Revision: 8946
http://matplotlib.svn.sourceforge.net/matplotlib/?rev=8946&view=rev
Author: efiring
Date: 2011-02-05 20:56:02 +0000 (Sat, 05 Feb 2011)
Log Message:
-----------
speed up Normalize, LogNorm, colormapping; modified patch by C. Gohlke
Modified Paths:
--------------
trunk/matplotlib/CHANGELOG
trunk/matplotlib/examples/pylab_examples/image_slices_viewer.py
trunk/matplotlib/lib/matplotlib/colors.py
Modified: trunk/matplotlib/CHANGELOG
===================================================================
--- trunk/matplotlib/CHANGELOG 2011-02-04 19:18:57 UTC (rev 8945)
+++ trunk/matplotlib/CHANGELOG 2011-02-05 20:56:02 UTC (rev 8946)
@@ -1,3 +1,8 @@
+2011-02-05 Speed up Normalize and LogNorm by using in-place
+ operations and by using float32 for float32 inputs
+ and for ints of 2 bytes or shorter; based on
+ patch by Christoph Gohlke. - EF
+
2011-02-04 Changed imshow to use rgba as uint8 from start to
finish, instead of going through an intermediate
step as double precision; thanks to Christoph Gohlke. - EF
Modified: trunk/matplotlib/examples/pylab_examples/image_slices_viewer.py
===================================================================
--- trunk/matplotlib/examples/pylab_examples/image_slices_viewer.py
2011-02-04 19:18:57 UTC (rev 8945)
+++ trunk/matplotlib/examples/pylab_examples/image_slices_viewer.py
2011-02-05 20:56:02 UTC (rev 8946)
@@ -1,5 +1,5 @@
import numpy
-from pylab import figure, show
+from matplotlib.pyplot import figure, show
@@ -7,7 +7,7 @@
class IndexTracker:
def __init__(self, ax, X):
self.ax = ax
- ax.set_title('use scroll wheen to navigate images')
+ ax.set_title('use scroll wheel to navigate images')
self.X = X
rows,cols,self.slices = X.shape
@@ -28,7 +28,7 @@
def update(self):
self.im.set_data(self.X[:,:,self.ind])
- ax.set_ylabel('slice %s'%self.ind)
+ ax.set_ylabel('slice %s'%self.ind)
self.im.axes.figure.canvas.draw()
Modified: trunk/matplotlib/lib/matplotlib/colors.py
===================================================================
--- trunk/matplotlib/lib/matplotlib/colors.py 2011-02-04 19:18:57 UTC (rev
8945)
+++ trunk/matplotlib/lib/matplotlib/colors.py 2011-02-05 20:56:02 UTC (rev
8946)
@@ -509,20 +509,22 @@
xa = np.array([X])
else:
vtype = 'array'
- # force a copy here -- the ma.array and filled functions
- # do force a cop of the data by default - JDH
- xma = ma.array(X, copy=True)
- xa = xma.filled(0)
- mask_bad = ma.getmask(xma)
+ xma = ma.array(X, copy=False)
+ mask_bad = xma.mask
+ xa = xma.data.copy() # Copy here to avoid side effects.
+ del xma
+ # masked values are substituted below; no need to fill them here
+
if xa.dtype.char in np.typecodes['Float']:
np.putmask(xa, xa==1.0, 0.9999999) #Treat 1.0 as slightly less
than 1.
# The following clip is fast, and prevents possible
# conversion of large positive values to negative integers.
+ xa *= self.N
if NP_CLIP_OUT:
- np.clip(xa * self.N, -1, self.N, out=xa)
+ np.clip(xa, -1, self.N, out=xa)
else:
- xa = np.clip(xa * self.N, -1, self.N)
+ xa = np.clip(xa, -1, self.N)
# ensure that all 'under' values will still have negative
# value after casting to int
@@ -532,8 +534,11 @@
# otherwise the under-range values get converted to over-range.
np.putmask(xa, xa>self.N-1, self._i_over)
np.putmask(xa, xa<0, self._i_under)
- if mask_bad is not None and mask_bad.shape == xa.shape:
- np.putmask(xa, mask_bad, self._i_bad)
+ if mask_bad is not None:
+ if mask_bad.shape == xa.shape:
+ np.putmask(xa, mask_bad, self._i_bad)
+ elif mask_bad:
+ xa.fill(self._i_bad)
if bytes:
lut = (self._lut * 255).astype(np.uint8)
else:
@@ -797,32 +802,60 @@
self.vmax = vmax
self.clip = clip
+ @staticmethod
+ def process_value(value):
+ """
+ Homogenize the input *value* for easy and efficient normalization.
