accelerate plotly JSON encoder for numpy arrays without nans (#2880)

* accelerate plotly JSON encoder for numpy arrays without nans

* limit change to numerical data types

* solution that works for 99% of cases but lists with composite types won't work

* other method: brute force string matching of Infinity or NaN

* black

* added test

* black

* removed ununsed variable

* changelog entry

* NaN tested before Infinity

Author: emmanuelle

CHANGELOG.md

@@ -46,6 +46,11 @@ This project adheres to [Semantic Versioning](http://semver.org/).
 
 ### Updated
 
+- The JSON serialization of plotly figures had been accelerated by handling
+  differently figures with and without NaN and Inf values ([#2880](https://github.com/plotly/plotly.py/pull/2880)).
+  
+### Updated
+
 - Updated Plotly.js to version 1.55.2. See the [plotly.js CHANGELOG](https://github.com/plotly/plotly.js/blob/v1.55.2/CHANGELOG.md) for more information. These changes are reflected in the auto-generated `plotly.graph_objects` module.
 - `px.imshow` has a new `binary_string` boolean argument, which passes the
   image data as a b64 binary string when True. Using binary strings allow for

packages/python/plotly/_plotly_utils/utils.py

@@ -55,10 +55,14 @@ def encode(self, o):
         Note that setting invalid separators will cause a failure at this step.
 
         """
-
         # this will raise errors in a normal-expected way
         encoded_o = super(PlotlyJSONEncoder, self).encode(o)
-
+        # Brute force guessing whether NaN or Infinity values are in the string
+        # We catch false positive cases (e.g. strings such as titles, labels etc.)
+        # but this is ok since the intention is to skip the decoding / reencoding
+        # step when it's completely safe
+        if not ("NaN" in encoded_o or "Infinity" in encoded_o):
+            return encoded_o
         # now:
         #    1. `loads` to switch Infinity, -Infinity, NaN to None
         #    2. `dumps` again so you get 'null' instead of extended JSON

packages/python/plotly/plotly/tests/test_core/test_utils/test_utils.py

@@ -1,11 +1,13 @@
 from __future__ import absolute_import
 
-from inspect import getargspec
 from unittest import TestCase
 
 import json as _json
 
 from plotly.utils import PlotlyJSONEncoder, get_by_path, node_generator
+from time import time
+import numpy as np
+import plotly.graph_objects as go
 
 
 class TestJSONEncoder(TestCase):
@@ -19,6 +21,38 @@ def test_invalid_encode_exception(self):
         with self.assertRaises(TypeError):
             _json.dumps({"a": {1}}, cls=PlotlyJSONEncoder)
 
+    def test_fast_track_finite_arrays(self):
+        # if NaN or Infinity is found in the json dump
+        # of a figure, it is decoded and re-encoded to replace these values
+        # with null. This test checks that NaN and Infinity values are
+        # indeed converted to null, and that the encoding of figures
+        # without inf or nan is faster (because we can avoid decoding
+        # and reencoding).
+        z = np.random.randn(100, 100)
+        x = np.arange(100.0)
+        fig_1 = go.Figure(go.Heatmap(z=z, x=x))
+        t1 = time()
+        json_str_1 = _json.dumps(fig_1, cls=PlotlyJSONEncoder)
+        t2 = time()
+        x[0] = np.nan
+        x[1] = np.inf
+        fig_2 = go.Figure(go.Heatmap(z=z, x=x))
+        t3 = time()
+        json_str_2 = _json.dumps(fig_2, cls=PlotlyJSONEncoder)
+        t4 = time()
+        assert t2 - t1 < t4 - t3
+        assert "null" in json_str_2
+        assert "NaN" not in json_str_2
+        assert "Infinity" not in json_str_2
+        x = np.arange(100.0)
+        fig_3 = go.Figure(go.Heatmap(z=z, x=x))
+        fig_3.update_layout(title_text="Infinity")
+        t5 = time()
+        json_str_3 = _json.dumps(fig_3, cls=PlotlyJSONEncoder)
+        t6 = time()
+        assert t2 - t1 < t6 - t5
+        assert "Infinity" in json_str_3
+
 
 class TestGetByPath(TestCase):
     def test_get_by_path(self):