Source code for holoviews.element.selection

"""
Defines mix-in classes to handle support for linked brushing on
elements.
"""

import sys
from importlib.util import find_spec

import numpy as np
import pandas as pd

from ..core import Dataset, NdOverlay, util
from ..streams import Lasso, Selection1D, SelectionXY
from ..util.transform import dim
from .annotation import HSpan, VSpan


class SelectionIndexExpr:

    _selection_dims = None

    _selection_streams = (Selection1D,)

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self._index_skip = False

    def _empty_region(self):
        return None

    def _get_index_selection(self, index, index_cols):
        self._index_skip = True
        if not index:
            return None, None, None
        ds = self.clone(kdims=index_cols, new_type=Dataset)
        if len(index_cols) == 1:
            index_dim = index_cols[0]
            vals = dim(index_dim).apply(ds.iloc[index], expanded=False)
            if vals.dtype.kind == 'O' and all(isinstance(v, np.ndarray) for v in vals):
                vals = [v for arr in vals for v in util.unique_iterator(arr)]
            expr = dim(index_dim).isin(list(util.unique_iterator(vals)))
        else:
            get_shape = dim(self.dataset.get_dimension(index_cols[0]), np.shape)
            index_cols = [dim(self.dataset.get_dimension(c), np.ravel) for c in index_cols]
            vals = dim(index_cols[0], util.unique_zip, *index_cols[1:]).apply(
                ds.iloc[index], expanded=True, flat=True
            )
            contains = dim(index_cols[0], util.lzip, *index_cols[1:]).isin(vals, object=True)
            expr = dim(contains, np.reshape, get_shape)
        return expr, None, None

    def _get_selection_expr_for_stream_value(self, **kwargs):
        index = kwargs.get('index')
        index_cols = kwargs.get('index_cols')
        if index is None or index_cols is None:
            return None, None, None
        return self._get_index_selection(index, index_cols)

    @staticmethod
    def _merge_regions(region1, region2, operation):
        return None


def spatial_select_gridded(xvals, yvals, geometry):
    rectilinear = (np.diff(xvals, axis=0) == 0).all()
    if rectilinear:
        from .path import Polygons
        from .raster import Image
        try:
            from ..operation.datashader import rasterize
        except ImportError:
            raise ImportError("Lasso selection on gridded data requires "
                              "datashader to be available.") from None
        xs, ys = xvals[0], yvals[:, 0]
        target = Image((xs, ys, np.empty(ys.shape+xs.shape)))
        poly = Polygons([geometry])
        sel_mask = rasterize(poly, target=target, dynamic=False, aggregator='any')
        return sel_mask.dimension_values(2, flat=False)
    else:
        sel_mask = spatial_select_columnar(xvals.flatten(), yvals.flatten(), geometry)
        return sel_mask.reshape(xvals.shape)

def spatial_select_columnar(xvals, yvals, geometry, geom_method=None):
    if 'cudf' in sys.modules:
        import cudf
        if isinstance(xvals, cudf.Series):
            xvals = xvals.values.astype('float')
            yvals = yvals.values.astype('float')
            try:
                import cuspatial
                result = cuspatial.point_in_polygon(
                    xvals,
                    yvals,
                    cudf.Series([0], index=["selection"]),
                    [0],
                    geometry[:, 0],
                    geometry[:, 1],
                )
                return result.values
            except ImportError:
                xvals = np.asarray(xvals)
                yvals = np.asarray(yvals)
    if 'dask' in sys.modules:
        import dask.dataframe as dd
        if isinstance(xvals, dd.Series):
            try:
                xvals.name = "xvals"
                yvals.name = "yvals"
                df = xvals.to_frame().join(yvals)
                return df.map_partitions(
                    lambda df, geometry: spatial_select_columnar(df.xvals, df.yvals, geometry),
                    geometry,
                    meta=pd.Series(dtype=bool)
                )
            except Exception:
                xvals = np.asarray(xvals)
                yvals = np.asarray(yvals)
    x0, x1 = geometry[:, 0].min(), geometry[:, 0].max()
    y0, y1 = geometry[:, 1].min(), geometry[:, 1].max()
    sel_mask = (xvals>=x0) & (xvals<=x1) & (yvals>=y0) & (yvals<=y1)
    masked_xvals = xvals[sel_mask]
    masked_yvals = yvals[sel_mask]
    if geom_method is None:
        if find_spec("spatialpandas") is not None:
            geom_method = "spatialpandas"
        elif find_spec("shapely") is not None:
            geom_method = "shapely"
        else:
            msg = "Lasso selection on tabular data requires either spatialpandas or shapely to be available."
            raise ImportError(msg) from None
    geom_function = {"spatialpandas": _mask_spatialpandas, "shapely": _mask_shapely}[geom_method]
    geom_mask = geom_function(masked_xvals, masked_yvals, geometry)
    if isinstance(xvals, pd.Series):
        sel_mask[sel_mask.index[np.where(sel_mask)[0]]] = geom_mask
    else:
        sel_mask[np.where(sel_mask)[0]] = geom_mask
    return sel_mask


