| 1 |
""" |
| 2 |
.. _interpolate_sample_example: |
| 3 |
|
| 4 |
Compare interpolation/sampling methods |
| 5 |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 6 |
|
| 7 |
There are two main methods of interpolating or sampling data from a target mesh |
| 8 |
in PyVista. :func:`pyvista.DataSetFilters.interpolate` uses a distance weighting |
| 9 |
kernel to interpolate point data from nearby points of the target mesh onto |
| 10 |
the desired points. |
| 11 |
:func:`pyvista.DataObjectFilters.sample` interpolates data using the |
| 12 |
interpolation scheme of the enclosing cell from the target mesh. |
| 13 |
|
| 14 |
If the target mesh is a point cloud, i.e. there is no connectivity in the cell |
| 15 |
structure, then :func:`pyvista.DataSetFilters.interpolate` is typically |
| 16 |
preferred. If interpolation is desired within the cells of the target mesh, then |
| 17 |
:func:`pyvista.DataObjectFilters.sample` is typically desired. |
| 18 |
|
| 19 |
Here the two methods are compared and contrasted using a simple example of |
| 20 |
sampling data from a mesh in a rectangular domain. This example demonstrates the |
| 21 |
main differences above. For more complex uses, see :ref:`interpolate_example` |
| 22 |
and :ref:`resampling_example`. |
| 23 |
|
| 24 |
""" |
| 25 |
|
| 26 |
# sphinx_gallery_thumbnail_number = 7 |
| 27 |
import numpy as np |
| 28 |
import pyvista as pv |
| 29 |
|
| 30 |
# %% |
| 31 |
# Interpolating from point cloud |
| 32 |
# ++++++++++++++++++++++++++++++ |
| 33 |
# A point cloud is a collection of points that have no connectivity in |
| 34 |
# the mesh, i.e. the mesh contains no cells or the cells are 0D |
| 35 |
# (vertex or polyvertex). The filter :func:`pyvista.DataSetFilters.interpolate` |
| 36 |
# uses a distance-based weighting methodology to interpolate between the |
| 37 |
# unconnected points. |
| 38 |
# |
| 39 |
# First, generate a point cloud mesh in a rectangular domain from |
| 40 |
# ``(0, 0)`` to ``(3, 1)``. The data to be sampled is the square of the y position. |
| 41 |
|
| 42 |
rng = np.random.default_rng(seed=0) |
| 43 |
points = rng.uniform(low=[0, 0], high=[3, 1], size=(250, 2)) |
| 44 |
# Make points be z=0 |
| 45 |
points = np.hstack((points, np.zeros((250, 1)))) |
| 46 |
point_mesh = pv.PolyData(points) |
| 47 |
point_mesh['ysquared'] = points[:, 1] ** 2 |
| 48 |
|
| 49 |
# %% |
| 50 |
# The point cloud data looks like this. |
| 51 |
|
| 52 |
pl = pv.Plotter() |
| 53 |
pl.add_mesh(point_mesh, render_points_as_spheres=True, point_size=10) |
| 54 |
pl.view_xy() |
| 55 |
pl.show() |
| 56 |
|
| 57 |
# %% |
| 58 |
# Now estimate data on a regular grid from the point data. Note |
| 59 |
# that the distance parameter ``radius`` determines how far away to |
| 60 |
# look for point cloud data. |
| 61 |
|
| 62 |
grid = pv.ImageData(dimensions=(11, 11, 1), spacing=[3 / 10, 1 / 10, 1]) |
| 63 |
output = grid.interpolate(point_mesh, radius=0.1, null_value=-1) |
| 64 |
output |
| 65 |
|
| 66 |
# %% |
| 67 |
# When using ``radius=0.1``, the expected extents of the data are |
| 68 |
# captured reasonably well over the domain, but there are holes in the |
| 69 |
# data (represented by the darkest blue colors) caused by no points within |
| 70 |
# the ``radius`` to interpolate from. |
| 71 |
|
| 72 |
pl = pv.Plotter() |
| 73 |
pl.add_mesh(output, clim=[0, 1]) |
| 74 |
pl.add_mesh(points, render_points_as_spheres=True, point_size=10, color='red') |
| 75 |
pl.view_xy() |
| 76 |
pl.show() |
| 77 |
|
| 78 |
|
| 79 |
# %% |
| 80 |
# Now repeat with ``radius=0.25``. |
| 81 |
# There are no holes but the extents of the data is much narrower |
| 82 |
# than ``[0, 1]``. This is caused by more interior points involved |
| 83 |
# in the weighting near the lower and upper edges of the domain. |
| 84 |
# Other parameters such as ``sharpness`` could be tuned to try to |
| 85 |
# lessen the issue. |
| 86 |
|
| 87 |
grid = pv.ImageData(dimensions=(11, 11, 1), spacing=[3 / 10, 1 / 10, 1]) |
| 88 |
output = grid.interpolate(point_mesh, radius=0.25, null_value=-1) |
| 89 |
|
| 90 |
pl = pv.Plotter() |
| 91 |
pl.add_mesh(output, clim=[0, 1]) |
| 92 |
pl.add_mesh(points, render_points_as_spheres=True, point_size=10, color='red') |
| 93 |
pl.view_xy() |
| 94 |
pl.show() |
| 95 |
|
| 96 |
# %% |
| 97 |
# While this filter is very useful for point clouds, it is possible to use |
| 98 |
# it to interpolate from the points on other mesh types. With |
| 99 |