+
+ *value* can be a scalar or sequence.
+
+ Returns *result*, *is_scalar*, where *result* is a
+ masked array matching *value*. Float dtypes are preserved;
+ integer types with two bytes or smaller are converted to
+ np.float32, and larger types are converted to np.float.
+ Preserving float32 when possible, and using in-place operations,
+ can greatly improve speed for large arrays.
+
+ Experimental; we may want to add an option to force the
+ use of float32.
+ """
+ if cbook.iterable(value):
+ is_scalar = False
+ result = ma.asarray(value)
+ if result.dtype.kind == 'f':
+ if isinstance(value, np.ndarray):
+ result = result.copy()
+ elif result.dtype.itemsize > 2:
+ result = result.astype(np.float)
+ else:
+ result = result.astype(np.float32)
+ else:
+ is_scalar = True
+ result = ma.array([value]).astype(np.float)
+ return result, is_scalar
+
def __call__(self, value, clip=None):
if clip is None:
clip = self.clip
- if cbook.iterable(value):
- vtype = 'array'
- val = ma.asarray(value).astype(np.float)
- else:
- vtype = 'scalar'
- val = ma.array([value]).astype(np.float)
+ result, is_scalar = self.process_value(value)
- self.autoscale_None(val)
+ self.autoscale_None(result)
vmin, vmax = self.vmin, self.vmax
if vmin > vmax:
raise ValueError("minvalue must be less than or equal to maxvalue")
- elif vmin==vmax:
- result = 0.0 * val
+ elif vmin == vmax:
+ result.fill(0) # Or should it be all masked? Or 0.5?
else:
vmin = float(vmin)
vmax = float(vmax)
if clip:
- mask = ma.getmask(val)
- val = ma.array(np.clip(val.filled(vmax), vmin, vmax),
- mask=mask)
- result = (val-vmin) / (vmax-vmin)
- if vtype == 'scalar':
+ mask = ma.getmask(result)
+ result = ma.array(np.clip(result.filled(vmax), vmin, vmax),
+ mask=mask)
+ result -= vmin
+ result /= vmax - vmin
+ if is_scalar:
result = result[0]
return result
@@ -847,8 +880,10 @@
def autoscale_None(self, A):
' autoscale only None-valued vmin or vmax'
- if self.vmin is None: self.vmin = ma.min(A)
- if self.vmax is None: self.vmax = ma.max(A)
+ if self.vmin is None:
+ self.vmin = ma.min(A)
+ if self.vmax is None:
+ self.vmax = ma.max(A)
def scaled(self):
'return true if vmin and vmax set'
@@ -862,30 +897,29 @@
if clip is None:
clip = self.clip
- if cbook.iterable(value):
- vtype = 'array'
- val = ma.asarray(value).astype(np.float)
- else:
- vtype = 'scalar'
- val = ma.array([value]).astype(np.float)
+ result, is_scalar = self.process_value(value)
- val = ma.masked_less_equal(val, 0, copy=False)
+ result = ma.masked_less_equal(result, 0, copy=False)
- self.autoscale_None(val)
+ self.autoscale_None(result)
vmin, vmax = self.vmin, self.vmax
if vmin > vmax:
raise ValueError("minvalue must be less than or equal to maxvalue")
elif vmin<=0:
raise ValueError("values must all be positive")
elif vmin==vmax:
- result = 0.0 * val
+ result.fill(0)
else:
if clip:
- mask = ma.getmask(val)
- val = ma.array(np.clip(val.filled(vmax), vmin, vmax),
+ mask = ma.getmask(result)
+ val = ma.array(np.clip(result.filled(vmax), vmin, vmax),
mask=mask)
- result = (ma.log(val)-np.log(vmin))/(np.log(vmax)-np.log(vmin))
- if vtype == 'scalar':
+ #result = (ma.log(result)-np.log(vmin))/(np.log(vmax)-np.log(vmin))
+ # in-place equivalent of above can be much faster
+ np.ma.log(result, result)
+ result -= np.log(vmin)
+ result /= (np.log(vmax) - np.log(vmin))
+ if is_scalar:
result = result[0]
return result
This was sent by the SourceForge.net collaborative development platform, the
world's largest Open Source development site.
------------------------------------------------------------------------------
The modern datacenter depends on network connectivity to access resources
and provide services. The best practices for maximizing a physical server's
connectivity to a physical network are well understood - see how these
rules translate into the virtual world?
http://p.sf.net/sfu/oracle-sfdevnlfb
_______________________________________________
Matplotlib-checkins mailing list
Matplotlib-checkins@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/matplotlib-checkins