def _mask_spatialpandas(masked_xvals, masked_yvals, geometry):
    from spatialpandas.geometry import PointArray, Polygon
    points = PointArray((masked_xvals.astype('float'), masked_yvals.astype('float')))
    poly = Polygon([np.concatenate([geometry, geometry[:1]]).flatten()])
    return points.intersects(poly)


def _mask_shapely(masked_xvals, masked_yvals, geometry):
    from shapely.geometry import Point, Polygon
    points = (Point(x, y) for x, y in zip(masked_xvals, masked_yvals))
    poly = Polygon(geometry)
    return np.array([poly.contains(p) for p in points], dtype=bool)


def spatial_select(xvals, yvals, geometry):
    if xvals.ndim > 1:
        return spatial_select_gridded(xvals, yvals, geometry)
    else:
        return spatial_select_columnar(xvals, yvals, geometry)

def spatial_geom_select(x0vals, y0vals, x1vals, y1vals, geometry):
    try:
        from shapely.geometry import Polygon, box
        boxes = (box(x0, y0, x1, y1) for x0, y0, x1, y1 in
                 zip(x0vals, y0vals, x1vals, y1vals))
        poly = Polygon(geometry)
        return np.array([poly.contains(p) for p in boxes])
    except ImportError:
        raise ImportError("Lasso selection on geometry data requires "
                          "shapely to be available.") from None

def spatial_poly_select(xvals, yvals, geometry):
    try:
        from shapely.geometry import Polygon
        boxes = (Polygon(np.column_stack([xs, ys])) for xs, ys in zip(xvals, yvals))
        poly = Polygon(geometry)
        return np.array([poly.contains(p) for p in boxes])
    except ImportError:
        raise ImportError("Lasso selection on geometry data requires "
                          "shapely to be available.") from None

def spatial_bounds_select(xvals, yvals, bounds):
    x0, y0, x1, y1 = bounds
    return np.array([((x0<=np.nanmin(xs)) & (y0<=np.nanmin(ys)) &
                      (x1>=np.nanmax(xs)) & (y1>=np.nanmax(ys)))
                     for xs, ys in zip(xvals, yvals)])