# unstuitable choice of ``radius`` the interpolation doesn't look very good. |
| 100 |
# It is recommended consider using :func:`pyvista.DataObjectFilters.sample` in a |
| 101 |
# case like this (see next section below). However, there may be cases with |
| 102 |
# non-point cloud meshes where :func:`pyvista.DataSetFilters.interpolate` is |
| 103 |
# still preferred. |
| 104 |
|
| 105 |
sphere = pv.SolidSphere(center=(0.5, 0.5, 0), outer_radius=1.0) |
| 106 |
sphere['ysquared'] = sphere.points[:, 1] ** 2 |
| 107 |
output = grid.interpolate(sphere, radius=0.1) |
| 108 |
|
| 109 |
pl = pv.Plotter() |
| 110 |
pl.add_mesh(output, clim=[0, 1]) |
| 111 |
pl.add_mesh(sphere, style='wireframe', color='white') |
| 112 |
pl.view_xy() |
| 113 |
pl.show() |
| 114 |
|
| 115 |
# %% |
| 116 |
# Sampling from a mesh with connectivity |
| 117 |
# ++++++++++++++++++++++++++++++++++++++ |
| 118 |
# This example is in many ways the opposite of the prior one. |
| 119 |
# A mesh with cell connectivity that spans 2 dimensions is |
| 120 |
# sampled at discrete points using :func:`pyvista.DataObjectFilters.sample`. |
| 121 |
# Importantly, the cell connectivity enables direct interpolation |
| 122 |
# inside the domain without needing distance or weighting parametization. |
| 123 |
# |
| 124 |
# First, show that sample does not work with point clouds with data. |
| 125 |
# Either :func:`pyvista.DataSetFilters.interpolate` or the |
| 126 |
# ``snap_to_closest_point`` parameter must be used. |
| 127 |
|
| 128 |
grid = pv.ImageData(dimensions=(11, 11, 1), spacing=[3 / 10, 1 / 10, 1]) |
| 129 |
output = grid.sample(point_mesh) |
| 130 |
# value of (0, 0) shows that no data was sampled |
| 131 |
print(f'(min, max): {output["ysquared"].min()}, {output["ysquared"].min()}') |
| 132 |
|
| 133 |
# %% |
| 134 |
# Create the non-point cloud mesh that will be sampled from and plot it. |
| 135 |
|
| 136 |
grid = pv.ImageData(dimensions=(11, 11, 1), spacing=[3 / 10, 1 / 10, 1]) |
| 137 |
grid['ysquared'] = grid.points[:, 1] ** 2 |
| 138 |
|
| 139 |
pl = pv.Plotter() |
| 140 |
pl.add_mesh(grid, clim=[0, 1]) |
| 141 |
pl.view_xy() |
| 142 |
pl.show() |
| 143 |
|
| 144 |
# %% |
| 145 |
# Now sample it at the discrete points used in the first example. |
| 146 |
point_mesh = pv.PolyData(points) |
| 147 |
output = point_mesh.sample(grid) |
| 148 |
output |
| 149 |
|
| 150 |
# %% |
| 151 |
# This looks identical to the first plot of the first example as the |
| 152 |
# data is not noisy, and there is little interpolation error. |
| 153 |
|
| 154 |
pl = pv.Plotter() |
| 155 |
pl.add_mesh(output, render_points_as_spheres=True, point_size=10) |
| 156 |
pl.view_xy() |
| 157 |
pl.show() |
| 158 |
|
| 159 |
# %% |
| 160 |
# Instead of sampling onto a point cloud, :func:`pyvista.DataObjectFilters.sample` |
| 161 |
# can sample using other mesh types. For example, sampling onto a rotated subset |
| 162 |
# of the grid. |
| 163 |
# |
| 164 |
# Make subset (0.7, 0.7, 0) units in dimension and then rotate by 45 degrees around |
| 165 |
# its center. |
| 166 |
subset = pv.ImageData(dimensions=(8, 8, 1), spacing=[0.1, 0.1, 0], origin=(0.15, 0.15, 0)) |
| 167 |
rotated_subset = subset.rotate_vector(vector=(0, 0, 1), angle=45, point=(0.5, 0.5, 0)) |
| 168 |
output = rotated_subset.sample(grid) |
| 169 |
output |
| 170 |
|
| 171 |
# %% |
| 172 |
# The data in the sampled region looks identical to the original grid |
| 173 |
# due to the well-behaved nature of the data and |
| 174 |
# low interpolation error. |
| 175 |
|
| 176 |
pl = pv.Plotter() |
| 177 |
pl.add_mesh(grid, style='wireframe', clim=[0, 1]) |
| 178 |
pl.add_mesh(output, clim=[0, 1]) |
| 179 |
pl.view_xy() |
| 180 |
pl.show() |
| 181 |
|
| 182 |
# %% |
| 183 |
# Repeat the sphere interpolation example, but using |
| 184 |
# :func:`pyvista.DataObjectFilters.sample`. This method |
| 185 |
# is directly able to sample from the mesh in this case without |
| 186 |
# fiddling with distance weighting parameters. |
| 187 |
|
| 188 |
sphere = pv.SolidSphere(center=(0.5, 0.5, 0), outer_radius=1.0) |
| 189 |
sphere['ysquared'] = sphere.points[:, 1] ** 2 |
| 190 |
output = grid.sample(sphere) |
| 191 |
|
| 192 |
pl = pv.Plotter() |
| 193 |
pl.add_mesh(output, clim=[0, 1]) |
| 194 |
pl.add_mesh(sphere, style='wireframe', color='white') |
| 195 |
pl.view_xy() |
| 196 |
pl.show() |
| 197 |
|
| 198 |
|
| 199 |
# %% |
| 200 |
# .. tags:: filter |