Add performance doc

doc/python/webgl-vs-svg.md → doc/python/performance.md

@@ -6,7 +6,7 @@ jupyter:
       extension: .md
       format_name: markdown
       format_version: '1.3'
-      jupytext_version: 1.16.1
+      jupytext_version: 1.16.4
   kernelspec:
     display_name: Python 3 (ipykernel)
     language: python
@@ -20,23 +20,28 @@ jupyter:
     name: python
     nbconvert_exporter: python
     pygments_lexer: ipython3
-    version: 3.10.11
+    version: 3.11.10
   plotly:
-    description: Using WebGL for increased speed, improved interactivity, and the
-      ability to plot even more data!
+    description: Recommendations for increased speed, improved interactivity, and
+      the ability to plot even more data!
     display_as: basic
     language: python
     layout: base
-    name: WebGL vs SVG
+    name: High Performance Visualization
     order: 14
-    permalink: python/webgl-vs-svg/
-    redirect_from: python/compare-webgl-svg/
+    permalink: python/performance/
+    redirect_from: python/webgl-vs-svg/
     thumbnail: thumbnail/webgl.jpg
 ---
 
-### SVG and canvas/WebGL: two browser capabilities for rendering
+## WebGL
 
-`plotly` figures are rendered by web browsers, which broadly speaking have two families of capabilities for rendering graphics: the SVG API which supports vector rendering, and the Canvas API which supports raster rendering, and can exploit GPU hardware acceleration via a browser technology known as WebGL. Each `plotly` trace type is primarily rendered with either SVG or WebGL, although WebGL-powered traces also use some SVG. The following trace types use WebGL for part or all of the rendering:
+`plotly` figures are rendered by web browsers, which broadly speaking have two families of capabilities for rendering graphics:
+
+- The SVG API, which supports vector rendering.
+- The Canvas API, which supports raster rendering, and can exploit GPU hardware acceleration via a browser technology known as WebGL.
+
+Each `plotly` trace type is rendered with either SVG or WebGL. The following trace types use WebGL for rendering:
 
 * Accelerated versions of SVG trace types: `scattergl`, `scatterpolargl`,
 * High-performance multidimensional trace types: `splom`, or `parcoords`
@@ -47,18 +52,18 @@ jupyter:
 
 WebGL is a powerful technology for accelerating computation but comes with some strict limitations:
 
-1. GPU requirement: WebGL is a GPU (graphics card) technology and therefore requires specific hardware which is available in most but not all cases and is supported by most but not all browsers
+1. GPU requirement: WebGL is a GPU (graphics card) technology and therefore requires specific hardware which is available in most but not all cases and is supported by most but not all browsers.
 2. Rasterization: WebGL-rendered data is drawn as a grid of pixels rather than as individual shapes, so can appear pixelated or fuzz in certain cases, and when exported to static file formats will appear pixelated on zoom. In addition: text rendering will differ between SVG and WebGL-powered traces.
-3. Context limits: browsers impose a strict limit on the number of WebGL "contexts" that any given web document can access. WebGL-powered traces in `plotly` can use multiple contexts in some cases but as a general rule, **it may not be possible to render more than 8 WebGL-involving figures on the same page at the same time.**
-4. Size limits: browsers impose hardware-dependent limits on the height and width of figures using WebGL which users may encounter with extremely large plots (e.g. tens of thousands of pixels of height)
+3. Context limits: browsers impose a strict limit on the number of WebGL "contexts" that any given web document can access. WebGL-powered traces in `plotly` can use multiple contexts in some cases but as a general rule, **it may not be possible to render more than 8 WebGL-involving figures on the same page at the same time.** See the following section, Multiple WebGL Contexts, for more details.
+4. Size limits: browsers impose hardware-dependent limits on the height and width of figures using WebGL which users may encounter with extremely large plots (e.g. tens of thousands of pixels of height).
 
 In addition to the above limitations, the WebGL-powered version of certain SVG-powered trace types (`scattergl`, `scatterpolargl`) are not complete drop-in replacements for their SVG counterparts yet
-* Available symbols will differ
-* Area fills are not yet supported in WebGL
-* Range breaks on time-series axes are not yet supported
-* Axis range heuristics may differ
+* Available symbols will differ.
+* Area fills are not yet supported in WebGL.
+* Range breaks on time-series axes are not yet supported.
+* Axis range heuristics may differ.
 
-### Multiple WebGL Contexts
+#### Multiple WebGL Contexts
 
 *New in 5.19*
 
@@ -72,12 +77,12 @@ To use it, in the environment where your Plotly figures are being rendered, load
 
 In a Jupyter notebook environment that supports magic commands, you can load it with the [HTML magic command](https://ipython.readthedocs.io/en/stable/interactive/magics.html#cellmagic-html):
 
-<!-- #region -->
+
 ```
 %%html
 <script src=“https://unpkg.com/virtual-webgl@1.0.6/src/virtual-webgl.js”></script>
 ```
-<!-- #endregion -->
+
 
 ### WebGL for Scatter Performance
 
@@ -87,9 +92,9 @@ it is also possible to use [datashader](/python/datashader/).
 
