Don't call np.identity() in transforms.

anntzer · anntzer · commit f2b6b1fd89b3 · 2018-01-26T00:05:48.000-08:00
Speeds up setting up of subplots by ~3% and their drawing by ~2% at a
reasonably limited cost in legibility.

The difference is actually not easy to observe because variation in
timings are much greater than 3% from run to run.

In a call like `subplots(10, 10)`, just setting up the figure calls
`np.identity(3)` 1801 times and actually drawing the figure another 1705
times (measured by manually instrumenting the calls to identity() in the
old version.  At the same time,

```
%timeit np.identity(3)
```
and
```
%%timeit t = np.identity(3)
t.copy()
```
show that on my machine, a call to np.identity(3) is approximately 16us
slower than copying a preexisting matrix; multiplying this by 1801 shows
that the calls to identity correspond to an excess time of ~27ms.

Finally, setting up the axes and drawing it can be timed using
```
from time import perf_counter
import matplotlib; matplotlib.use("agg")
from matplotlib import pyplot as plt

N = 10
 # First draw always seems slower, so skip it.
figure = plt.figure()
figure.subplots(N, N)
figure.canvas.draw()
plt.close("all")

for _ in range(5):
    figure = plt.figure()
    start = perf_counter()
    figure.subplots(N, N)
    print("{: 4}".format(int(1000 * (perf_counter() - start))), end=" ")
    start = perf_counter()
    figure.canvas.draw()
    print("{: 4}".format(int(1000 * (perf_counter() - start))))
    plt.close("all")
```
which shows that (on my machine) setting up the subplots takes ~850ms
and drawing them ~1300ms (but again, with a lot of jitter).

Hence, the gain from the patch should be ~3% for the setup and ~2% for
the draw.
diff --git a/lib/matplotlib/transforms.py b/lib/matplotlib/transforms.py
@@ -1928,7 +1928,8 @@ def __init__(self, matrix=None, **kwargs):
         """
         Affine2DBase.__init__(self, **kwargs)
         if matrix is None:
-            matrix = np.identity(3)
+            # Significantly faster than np.identity(3).
+            matrix = IdentityTransform._mtx.copy()
         self._mtx = matrix
         self._invalid = 0
 
@@ -1999,13 +2000,14 @@ def identity():
         Unless this transform will be mutated later on, consider using
         the faster :class:`IdentityTransform` class instead.
         """
-        return Affine2D(np.identity(3))
+        return Affine2D()
 
     def clear(self):
         """
         Reset the underlying matrix to the identity transform.
         """
-        self._mtx = np.identity(3)
+        # Significantly faster than np.identity(3).
+        self._mtx = IdentityTransform._mtx.copy()
         self.invalidate()
         return self
 
