Update performance thresholds

Author: marthacryan

packages/python/plotly/plotly/tests/test_core/test_graph_objs/test_performance.py

@@ -1,6 +1,7 @@
 import time
 import numpy as np
 import plotly.graph_objs as go
+import pytest
 
 np.random.seed(1)
 
@@ -14,14 +15,18 @@ def test_performance_b64_scatter3d():
     c = np.random.randint(size=N, low=-10, high=10, dtype="int8")
 
     # Test the performance with lists
+    x_list = x.tolist()
+    y_list = y.tolist()
+    z_list = z.tolist()
+    c_list = c.tolist()
     list_start = time.time()
     fig = go.Figure(
         data=[
             go.Scatter3d(
-                x=x.tolist(),
-                y=y.tolist(),
-                z=z.tolist(),
-                marker=dict(color=c.tolist()),
+                x=x_list,
+                y=y_list,
+                z=z_list,
+                marker=dict(color=c_list),
                 mode="markers",
                 opacity=0.2,
             )
@@ -41,31 +46,46 @@ def test_performance_b64_scatter3d():
     )
     np_time_elapsed = time.time() - np_start
 
-    # np should be faster than raw
-    assert (np_time_elapsed / list_time_elapsed) < 0.75
+    # np should be faster than lists
+    assert (np_time_elapsed / list_time_elapsed) < 0.005
 
 
 def test_performance_b64_float64():
     np_arr_1 = np.random.random(10000)
     np_arr_2 = np.random.random(10000)
+    list_1 = np_arr_1.tolist()
+    list_2 = np_arr_2.tolist()
 
     # Test the performance of the base64 arrays
     np_start = time.time()
     fig = go.Scatter(x=np_arr_1, y=np_arr_2)
     np_time_elapsed = time.time() - np_start
 
-    # Test the performance of the raw arrays
+    # Test the performance of the normal lists
     list_start = time.time()
-    fig = go.Scatter(x=np_arr_1.tolist(), y=np_arr_2.tolist())
+    fig = go.Scatter(x=list_1, y=list_2)
     list_time_elapsed = time.time() - list_start
 
-    # np should be faster than raw
-    assert (np_time_elapsed / list_time_elapsed) < 0.75
+    # np should be faster than lists
+    assert (np_time_elapsed / list_time_elapsed) < 0.3
 
 
-def test_size_performance_b64_uint8():
-    np_arr_1 = (np.random.random(100000) * 256).astype("uint8")
-    np_arr_2 = (np.random.random(100000) * 256).astype("uint8")
+DTYPE_TEST_CASES = [
+    (
+        "uint8", # dtype
+        256, # max_val
+        400000 # difference threshold
+    ),
+    (
+        'uint32',
+        2**32,
+        900000
+    )
+]
+@pytest.mark.parametrize('dtype, max_val, expected_size_difference', DTYPE_TEST_CASES)
+def test_size_performance_b64_uint8(dtype, max_val, expected_size_difference):
+    np_arr_1 = (np.random.random(100000) * max_val).astype(dtype)
+    np_arr_2 = (np.random.random(100000) * max_val).astype(dtype)
 
     # Measure the size of figures with numpy arrays
     fig_np = go.Scatter(x=np_arr_1, y=np_arr_2)
@@ -75,5 +95,5 @@ def test_size_performance_b64_uint8():
     fig_list = go.Scatter(x=np_arr_1.tolist(), y=np_arr_2.tolist())
     size_list = fig_list.to_json().__sizeof__()
 
-    # np should be smaller than raw
-    assert size_list - size_np > 1000
+    # np should be smaller than lists
+    assert size_list - size_np > expected_size_difference