Export Mesh Data to a DataFrame

Convert a mesh’s point_data or cell_data to a pandas.DataFrame or pyarrow.Table for downstream analytics, export, or interactive exploration in IDEs like Positron’s Data Explorer, JupyterLab, or VS Code’s data viewer.

This example uses a classic CFD-style workflow: a scalar field and a vector field attached to a mesh, then filtered / aggregated / exported using pandas idioms.

import numpy as np
from pyvista import examples

Load a mesh and attach some data

Use the hexbeam fixture mesh and attach a pressure scalar plus a velocity vector.

mesh = examples.load_hexbeam()
rng = np.random.default_rng(seed=0)
mesh.point_data['pressure'] = rng.normal(loc=100.0, scale=20.0, size=mesh.n_points)
mesh.point_data['velocity'] = rng.normal(size=(mesh.n_points, 3))

Convert to a pandas.DataFrame

mesh.to_pandas() returns one row per point. Multi-component arrays are expanded to one column per component, named {array_name}_{i}.

df = mesh.to_pandas()
df.head()

	sample_point_scalars	VTKorigID	pressure	velocity_0	velocity_1	velocity_2
0	1	0	102.514604	-1.401520	0.502683	0.989713
1	2	1	97.357903	-0.164295	-1.074365	0.873042
2	4	2	112.808453	-1.280394	-0.713068	0.621018
3	6	3	102.098002	-2.250141	0.386370	-0.581641
4	8	4	89.286613	0.109280	-0.075702	0.202114

The DataFrame has n_points rows and one column per (expanded) array.

df.shape

(99, 6)

Ad-hoc analytics

Since we have a DataFrame, the full pandas API is available: filter, derive new columns, summarize.

velocity = df[['velocity_0', 'velocity_1', 'velocity_2']].to_numpy()
df['speed'] = np.linalg.norm(velocity, axis=1)
df[['pressure', 'speed']].describe()

	pressure	speed
count	99.000000	99.000000
mean	101.921452	1.612009
std	19.203438	0.670339
min	53.499385	0.241915
25%	87.435314	1.174939
50%	101.880246	1.521954
75%	114.773541	1.927140
max	140.047852	3.781354

Find the 5 highest-pressure points.

df.nlargest(5, 'pressure')[['pressure', 'speed']]

	pressure	speed
79	140.047852	1.633129
47	139.205166	1.591161
74	136.440227	1.271196
48	136.032697	1.838132
62	134.787358	1.471332

Cell-level export

Pass association='cell' to export cell_data instead. The result has n_cells rows.

mesh.cell_data['quality'] = rng.random(mesh.n_cells)
cell_df = mesh.to_pandas('cell')
cell_df.head()

	sample_cell_scalars	quality
0	1	0.322921
1	2	0.990745
2	3	0.264695
3	4	0.830694
4	5	0.173114

Export to disk

A DataFrame gives you one-liner access to every pandas I/O backend: Parquet, CSV, Feather, Excel, SQL, HDF5, and more. Commented out here to keep the gallery build clean.

# df.to_parquet('point_data.parquet')
# df.to_csv('point_data.csv', index=False)

Zero-copy Arrow interchange

mesh, mesh.point_data, mesh.cell_data, and pyvista.Table all implement the Arrow PyCapsule interface, so any Arrow-aware library (polars, DuckDB, ibis, narwhals) can consume them without pyvista depending on those libraries. For example, with pyarrow:

import pyarrow as pa

arrow_table = pa.table(mesh)
arrow_table.schema

sample_point_scalars: int64
VTKorigID: int64
pressure: double
velocity_0: double
velocity_1: double
velocity_2: double

The mesh, mesh.point_data, and mesh.cell_data objects can also be opened directly in data-science IDE variable explorers (Positron, Jupyter, VS Code) after calling to_pandas(). The returned DataFrame renders as an interactive, sortable, filterable table.