[docs]class Selection2DExpr(SelectionIndexExpr): """ Mixin class for Cartesian 2D elements to add basic support for SelectionExpr streams. """ _selection_dims = 2 _selection_streams = (SelectionXY, Lasso, Selection1D) def _empty_region(self): from .geom import Rectangles from .path import Path return Rectangles([]) * Path([]) def _get_selection(self, **kwargs): xcats, ycats = None, None x0, y0, x1, y1 = kwargs['bounds'] if 'x_selection' in kwargs: xsel = kwargs['x_selection'] if isinstance(xsel, list): xcats = xsel x0, x1 = int(round(x0)), int(round(x1)) ysel = kwargs['y_selection'] if isinstance(ysel, list): ycats = ysel y0, y1 = int(round(y0)), int(round(y1)) # Handle invert_xaxis/invert_yaxis if x0 > x1: x0, x1 = x1, x0 if y0 > y1: y0, y1 = y1, y0 return (x0, x1), xcats, (y0, y1), ycats def _get_index_expr(self, index_cols, sel): if len(index_cols) == 1: index_dim = index_cols[0] vals = dim(index_dim).apply(sel, expanded=False, flat=True) expr = dim(index_dim).isin(list(util.unique_iterator(vals))) else: get_shape = dim(self.dataset.get_dimension(), np.shape) index_cols = [dim(self.dataset.get_dimension(c), np.ravel) for c in index_cols] vals = dim(index_cols[0], util.unique_zip, *index_cols[1:]).apply( sel, expanded=True, flat=True ) contains = dim(index_cols[0], util.lzip, *index_cols[1:]).isin(vals, object=True) expr = dim(contains, np.reshape, get_shape) return expr def _get_bounds_selection(self, xdim, ydim, **kwargs): from .geom import Rectangles (x0, x1), xcats, (y0, y1), ycats = self._get_selection(**kwargs) xsel = xcats or (x0, x1) ysel = ycats or (y0, y1) bbox = {xdim.name: xsel, ydim.name: ysel} index_cols = kwargs.get('index_cols') if index_cols: selection = self.dataset.clone(datatype=['dataframe', 'dictionary']).select(**bbox) selection_expr = self._get_index_expr(index_cols, selection) region_element = None else: if xcats: xexpr = dim(xdim).isin(xcats) else: xexpr = (dim(xdim) >= x0) & (dim(xdim) <= x1) if ycats: yexpr = dim(ydim).isin(ycats) else: yexpr = (dim(ydim) >= y0) & (dim(ydim) <= y1) selection_expr = (xexpr & yexpr) region_element = Rectangles([(x0, y0, x1, y1)]) return selection_expr, bbox, region_element def _get_lasso_selection(self, xdim, ydim, geometry, **kwargs): from .path import Path bbox = {xdim.name: geometry[:, 0], ydim.name: geometry[:, 1]} expr = dim.pipe(spatial_select, xdim, dim(ydim), geometry=geometry) index_cols = kwargs.get('index_cols') if index_cols: selection = self[expr.apply(self)] selection_expr = self._get_index_expr(index_cols, selection) return selection_expr, bbox, None return expr, bbox, Path([np.concatenate([geometry, geometry[:1]])]) def _get_selection_dims(self): from .graphs import Graph if isinstance(self, Graph): xdim, ydim = self.nodes.dimensions()[:2] else: xdim, ydim = self.dimensions()[:2] invert_axes = self.opts.get('plot').kwargs.get('invert_axes', False) if invert_axes: xdim, ydim = ydim, xdim return (xdim, ydim) def _skip(self, **kwargs): skip = kwargs.get('index_cols') and self._index_skip if skip: self._index_skip = False return skip def _get_selection_expr_for_stream_value(self, **kwargs): from .geom import Rectangles from .path import Path if (kwargs.get('bounds') is None and kwargs.get('x_selection') is None and kwargs.get('geometry') is None and not kwargs.get('index')): return None, None, Rectangles([]) * Path([]) index_cols = kwargs.get('index_cols') dims = self._get_selection_dims() if kwargs.get('index') is not None and index_cols is not None: expr, _, _ = self._get_index_selection(kwargs['index'], index_cols) return expr, None, self._empty_region() elif self._skip(**kwargs): return None elif 'bounds' in kwargs: expr, bbox, region = self._get_bounds_selection(*dims, **kwargs) return expr, bbox, None if region is None else region * Path([]) elif 'geometry' in kwargs: expr, bbox, region = self._get_lasso_selection(*dims, **kwargs) return expr, bbox, None if region is None else Rectangles([]) * region @staticmethod def _merge_regions(region1, region2, operation): if region1 is None or operation == "overwrite": return region2 rect1 = region1.get(0) rect2 = region2.get(0) rects = rect1.clone(rect1.interface.concatenate([rect1, rect2])) poly1 = region1.get(1) poly2 = region2.get(1) polys = poly1.clone([poly1, poly2]) return rects * polys
[docs]class SelectionGeomExpr(Selection2DExpr): def _get_selection_dims(self): x0dim, y0dim, x1dim, y1dim = self.kdims invert_axes = self.opts.get('plot').kwargs.get('invert_axes', False) if invert_axes: x0dim, x1dim, y0dim, y1dim = y0dim, y1dim, x0dim, x1dim return (x0dim, y0dim, x1dim, y1dim) def _get_bounds_selection(self, x0dim, y0dim, x1dim, y1dim, **kwargs): from .geom import Rectangles (x0, x1), xcats, (y0, y1), ycats = self._get_selection(**kwargs) xsel = xcats or (x0, x1) ysel = ycats or (y0, y1) bbox = {x0dim.name: xsel, y0dim.name: ysel, x1dim.name: xsel, y1dim.name: ysel} index_cols = kwargs.get('index_cols') if index_cols: selection = self.dataset.clone(datatype=['dataframe', 'dictionary']).select(**bbox) selection_expr = self._get_index_expr(index_cols, selection) region_element = None else: x0expr = (dim(x0dim) >= x0) & (dim(x0dim) <= x1) y0expr = (dim(y0dim) >= y0) & (dim(y0dim) <= y1) x1expr = (dim(x1dim) >= x0) & (dim(x1dim) <= x1) y1expr = (dim(y1dim) >= y0) & (dim(y1dim) <= y1) selection_expr = (x0expr & y0expr & x1expr & y1expr) region_element = Rectangles([(x0, y0, x1, y1)]) return selection_expr, bbox, region_element def _get_lasso_selection(self, x0dim, y0dim, x1dim, y1dim, geometry, **kwargs): from .path import Path bbox = { x0dim.name: geometry[:, 0], y0dim.name: geometry[:, 1], x1dim.name: geometry[:, 0], y1dim.name: geometry[:, 1] } expr = dim.pipe(spatial_geom_select, x0dim, dim(y0dim), dim(x1dim), dim(y1dim), geometry=geometry) index_cols = kwargs.get('index_cols') if index_cols: selection = self[expr.apply(self)] selection_expr = self._get_index_expr(index_cols, selection) return selection_expr, bbox, None return expr, bbox, Path([np.concatenate([geometry, geometry[:1]])])
[docs]class SelectionPolyExpr(Selection2DExpr): def _skip(self, **kwargs): """ Do not skip geometry selections until polygons support returning indexes on lasso based selections. """ skip = kwargs.get('index_cols') and self._index_skip and 'geometry' not in kwargs if skip: self._index_skip = False return skip def _get_bounds_selection(self, xdim, ydim, **kwargs): from .geom import Rectangles (x0, x1), _, (y0, y1), _ = self._get_selection(**kwargs) bbox = {xdim.name: (x0, x1), ydim.name: (y0, y1)} index_cols = kwargs.get('index_cols') expr = dim.pipe(spatial_bounds_select, xdim, dim(ydim), bounds=(x0, y0, x1, y1)) if index_cols: selection = self[expr.apply(self, expanded=False)] selection_expr = self._get_index_expr(index_cols, selection) return selection_expr, bbox, None return expr, bbox, Rectangles([(x0, y0, x1, y1)]) def _get_lasso_selection(self, xdim, ydim, geometry, **kwargs): from .path import Path bbox = {xdim.name: geometry[:, 0], ydim.name: geometry[:, 1]} expr = dim.pipe(spatial_poly_select, xdim, dim(ydim), geometry=geometry) index_cols = kwargs.get('index_cols') if index_cols: selection = self[expr.apply(self, expanded=False)] selection_expr = self._get_index_expr(index_cols, selection) return selection_expr, bbox, None return expr, bbox, Path([np.concatenate([geometry, geometry[:1]])])
[docs]class Selection1DExpr(Selection2DExpr): """ Mixin class for Cartesian 1D Chart elements to add basic support for SelectionExpr streams. """ _selection_dims = 1 _inverted_expr = False _selection_streams = (SelectionXY,) def _empty_region(self): invert_axes = self.opts.get('plot').kwargs.get('invert_axes', False) if ((invert_axes and not self._inverted_expr) or (not invert_axes and self._inverted_expr)): region_el = HSpan else: region_el = VSpan return NdOverlay({0: region_el()}) def _get_selection_expr_for_stream_value(self, **kwargs): invert_axes = self.opts.get('plot').kwargs.get('invert_axes', False) if ((invert_axes and not self._inverted_expr) or (not invert_axes and self._inverted_expr)): region_el = HSpan else: region_el = VSpan if kwargs.get('bounds', None) is None: region = None if 'index_cols' in kwargs else NdOverlay({0: region_el()}) return None, None, region x0, y0, x1, y1 = kwargs['bounds'] # Handle invert_xaxis/invert_yaxis if y0 > y1: y0, y1 = y1, y0 if x0 > x1: x0, x1 = x1, x0 if len(self.dimensions()) == 1: xdim = self.dimensions()[0] ydim = None else: xdim, ydim = self.dimensions()[:2] if invert_axes: x0, x1, y0, y1 = y0, y1, x0, x1 cat_kwarg = 'y_selection' else: cat_kwarg = 'x_selection' if self._inverted_expr: if ydim is not None: xdim = ydim x0, x1 = y0, y1 cat_kwarg = ('y' if invert_axes else 'x') + '_selection' cats = kwargs.get(cat_kwarg) bbox = {xdim.name: (x0, x1)} if cats is not None and len(self.kdims) == 1: bbox[self.kdims[0].name] = cats index_cols = kwargs.get('index_cols') if index_cols: selection = self.dataset.clone(datatype=['dataframe', 'dictionary']).select(**bbox) selection_expr = self._get_index_expr(index_cols, selection) region_element = None else: if isinstance(cats, list) and xdim in self.kdims[:1]: selection_expr = dim(xdim).isin(cats) else: selection_expr = ((dim(xdim) >= x0) & (dim(xdim) <= x1)) if isinstance(cats, list) and len(self.kdims) == 1: selection_expr &= dim(self.kdims[0]).isin(cats) region_element = NdOverlay({0: region_el(x0, x1)}) return selection_expr, bbox, region_element @staticmethod def _merge_regions(region1, region2, operation): if region1 is None or operation == "overwrite": return region2 data = [d.data for d in region1] + [d.data for d in region2] prev = len(data) new = None while prev != new: prev = len(data) contiguous = [] for l, u in data: if not util.isfinite(l) or not util.isfinite(u): continue overlap = False for i, (pl, pu) in enumerate(contiguous): if l >= pl and l <= pu: pu = max(u, pu) overlap = True elif u <= pu and u >= pl: pl = min(l, pl) overlap = True if overlap: contiguous[i] = (pl, pu) if not overlap: contiguous.append((l, u)) new = len(contiguous) data = contiguous return NdOverlay([(i, region1.last.clone(l, u)) for i, (l, u) in enumerate(data)])