np.newaxis, :, :], axis=2)
+
+        with np.errstate(divide='ignore', invalid='ignore'):
+            t = np.clip(dots / self.edge_lengths_sq[np.newaxis, :], 0, 1)
+            t = np.where(self.edge_lengths_sq[np.newaxis, :] == 0, 0, t)
+
+        projections = (
+            self.edge_starts[np.newaxis, :, :] +
+            t[:, :, np.newaxis] * self.edge_vecs[np.newaxis, :, :]
+        )
+        distances = np.linalg.norm(points[:, np.newaxis, :] - projections, axis=2)
+
+        return np.min(distances, axis=1)
+    
+    def closest_point_on_edge(self, point: Point, edge_idx: int) -> Tuple[Point, float]:
+        """Find closest point on a specific edge and return it with distance."""
+        start = self.edge_starts[edge_idx]
+        vec = self.edge_vecs[edge_idx]
+        length_sq = self.edge_lengths_sq[edge_idx]
+        
+        if length_sq == 0:
+            return start, np.linalg.norm(point - start)
+        
+        t = np.clip(np.dot(point - start, vec) / length_sq, 0, 1)
+        closest = start + t * vec
+        return closest, np.linalg.norm(point - closest)
+
+
+@dataclass
+class SpatialIndex:
+    """Grid-based spatial index for efficient collision detection."""
+    cell_size: float
+    origin: np.ndarray
+    mega_threshold: float
+    grid: Dict[GridKey, List[int]] = field(default_factory=dict)
+    mega_circles: List[int] = field(default_factory=list)
+
+    _centers: np.ndarray = field(default_factory=lambda: np.empty((0, 2)))
+    _radii: np.ndarray = field(default_factory=lambda: np.empty(0))
+
+    def add_circle(self, index: int, center: Point, radius: float) -> None:
+        self._centers = np.vstack([self._centers, center]) if len(self._centers) > 0 else center.reshape(1, 2)
+        self._radii = np.append(self._radii, radius)
+
+        if radius > self.cell_size * self.mega_threshold:
+            self.mega_circles.append(index)
+        else:
+            key = self._get_cell_key(center)
+            self.grid.setdefault(key, []).append(index)
+
+    def get_nearby_indices(self, point: Point) -> Iterator[int]:
+        yield from self.mega_circles
+        center_key = self._get_cell_key(point)
+        for dx in range(-1, 2):
+            for dy in range(-1, 2):
+                neighbor_key = (center_key[0] + dx, center_key[1] + dy)
+                if neighbor_key in self.grid:
+                    yield from self.grid[neighbor_key]
+
+    def distance_to_circles(self, point: Point) -> float:
+        """Get minimum distance from point to any existing circle's edge."""
+        if len(self._centers) == 0:
+            return float('inf')
+
+        indices = list(self.get_nearby_indices(point))
+        if not indices:
+            return float('inf')
+
+        centers = self._centers[indices]
+        radii = self._radii[indices]
+
+        distances = np.linalg.norm(centers - point, axis=1) - radii
+        return float(np.min(distances))
+
+    def _get_cell_key(self, point: Point) -> GridKey:
+        cell_coords = ((point - self.origin) // self.cell_size).astype(int)
+        return (int(cell_coords[0]), int(cell_coords[1]))
+
+
+@dataclass(order=True)
+class FrontCandidate:
+    """A candidate position for placing a circle in front-based packing."""
+    priority: float  # Negative radius for max-heap behavior with min-heap
+    center: Point = field(compare=False)
+    radius: float = field(compare=False)
+    source_type: str = field(compare=False)  # 'edge', 'circle-circle', 'circle-edge'
+    source_ids: Tuple = field(compare=False)  # IDs of circles/edges that generated this candidate
+
+
+class CirclePacker:
+    """Packs circles within polygon boundaries using various strategies."""