 ### WebGL with Plotly Express
 
-The `rendermode` argument to supported Plotly Express functions (e.g. `scatter` and `scatter_polar`) can be used to enable WebGL rendering.
+The `render_mode` argument to supported Plotly Express functions (e.g. `scatter` and `scatter_polar`) can be used to enable WebGL rendering.
 
-> **Note** The default `rendermode` is `"auto"`, in which case Plotly Express will automatically set `rendermode="webgl"` if the input data is more than 1,000 rows long. If WebGL acceleration is *not* desired in this case, `rendermode` can be forced to `"svg"` for vectorized, if slower, rendering.
+> **Note** The default `render_mode` is `"auto"`, in which case Plotly Express will automatically set `render_mode="webgl"` if the input data is more than 1,000 rows long. In this case, WebGl can be disabled by setting `render_mode=svg`.
 
 Here is an example that creates a 100,000 point scatter plot using Plotly Express with WebGL rendering explicitly enabled.
 
@@ -98,6 +103,7 @@ import plotly.express as px
 
 import pandas as pd
 import numpy as np
+
 np.random.seed(1)
 
 N = 100000
@@ -113,7 +119,7 @@ fig.show()
 ```
 
 
-#### WebGL with 100,000 points with Graph Objects
+#### WebGL with 1,000,000 points with Graph Objects
 
 If Plotly Express does not provide a good starting point, it is also possible to use [the more generic `go.Scattergl` class from `plotly.graph_objects`](/python/graph-objects/).
 
@@ -122,9 +128,8 @@ import plotly.graph_objects as go
 
 import numpy as np
 
-N = 100000
+N = 1_000_000
 
-# Create figure
 fig = go.Figure()
 
 fig.add_trace(
@@ -143,58 +148,69 @@ fig.add_trace(
 fig.show()
 ```
 
-#### WebGL Rendering with 1 Million Points
+See https://plotly.com/python/reference/scattergl/ for more information and chart attribute options!
 
-```python
-import plotly.graph_objects as go
+## NumPy and NumPy Convertible Arrays for Improved Performance
 
-import numpy as np
+Improve the performance of generating Plotly figures that use a large number of data points by using NumPy arrays and other objects that Plotly can convert to NumPy arrays, such as Pandas and Polars Series.
 
-N = 1000000
+Plotly.py uses Plotly.js for rendering, which supports typed arrays. In Plotly.py, NumPy array and NumPy-convertible arrays are base64 encoded before being passed to Plotly.js for rendering.
 
-# Create figure
-fig = go.Figure()
+### Arrays and Data Types Supported
 
-fig.add_trace(
-    go.Scattergl(
-        x = np.random.randn(N),
-        y = np.random.randn(N),
-        mode = 'markers',
-        marker = dict(
-            line = dict(
-                width = 1,
-                color = 'DarkSlateGrey')
-        )
-    )
-)
+The following types of objects in Python are supported for improved performance:
+
+- Numpy `numpy.ndarray` objects.
+- Pandas Index, Pandas Series, Polars Series, and PyArrow Chunked Array objects. 
+- Array objects that can be converted to `numpy.ndarray` objects. i.e., they implement `"__array__"` or `"__array_interface__"` and return a `numpy.ndarray`.
+
+The following [array data types](https://numpy.org/devdocs/reference/arrays.scalars.html) are supported:
+
+- float32
+- float64
+- int8
+- uint8
+- int16
+- uint16
+- int32
+- uint32
+
+*If the array dtype is **int64** and **uint64**, often the default dtype for arrays in NumPy when no dtype is specified, those dtypes will be changed to other types internally by Plotly.py where possible. When working with NumPY directly, you can [specify the `dtype`](https://numpy.org/doc/stable/user/basics.types.html#array-types-and-conversions-between-types) when creating `ndarray` objects.
+
+### Unsupported Attributes
+
+Arrays passed to attributes with the following names are not supported:
+
+`geojson`, `layers`, and `range`.
 
-fig.show()
-```
 
-#### WebGL with many traces
+### Example with NumPy Arrays
+
+Here, we use NumPy arrays with a `go.Scatter3d` figure.
 
 ```python
 import plotly.graph_objects as go
-
 import numpy as np
 
-fig = go.Figure()
+np.random.seed(1)
 
-trace_num = 10
-point_num = 5000
-for i in range(trace_num):
-    fig.add_trace(
-        go.Scattergl(
-                x = np.linspace(0, 1, point_num),
-                y = np.random.randn(point_num)+(i*5)
-        )
-    )
+# Number of data points
+N = 10000
+
+# Generate random data
+x = np.random.randn(N)
+y = np.random.randn(N).astype('float32')
+z = np.random.randint(size=N, low=0, high=256, dtype='uint8')
+c = np.random.randint(size=N, low=-10, high=10, dtype='int8')
 
-fig.update_layout(showlegend=False)
+fig = go.Figure(data=[go.Scatter3d(
+    x=x,
+    y=y,
+    z=z,
+    marker=dict(color=c),
+    mode='markers',
+    opacity=0.2
+)])
 
 fig.show()
 ```
-
-### Reference
-
-See https://plotly.com/python/reference/scattergl/ for more information and chart attribute options!