+
+    def __init__(self, polygons: List[Polygon], config: Optional[PackingConfig] = None):
+        self.config = config or PackingConfig()
+        self.geometry = PolygonGeometry(polygons, self.config.ray_cast_epsilon)
+        self.centers: List[Point] = []
+        self.radii: List[float] = []
+        self.progress = PackingProgress(max_failed_attempts=self.config.max_failed_attempts)
+
+        extent = max(self.geometry.max_coords - self.geometry.min_coords)
+        cell_size = extent / self.config.grid_resolution_divisor
+        self.spatial_index = SpatialIndex(
+            cell_size=cell_size,
+            origin=self.geometry.min_coords,
+            mega_threshold=self.config.mega_circle_threshold
+        )
+
+        # Cache for numpy arrays
+        self._centers_arr: Optional[np.ndarray] = None
+        self._radii_arr: Optional[np.ndarray] = None
+        self._cache_valid = False
+
+    def _invalidate_cache(self) -> None:
+        self._cache_valid = False
+
+    def _get_arrays(self) -> Tuple[np.ndarray, np.ndarray]:
+        if not self._cache_valid or self._centers_arr is None:
+            if len(self.centers) > 0:
+                self._centers_arr = np.array(self.centers)
+                self._radii_arr = np.array(self.radii)
+            else:
+                self._centers_arr = np.empty((0, 2))
+                self._radii_arr = np.empty(0)
+            self._cache_valid = True
+        return self._centers_arr, self._radii_arr
+
+    def _sample_candidate_points(self, count: int) -> np.ndarray:
+        points = np.random.uniform(
+            self.geometry.min_coords,
+            self.geometry.max_coords,
+            size=(count, 2)
+        )
+        return points[self.geometry.contains_points(points)]
+
+    def _compute_max_radius(self, point: Point) -> float:
+        max_radius = self.geometry.distance_to_boundary(point)
+        circle_dist = self.spatial_index.distance_to_circles(point)
+        max_radius = min(max_radius, circle_dist)
+        return max_radius - self.config.padding
+
+    def _compute_max_radii_batch(self, points: np.ndarray) -> np.ndarray:
+        if len(points) == 0:
+            return np.array([])
+
+        max_radii = self.geometry.distances_to_boundary_batch(points)
+
+        if len(self.centers) == 0:
+            return max_radii - self.config.padding
+
+        centers_arr, radii_arr = self._get_arrays()
+
+        if len(self.centers) < VECTORIZED_THRESHOLD:
+            dists = np.linalg.norm(
+                points[:, np.newaxis, :] - centers_arr[np.newaxis, :, :],
+                axis=2
+            ) - radii_arr
+            min_circle_dists = np.min(dists, axis=1)
+            max_radii = np.minimum(max_radii, min_circle_dists)
+        else:
+            for i, point in enumerate(points):
+                indices = list(self.spatial_index.get_nearby_indices(point))
+                if indices:
+                    nearby_centers = centers_arr[indices]
+                    nearby_radii = radii_arr[indices]
+                    distances = np.linalg.norm(nearby_centers - point, axis=1) - nearby_radii
+                    max_radii[i] = min(max_radii[i], np.min(distances))
+
+        return max_radii - self.config.padding
+
+    def _find_best_placement(self, candidates: np.ndarray) -> Optional[Tuple[Point, float]]:
+        if len(candidates) == 0:
+            return None
+
+        radii = self._compute_max_radii_batch(candidates)
+        fixed = self.config.fixed_radius
+
+        if fixed is not None:
+            valid_mask = radii >= fixed
+            if not np.any(valid_mask):
+                return None
+            valid_indices = np.where(valid_mask)[0]
+            best_idx = valid_indices[0]
+            return candidates[best_idx], fixed
+        else:
+            best_idx = np.argmax(radii)
+            best_radius = radii[best_idx]
+
+            if best_radius >= self.config.min_radius:
+                return candidates[best_idx], best_radius
+            return None
+
+    def _place_circle(self, center: Point, radius: float) -> None:
+        idx = len(self.centers)
+        self.centers.append(center)
+        self.radii.append(radius)
+        self.spatial_index.add_circle(idx, center, radius)
+        self._invalidate_cache()
+
+    def _is_valid_placement(self, center: Point, radius: float) -> bool:
+        """Check if a circle placement is valid (inside polygon, no overlaps)."""
+        # Check if center is inside polygon
+        if not self.geometry._point_in_polygon_single(center):
+            return False
+        
+        # Check boundary distance
+        boundary_dist = self.geometry.distance_to_boundary(center)
+        if boundary_dist < radius + self.config.padding - 1e-9:
+            return False
+        
+        # Check circle overlaps
+        if len(self.centers) > 0:
+            centers_arr, radii_arr = self._get_arrays()
+            distances = np.linalg.norm(centers_arr - center, axis=1)
+            min_allowed = radii_arr + radius + self.config.padding
+            if np.any(distances < min_allowed - 1e-9):
+                return False
+        
+        return True
+
+    # =========================================================================
+    # Hex Grid Packing
+    # =========================================================================
+
+    def _generate_hex_grid(self, radius: float) -> np.ndarray:
+        spacing = (radius + self.config.padding) * 2
+        dy = spacing * np.sqrt(3) / 2
+
+        min_x, min_y = self.geometry.min_coords
+        max_x, max_y = self.geometry.max_coords
+
+        min_x -= spacing
+        min_y -= spacing
+        max_x += spacing
+        max_y += spacing
+
+        points = []
+        row = 0
+        y = min_y
+
+        while y <= max_y:
+            x_offset = (spacing / 2) if row % 2 else 0
+            x = min_x + x_offset
+
+            while x <= max_x:
+                points.append([x, y])
+                x += spacing
+
+            y += dy
+            row += 1
+
+        return np.array(points) if points else np.empty((0, 2))
+
+    def _pack_hex_grid(self) -> List[Circle]:
+        radius = self.config.fixed_radius
+        circles = []
+
+        grid_points = self._generate_hex_grid(radius)
+
+        if len(grid_points) == 0:
+            return circles
+
+        inside_mask = self.geometry.contains_points(grid_points)
+        interior_points = grid_points[inside_mask]
+
+        min_clearance = radius + self.config.padding
+        boundary_distances = self.geometry.distances_to_boundary_batch(interior_points)
+        valid_mask = boundary_distances >= min_clearance
+
+        valid_points = interior_points[valid_mask]
+
+        if self.config.verbose:
+            print(f"Hex grid: {len(grid_points)} total -> {len(interior_points)} inside -> {len(valid_points)} valid")
+
+        for point in valid_points:
+            self._place_circle(point, radius)
+            circles.append((float(point[0]), float(point[1]), float(radius)))
+
+        if self.config.verbose:
+            print(f"Done! Placed {len(circles)} circles")
+
+        return circles
+
+    # =========================================================================
+    # Front-Based Packing
+    # =========================================================================
+
+    def _find_tangent_circle_two_circles(
+        self, c1: Point, r1: float, c2: Point, r2: float, r: float
+    ) -> List[Point]:
+        """
+        Find positions where a circle of radius r is tangent to two existing circles.
+        Returns 0, 1, or 2 valid positions.
+        """
+        d = np.linalg.norm(c2 - c1)
+        
+        # Distance from each center to the new circle's center
+        d1 = r1 + r + self.config.padding
+        d2 = r2 + r + self.config.padding
+        
+        # Check if solution exists (triangle inequality)
+        if d > d1 + d2 or d < abs(d1 - d2) or d < 1e-10:
+            return []
+        
+        # Solve using law of cosines
+        # d1^2 = d^2 + d2^2 - 2*d*d2*cos(angle at c2)
+        # Actually easier: find intersection of two circles centered at c1, c2
+        
+        # Using the formula for circle-circle intersection
+        a = (d1**2 - d2**2 + d**2) / (2 * d)
+        h_sq = d1**2 - a**2
+        
+        if h_sq < 0:
+            return []
+        
+        h = np.sqrt(h_sq)
+        
+        # Unit vector from c1 to c2
+        u = (c2 - c1) / d
+        # Perpendicular
+        v = np.array([-u[1], u[0]])
+        
+        # Midpoint along c1-c2 axis
+        p = c1 + a * u
+        
+        # Two solutions
+        solutions = []
+        if h < 1e-10:
+            solutions.append(p)
+        else:
+            solutions.append(p + h * v)
+            solutions.append(p - h * v)
+        
+        return solutions
+
+    def _find_tangent_circle_edge(
+        self, edge_idx: int, r: float
+    ) -> List[Tuple[Point, float]]:
+        """
+        Find positions along an edge where circles of radius r can be placed.
+        Returns list of (center, t) where t is position along edge [0, 1].
+        """
+        start = self.geometry.edge_starts[edge_idx]
+        vec = self.geometry.edge_vecs[edge_idx]
+        normal = self.geometry.edge_normals[edge_idx]
+        length = self.geometry.edge_lengths[edge_idx]
+        
+        if length < 1e-10:
+            return []
+        
+        # Circle center is offset from edge by radius + padding
+        offset = r + self.config.padding
+        
+        # Generate positions along the edge
+        positions = []
+        spacing = (r + self.config.padding) * 2
+        
+        t = offset / length  # Start offset from edge start
+        while t < 1 - offset / length:
+            point_on_edge = start + t * vec
+            center = point_on_edge + offset * normal
+            positions.append((center, t))
+            t += spacing / length
+        
+        return positions
+
+    def _find_tangent_circle_circle_and_edge(
+        self, circle_idx: int, edge_idx: int, r: float
+    ) -> List[Point]:
+        """
+        Find positions where a circle of radius r is tangent to both
+        an existing circle and a polygon edge.
+        """
+        c = self.centers[circle_idx]
+        rc = self.radii[circle_idx]
+        
+        start = self.geometry.edge_starts[edge_idx]
+        vec = self.geometry.edge_vecs[edge_idx]
+        normal = self.geometry.edge_normals[edge_idx]
+        length = self.geometry.edge_lengths[edge_idx]
+        
+        if length < 1e-10:
+            return []
+        
+        # New circle must be:
+        # 1. At distance rc + r + padding from circle center
+        # 2. At distance r + padding from edge
+        
+        edge_offset = r + self.config.padding
+        circle_dist = rc + r + self.config.padding
+        
+        # The center lies on a line parallel to the edge, offset by edge_offset
+        # And on a circle around c with radius circle_dist
+        
+        # Line: point = start + t * vec + edge_offset * normal
+        # Circle: |point - c| = circle_dist
+        
+        # Substitute: |start + t * vec + edge_offset * normal - c| = circle_dist
+        # Let p0 = start + edge_offset * normal - c
+        # |p0 + t * vec| = circle_dist
+        # |p0|^2 + 2*t*(p0 . vec) + t^2*|vec|^2 = circle_dist^2
+        
+        p0 = start + edge_offset * normal - c
+        a = np.dot(vec, vec)  # |vec|^2
+        b = 2 * np.dot(p0, vec)
+        c_coef = np.dot(p0, p0) - circle_dist**2
+        
+        discriminant = b**2 - 4 * a * c_coef
+        
+        if discriminant < 0:
+            return []
+        
+        solutions = []
+        sqrt_disc = np.sqrt(discriminant)
+        
+        for t in [(-b + sqrt_disc) / (2 * a), (-b - sqrt_disc) / (2 * a)]:
+            if 0 <= t <= 1:
+                center = start + t * vec + edge_offset * normal
+                solutions.append(center)
+        
+        return solutions
+
+    def _get_max_radius_at_point(self, center: Point) -> float:
+        """Get the maximum radius that can fit at a given center point."""
+        # Distance to boundary
+        max_r = self.geometry.distance_to_boundary(center)
+        
+        # Distance to existing circles
+        if len(self.centers) > 0:
+            centers_arr, radii_arr = self._get_arrays()
+            distances = np.linalg.norm(centers_arr - center, axis=1) - radii_arr
+            max_r = min(max_r, np.min(distances))
+        
+        return max_r - self.config.padding
+
+    def _pack_front(self) -> List[Circle]:
+        """
+        Pack circles using front-based algorithm.
+        Achieves higher density by systematically filling from edges inward.
+        """
+        circles = []
+        min_r = self.config.min_radius
+        fixed_r = self.config.fixed_radius
+        
+        # Priority queue: (negative_radius, center, radius, source_info)
+        # Using negative radius for max-heap behavior
+        candidates: List[Tuple[float, int, Point, float]] = []
+        candidate_id = 0
+        
+        # Track which circle pairs we've already processed
+        processed_pairs: Set[Tuple[int, int]] = set()
+        processed_circle_edge: Set[Tuple[int, int]] = set()
+        
+        def add_candidate(center: Point, radius: float):
+            nonlocal candidate_id
+            if radius >= min_r:
+                heapq.heappush(candidates, (-radius, candidate_id, center, radius))
+                candidate_id += 1
+
+        # Phase 1: Seed candidates along all edges
+        if self.config.verbose:
+            print("Phase 1: Seeding edge candidates...")
+        
+        for edge_idx in range(len(self.geometry.edge_starts)):
+            if fixed_r is not None:
+                # Fixed radius mode: place along edges
+                edge_positions = self._find_tangent_circle_edge(edge_idx, fixed_r)
+                for center, t in edge_positions:
+                    if self._is_valid_placement(center, fixed_r):
+                        add_candidate(center, fixed_r)
+            else:
+                # Variable radius: sample points along edge and compute max radius
+                edge_positions = self._find_tangent_circle_edge(edge_idx, min_r)
+                for center, t in edge_positions:
+                    if self.geometry._point_in_polygon_single(center):
+                        max_r = self._get_max_radius_at_point(center)
+                        if max_r >= min_r:
+                            add_candidate(center, max_r)
+        
+        if self.config.verbose:
+            print(f"  Initial candidates: {len(candidates)}")
+        
+        # Phase 2: Main loop - place circles and generate new candidates
+        if self.config.verbose:
+            print("Phase 2: Placing circles...")
+        
+        iterations = 0
+        max_iterations = 100000  # Safety limit
+        
+        while candidates and iterations < max_iterations:
+            iterations += 1
+            
+            # Pop best candidate
+            neg_radius, _, center, radius = heapq.heappop(candidates)
+            
+            # Recompute max radius (things may have changed)
+            if fixed_r is not None:
+                actual_radius = fixed_r
+                if not self._is_valid_placement(center, actual_radius):
+                    continue
+            else:
+                actual_radius = self._get_max_radius_at_point(center)
+                if actual_radius < min_r:
+                    continue
+                if not self._is_valid_placement(center, actual_radius):
+                    continue
+            
+            # Place the circle
+            self._place_circle(center, actual_radius)
+            circles.append((float(center[0]), float(center[1]), float(actual_radius)))
+            
+            if self.config.verbose and len(circles) % 50 == 0:
+                print(f"  Placed {len(circles)} circles, {len(candidates)} candidates remaining")
+            
+            new_circle_idx = len(self.centers) - 1
+            
+            # Generate new candidates from circle-circle tangencies
+            for other_idx in range(new_circle_idx):
+                pair = (min(other_idx, new_circle_idx), max(other_idx, new_circle_idx))
+                if pair in processed_pairs:
+                    continue
+                processed_pairs.add(pair)
+                
+                other_center = self.centers[other_idx]
+                other_radius = self.radii[other_idx]
+                
+                # Try to find tangent circles
+                if fixed_r is not None:
+                    tangent_centers = self._find_tangent_circle_two_circles(
+                        center, actual_radius, other_center, other_radius, fixed_r
+                    )
+                    for tc in tangent_centers:
+                        if self._is_valid_placement(tc, fixed_r):
+                            add_candidate(tc, fixed_r)
+                else:
+                    # For variable radius, try a few different radii
+                    for test_r in [min_r, min_r * 2, min_r * 4]:
+                        tangent_centers = self._find_tangent_circle_two_circles(
+                            center, actual_radius, other_center, other_radius, test_r
+                        )
+                        for tc in tangent_centers:
+                            if self.geometry._point_in_polygon_single(tc):
+                                max_r = self._get_max_radius_at_point(tc)
+                                if max_r >= min_r:
+                                    add_candidate(tc, max_r)
+            
+            # Generate new candidates from circle-edge tangencies
+            for edge_idx in range(len(self.geometry.edge_starts)):
+                ce_pair = (new_circle_idx, edge_idx)
+                if ce_pair in processed_circle_edge:
+                    continue
+                processed_circle_edge.add(ce_pair)
+                
+                if fixed_r is not None:
+                    tangent_centers = self._find_tangent_circle_circle_and_edge(
+                        new_circle_idx, edge_idx, fixed_r
+                    )
+                    for tc in tangent_centers:
+                        if self._is_valid_placement(tc, fixed_r):
+                            add_candidate(tc, fixed_r)
+                else:
+                    for test_r in [min_r, min_r * 2]:
+                        tangent_centers = self._find_tangent_circle_circle_and_edge(
+                            new_circle_idx, edge_idx, test_r
+                        )
+                        for tc in tangent_centers:
+                            if self.geometry._point_in_polygon_single(tc):
+                                max_r = self._get_max_radius_at_point(tc)
+                                if max_r >= min_r:
+                                    add_candidate(tc, max_r)
+        
+        if self.config.verbose:
+            print(f"Done! Placed {len(circles)} circles in {iterations} iterations")
+        
+        return circles
+
+    # =========================================================================
+    # Random Sampling Packing
+    # =========================================================================
+
+    def _pack_random(self) -> Iterator[Circle]:
+        self.progress = PackingProgress(max_failed_attempts=self.config.max_failed_attempts)
+
+        while self.progress.failed_attempts < self.config.max_failed_attempts:
+            candidates = self._sample_candidate_points(self.config.sample_batch_size)
+            result = self._find_best_placement(candidates)
+
+            if result is not None:
+                center, radius = result
+                self._place_circle(center, radius)
+                self.progress.circles_placed += 1
+                self.progress.failed_attempts = 0
+
+                if self.config.verbose and self.progress.circles_placed % 25 == 0:
+                    print(self.progress)
+
+                yield (float(center[0]), float(center[1]), float(radius))
+            else:
+                self.progress.failed_attempts += 1
+
+                if self.config.verbose and self.progress.failed_attempts % 50 == 0:
+                    print(self.progress)
+
+        if self.config.verbose:
+            print(f"Done! {self.progress}")
+
+    # =========================================================================
+    # Main Entry Points
+    # =========================================================================
+
+    def generate(self) -> Iterator[Circle]:
+        """
+        Generate circles until no more can be placed.
+
+        Strategy selection:
+        1. If use_front_packing=True: use front-based algorithm (highest density)
+        2. Else if fixed_radius and use_hex_grid=True: use hex grid (fastest for fixed)
+        3. Else: use random sampling
+
+        Yields:
+            Tuples of (x, y, radius) for each placed circle.
+        """
+        # Front-based packing (highest density)
+        if self.config.use_front_packing:
+            yield from self._pack_front()
+            return
+        
+        # Hex grid for fixed radius (unless disabled)
+        if self.config.fixed_radius is not None and self.config.use_hex_grid:
+            yield from self._pack_hex_grid()
+            return
+
+        # Random sampling (original method)
+        yield from self._pack_random()
+
+    def pack(self) -> List[Circle]:
+        """Pack circles and return them as a list."""
+        return list(self.generate())

{diskpack-0.7.0 → diskpack-0.8.0/src/diskpack.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: diskpack
-Version: 0.7.0
+Version: 0.8.0
 Summary: A high-performance vectorized circle packer with spatial hashing.
 Author-email: James Kelly <mrkellyjam@gmail.com>
 Requires-Python: >=3.8

diskpack-0.7.0/src/diskpack/packer.py

@@ -1,430 +0,0 @@
-import numpy as np
-from dataclasses import dataclass, field
-from typing import List, Optional, Iterator, Tuple, Dict
-
-# Type aliases
-Polygon = np.ndarray
-Point = np.ndarray
-GridKey = Tuple[int, int]
-Circle = Tuple[float, float, float]
-
-# Threshold for switching between vectorized and spatial index approaches
-VECTORIZED_THRESHOLD = 750
-
-
-@dataclass
-class PackingConfig:
-    """Configuration parameters for the circle packing algorithm."""
-    padding: float = 1.5
-    min_radius: float = 1.0
-    grid_resolution_divisor: float = 25
-    max_failed_attempts: int = 200
-    mega_circle_threshold: float = 0.5
-    ray_cast_epsilon: float = 1e-10
-    sample_batch_size: int = 50
-    fixed_radius: Optional[float] = None
-    use_hex_grid: bool = True
-    verbose: bool = False
-
-
-@dataclass
-class PackingProgress:
-    """Tracks the current state of the packing algorithm."""
-    circles_placed: int = 0
-    failed_attempts: int = 0
-    max_failed_attempts: int = 200
-
-    @property
-    def progress_ratio(self) -> float:
-        """How close we are to stopping (0.0 = just started, 1.0 = about to stop)."""
-        return self.failed_attempts / self.max_failed_attempts
-
-    def __str__(self) -> str:
-        return f"Placed: {self.circles_placed} | Failed attempts: {self.failed_attempts}/{self.max_failed_attempts} ({self.progress_ratio:.0%})"
-
-
-class PolygonGeometry:
-    """Handles geometric calculations for polygon boundaries."""
-
-    def __init__(self, polygons: List[Polygon], epsilon: float = 1e-10):
-        self.polygons = [np.array(p, dtype=float) for p in polygons]
-        self.epsilon = epsilon
-        self._compute_bounds()
-        self._precompute_edges()
-
-    def _compute_bounds(self) -> None:
-        all_vertices = np.vstack(self.polygons)
-        self.min_coords = np.min(all_vertices, axis=0)
-        self.max_coords = np.max(all_vertices, axis=0)
-
-    def _precompute_edges(self) -> None:
-        """Precompute edge data for vectorized distance calculations."""
-        all_p1 = []
-        all_p2 = []
-        for poly in self.polygons:
-            n = len(poly)
-            for i in range(n):
-                all_p1.append(poly[i])
-                all_p2.append(poly[(i + 1) % n])
-
-        self.edge_starts = np.array(all_p1)
-        self.edge_ends = np.array(all_p2)
-        self.edge_vecs = self.edge_ends - self.edge_starts
-        self.edge_lengths_sq = np.sum(self.edge_vecs ** 2, axis=1)
-
-    def contains_points(self, points: np.ndarray) -> np.ndarray:
-        """Even-Odd Rule for interior detection, supports holes."""
-        x, y = points[:, 0], points[:, 1]
-        inside = np.zeros(len(points), dtype=bool)
-
-        for poly in self.polygons:
-            n = len(poly)
-            for i in range(n):
-                p1, p2 = poly[i], poly[(i + 1) % n]
-                crosses_edge = (p1[1] > y) != (p2[1] > y)
-                dy = p2[1] - p1[1] + self.epsilon
-                x_intercept = (p2[0] - p1[0]) * (y - p1[1]) / dy + p1[0]
-                inside ^= crosses_edge & (x < x_intercept)
-
-        return inside
-
-    def distance_to_boundary(self, point: Point) -> float:
-        """Vectorized distance to nearest polygon edge."""
-        to_point = point - self.edge_starts
-
-        dots = np.sum(to_point * self.edge_vecs, axis=1)
-
-        with np.errstate(divide='ignore', invalid='ignore'):
-            t = np.clip(dots / self.edge_lengths_sq, 0, 1)
-            t = np.where(self.edge_lengths_sq == 0, 0, t)
-
-        projections = self.edge_starts + t[:, np.newaxis] * self.edge_vecs
-        distances = np.linalg.norm(point - projections, axis=1)
-
-        return float(np.min(distances))
-
-    def distances_to_boundary_batch(self, points: np.ndarray) -> np.ndarray:
-        """
-        Vectorized distance calculation for multiple points at once.
-        """
-        n_points = len(points)
-        n_edges = len(self.edge_starts)
-
-        to_point = points[:, np.newaxis, :] - self.edge_starts[np.newaxis, :, :]
-        dots = np.sum(to_point * self.edge_vecs[np.newaxis, :, :], axis=2)
-
-        with np.errstate(divide='ignore', invalid='ignore'):
-            t = np.clip(dots / self.edge_lengths_sq[np.newaxis, :], 0, 1)
-            t = np.where(self.edge_lengths_sq[np.newaxis, :] == 0, 0, t)
-
-        projections = (
-            self.edge_starts[np.newaxis, :, :] +
-            t[:, :, np.newaxis] * self.edge_vecs[np.newaxis, :, :]
-        )
-
-        distances = np.linalg.norm(points[:, np.newaxis, :] - projections, axis=2)
-
-        return np.min(distances, axis=1)
-
-
-@dataclass
-class SpatialIndex:
-    """Grid-based spatial index for efficient collision detection."""
-    cell_size: float
-    origin: np.ndarray
-    mega_threshold: float
-    grid: Dict[GridKey, List[int]] = field(default_factory=dict)
-    mega_circles: List[int] = field(default_factory=list)
-
-    _centers: np.ndarray = field(default_factory=lambda: np.empty((0, 2)))
-    _radii: np.ndarray = field(default_factory=lambda: np.empty(0))
-
-    def add_circle(self, index: int, center: Point, radius: float) -> None:
-        self._centers = np.vstack([self._centers, center]) if len(self._centers) > 0 else center.reshape(1, 2)
-        self._radii = np.append(self._radii, radius)
-
-        if radius > self.cell_size * self.mega_threshold:
-            self.mega_circles.append(index)
-        else:
-            key = self._get_cell_key(center)
-            self.grid.setdefault(key, []).append(index)
-
-    def get_nearby_indices(self, point: Point) -> Iterator[int]:
-        yield from self.mega_circles
-        center_key = self._get_cell_key(point)
-        for dx in range(-1, 2):
-            for dy in range(-1, 2):
-                neighbor_key = (center_key[0] + dx, center_key[1] + dy)
-                if neighbor_key in self.grid:
-                    yield from self.grid[neighbor_key]
-
-    def distance_to_circles(self, point: Point) -> float:
-        """Get minimum distance from point to any existing circle's edge."""
-        if len(self._centers) == 0:
-            return float('inf')
-
-        indices = list(self.get_nearby_indices(point))
-        if not indices:
-            return float('inf')
-
-        centers = self._centers[indices]
-        radii = self._radii[indices]
-
-        distances = np.linalg.norm(centers - point, axis=1) - radii
-        return float(np.min(distances))
-
-    def _get_cell_key(self, point: Point) -> GridKey:
-        cell_coords = ((point - self.origin) // self.cell_size).astype(int)
-        return (int(cell_coords[0]), int(cell_coords[1]))
-
-
-class CirclePacker:
-    """Packs circles within polygon boundaries using random sampling."""
-
-    def __init__(self, polygons: List[Polygon], config: Optional[PackingConfig] = None):
-        self.config = config or PackingConfig()
-        self.geometry = PolygonGeometry(polygons, self.config.ray_cast_epsilon)
-        self.centers: List[Point] = []
-        self.radii: List[float] = []
-        self.progress = PackingProgress(max_failed_attempts=self.config.max_failed_attempts)
-
-        extent = max(self.geometry.max_coords - self.geometry.min_coords)
-        cell_size = extent / self.config.grid_resolution_divisor
-        self.spatial_index = SpatialIndex(
-            cell_size=cell_size,
-            origin=self.geometry.min_coords,
-            mega_threshold=self.config.mega_circle_threshold
-        )
-        
-        # Cache for numpy arrays (avoid repeated conversion)
-        self._centers_arr: Optional[np.ndarray] = None
-        self._radii_arr: Optional[np.ndarray] = None
-        self._cache_valid = False
-
-    def _invalidate_cache(self) -> None:
-        """Mark the numpy array cache as needing refresh."""
-        self._cache_valid = False
-
-    def _get_arrays(self) -> Tuple[np.ndarray, np.ndarray]:
-        """Get cached numpy arrays of centers and radii."""
-        if not self._cache_valid or self._centers_arr is None:
-            if len(self.centers) > 0:
-                self._centers_arr = np.array(self.centers)
-                self._radii_arr = np.array(self.radii)
-            else:
-                self._centers_arr = np.empty((0, 2))
-                self._radii_arr = np.empty(0)
-            self._cache_valid = True
-        return self._centers_arr, self._radii_arr
-
-    def _sample_candidate_points(self, count: int) -> np.ndarray:
-        points = np.random.uniform(
-            self.geometry.min_coords,
-            self.geometry.max_coords,
-            size=(count, 2)
-        )
-        return points[self.geometry.contains_points(points)]
-
-    def _compute_max_radius(self, point: Point) -> float:
-        """Compute max radius for a single point."""
-        max_radius = self.geometry.distance_to_boundary(point)
-        circle_dist = self.spatial_index.distance_to_circles(point)
-        max_radius = min(max_radius, circle_dist)
-        return max_radius - self.config.padding
-
-    def _compute_max_radii_batch(self, points: np.ndarray) -> np.ndarray:
-        """
-        Vectorized max radius computation for multiple points.
-        
-        Uses hybrid approach:
-        - Fully vectorized numpy for small circle counts (faster due to no Python loop)
-        - Spatial index for large circle counts (faster due to fewer distance calculations)
-        """
-        if len(points) == 0:
-            return np.array([])
-
-        # Boundary distances (fully vectorized)
-        max_radii = self.geometry.distances_to_boundary_batch(points)
-
-        if len(self.centers) == 0:
-            return max_radii - self.config.padding
-
-        centers_arr, radii_arr = self._get_arrays()
-
-        if len(self.centers) < VECTORIZED_THRESHOLD:
-            # Fully vectorized approach for small circle counts
-            # Compute distance from each point to each circle: (n_points, n_circles)
-            dists = np.linalg.norm(
-                points[:, np.newaxis, :] - centers_arr[np.newaxis, :, :],
-                axis=2
-            ) - radii_arr
-            
-            # Min distance to any circle for each point
-            min_circle_dists = np.min(dists, axis=1)
-            max_radii = np.minimum(max_radii, min_circle_dists)
-        else:
-            # Spatial index approach for large circle counts
-            for i, point in enumerate(points):
-                indices = list(self.spatial_index.get_nearby_indices(point))
-                if indices:
-                    nearby_centers = centers_arr[indices]
-                    nearby_radii = radii_arr[indices]
-                    distances = np.linalg.norm(nearby_centers - point, axis=1) - nearby_radii
-                    max_radii[i] = min(max_radii[i], np.min(distances))
-
-        return max_radii - self.config.padding
-
-    def _find_best_placement(self, candidates: np.ndarray) -> Optional[Tuple[Point, float]]:
-        if len(candidates) == 0:
-            return None
-
-        radii = self._compute_max_radii_batch(candidates)
-        fixed = self.config.fixed_radius
-
-        if fixed is not None:
-            valid_mask = radii >= fixed
-            if not np.any(valid_mask):
-                return None
-            valid_indices = np.where(valid_mask)[0]
-            best_idx = valid_indices[0]
-            return candidates[best_idx], fixed
-        else:
-            best_idx = np.argmax(radii)
-            best_radius = radii[best_idx]
-
-            if best_radius >= self.config.min_radius:
-                return candidates[best_idx], best_radius
-            return None
-
-    def _place_circle(self, center: Point, radius: float) -> None:
-        idx = len(self.centers)
-        self.centers.append(center)
-        self.radii.append(radius)
-        self.spatial_index.add_circle(idx, center, radius)
-        self._invalidate_cache()
-
-    def _generate_hex_grid(self, radius: float) -> np.ndarray:
-        """
-        Generate a hexagonal grid of points within the bounding box.
-        Hex grid is the optimal packing arrangement for equal circles.
-        """
-        spacing = (radius + self.config.padding) * 2
-        dy = spacing * np.sqrt(3) / 2
-
-        min_x, min_y = self.geometry.min_coords
-        max_x, max_y = self.geometry.max_coords
-
-        # Add margin to ensure coverage
-        min_x -= spacing
-        min_y -= spacing
-        max_x += spacing
-        max_y += spacing
-
-        points = []
-        row = 0
-        y = min_y
-
-        while y <= max_y:
-            # Offset every other row by half spacing
-            x_offset = (spacing / 2) if row % 2 else 0
-            x = min_x + x_offset
-
-            while x <= max_x:
-                points.append([x, y])
-                x += spacing
-
-            y += dy
-            row += 1
-
-        return np.array(points) if points else np.empty((0, 2))
-
-    def _pack_hex_grid(self) -> List[Circle]:
-        """
-        Pack circles using hexagonal grid placement.
-        Much faster and denser than random sampling for fixed radius.
-        """
-        radius = self.config.fixed_radius
-        circles = []
-
-        # Generate hex grid
-        grid_points = self._generate_hex_grid(radius)
-
-        if len(grid_points) == 0:
-            return circles
-
-        # Filter to points inside polygon
-        inside_mask = self.geometry.contains_points(grid_points)
-        interior_points = grid_points[inside_mask]
-
-        # Filter to points with enough clearance from boundary
-        min_clearance = radius + self.config.padding
-        boundary_distances = self.geometry.distances_to_boundary_batch(interior_points)
-        valid_mask = boundary_distances >= min_clearance
-
-        valid_points = interior_points[valid_mask]
-
-        if self.config.verbose:
-            print(f"Hex grid: {len(grid_points)} total -> {len(interior_points)} inside -> {len(valid_points)} valid")
-
-        # All valid points become circles (no collision check needed - hex grid guarantees no overlap)
-        for point in valid_points:
-            self._place_circle(point, radius)
-            circles.append((float(point[0]), float(point[1]), float(radius)))
-
-        if self.config.verbose:
-            print(f"Done! Placed {len(circles)} circles")
-
-        return circles
-
-    def _pack_random(self) -> Iterator[Circle]:
-        """
-        Pack circles using random sampling.
-        Used for variable radius mode or when organic placement is desired.
-        """
-        self.progress = PackingProgress(max_failed_attempts=self.config.max_failed_attempts)
-
-        while self.progress.failed_attempts < self.config.max_failed_attempts:
-            candidates = self._sample_candidate_points(self.config.sample_batch_size)
-            result = self._find_best_placement(candidates)
-
-            if result is not None:
-                center, radius = result
-                self._place_circle(center, radius)
-                self.progress.circles_placed += 1
-                self.progress.failed_attempts = 0
-
-                if self.config.verbose and self.progress.circles_placed % 25 == 0:
-                    print(self.progress)
-
-                yield (float(center[0]), float(center[1]), float(radius))
-            else:
-                self.progress.failed_attempts += 1
-
-                if self.config.verbose and self.progress.failed_attempts % 50 == 0:
-                    print(self.progress)
-
-        if self.config.verbose:
-            print(f"Done! {self.progress}")
-
-    def generate(self) -> Iterator[Circle]:
-        """
-        Generate circles until no more can be placed.
-
-        For fixed_radius mode with use_hex_grid=True (default), uses optimized hex grid.
-        For fixed_radius mode with use_hex_grid=False, uses random sampling for organic look.
-        For variable radius mode, uses random sampling with best-fit selection.
-
-        Yields:
-            Tuples of (x, y, radius) for each placed circle.
-        """
-        # Use hex grid for fixed radius (unless disabled)
-        if self.config.fixed_radius is not None and self.config.use_hex_grid:
-            yield from self._pack_hex_grid()
-            return
-
-        # Random sampling mode (variable radius OR fixed radius with organic placement)
-        yield from self._pack_random()
-
-    def pack(self) -> List[Circle]:
-        """Pack circles and return them as a list."""
-        return list(self.generate())