Removed all dependencies on scipy.

Merge branch 'make_scipy_optional' into develop
This commit is contained in:
Luke Campagnola 2014-03-11 19:05:30 -04:00
commit 1edf1375ed
21 changed files with 413 additions and 273 deletions

View File

@ -23,6 +23,7 @@ pyqtgraph-0.9.9 [unreleased]
- Added ViewBox.setLimits() method
- Adde ImageItem downsampling
- New HDF5 example for working with very large datasets
- Removed all dependency on scipy
- Added Qt.loadUiType function for PySide
- Simplified Profilers; can be activated with environmental variables
- Added Dock.raiseDock() method

View File

@ -11,7 +11,6 @@ a 2D plane and interpolate data along that plane to generate a slice image
import initExample
import numpy as np
import scipy
from pyqtgraph.Qt import QtCore, QtGui
import pyqtgraph as pg

View File

@ -12,7 +12,6 @@ from pyqtgraph.flowchart.library.common import CtrlNode
from pyqtgraph.Qt import QtGui, QtCore
import pyqtgraph as pg
import numpy as np
import scipy.ndimage
app = QtGui.QApplication([])
@ -44,7 +43,7 @@ win.show()
## generate random input data
data = np.random.normal(size=(100,100))
data = 25 * scipy.ndimage.gaussian_filter(data, (5,5))
data = 25 * pg.gaussianFilter(data, (5,5))
data += np.random.normal(size=(100,100))
data[40:60, 40:60] += 15.0
data[30:50, 30:50] += 15.0
@ -90,7 +89,7 @@ class ImageViewNode(Node):
## CtrlNode is just a convenience class that automatically creates its
## control widget based on a simple data structure.
class UnsharpMaskNode(CtrlNode):
"""Return the input data passed through scipy.ndimage.gaussian_filter."""
"""Return the input data passed through pg.gaussianFilter."""
nodeName = "UnsharpMask"
uiTemplate = [
('sigma', 'spin', {'value': 1.0, 'step': 1.0, 'range': [0.0, None]}),
@ -110,7 +109,7 @@ class UnsharpMaskNode(CtrlNode):
# CtrlNode has created self.ctrls, which is a dict containing {ctrlName: widget}
sigma = self.ctrls['sigma'].value()
strength = self.ctrls['strength'].value()
output = dataIn - (strength * scipy.ndimage.gaussian_filter(dataIn, (sigma,sigma)))
output = dataIn - (strength * pg.gaussianFilter(dataIn, (sigma,sigma)))
return {'dataOut': output}

View File

@ -12,7 +12,6 @@ from pyqtgraph.Qt import QtCore, QtGui
import pyqtgraph.opengl as gl
import pyqtgraph as pg
import numpy as np
import scipy.ndimage as ndi
app = QtGui.QApplication([])
w = gl.GLViewWidget()
@ -22,8 +21,8 @@ w.setWindowTitle('pyqtgraph example: GLImageItem')
## create volume data set to slice three images from
shape = (100,100,70)
data = ndi.gaussian_filter(np.random.normal(size=shape), (4,4,4))
data += ndi.gaussian_filter(np.random.normal(size=shape), (15,15,15))*15
data = pg.gaussianFilter(np.random.normal(size=shape), (4,4,4))
data += pg.gaussianFilter(np.random.normal(size=shape), (15,15,15))*15
## slice out three planes, convert to RGBA for OpenGL texture
levels = (-0.08, 0.08)

View File

@ -10,7 +10,6 @@ import initExample
from pyqtgraph.Qt import QtCore, QtGui
import pyqtgraph as pg
import pyqtgraph.opengl as gl
import scipy.ndimage as ndi
import numpy as np
## Create a GL View widget to display data
@ -29,7 +28,7 @@ w.addItem(g)
## Simple surface plot example
## x, y values are not specified, so assumed to be 0:50
z = ndi.gaussian_filter(np.random.normal(size=(50,50)), (1,1))
z = pg.gaussianFilter(np.random.normal(size=(50,50)), (1,1))
p1 = gl.GLSurfacePlotItem(z=z, shader='shaded', color=(0.5, 0.5, 1, 1))
p1.scale(16./49., 16./49., 1.0)
p1.translate(-18, 2, 0)
@ -46,7 +45,7 @@ w.addItem(p2)
## Manually specified colors
z = ndi.gaussian_filter(np.random.normal(size=(50,50)), (1,1))
z = pg.gaussianFilter(np.random.normal(size=(50,50)), (1,1))
x = np.linspace(-12, 12, 50)
y = np.linspace(-12, 12, 50)
colors = np.ones((50,50,4), dtype=float)

View File

@ -7,7 +7,6 @@ Use a HistogramLUTWidget to control the contrast / coloration of an image.
import initExample
import numpy as np
import scipy.ndimage as ndi
from pyqtgraph.Qt import QtGui, QtCore
import pyqtgraph as pg
@ -34,7 +33,7 @@ l.addWidget(v, 0, 0)
w = pg.HistogramLUTWidget()
l.addWidget(w, 0, 1)
data = ndi.gaussian_filter(np.random.normal(size=(256, 256)), (20, 20))
data = pg.gaussianFilter(np.random.normal(size=(256, 256)), (20, 20))
for i in range(32):
for j in range(32):
data[i*8, j*8] += .1

View File

@ -14,7 +14,6 @@ displaying and analyzing 2D and 3D data. ImageView provides:
import initExample
import numpy as np
import scipy
from pyqtgraph.Qt import QtCore, QtGui
import pyqtgraph as pg
@ -29,7 +28,7 @@ win.show()
win.setWindowTitle('pyqtgraph example: ImageView')
## Create random 3D data set with noisy signals
img = scipy.ndimage.gaussian_filter(np.random.normal(size=(200, 200)), (5, 5)) * 20 + 100
img = pg.gaussianFilter(np.random.normal(size=(200, 200)), (5, 5)) * 20 + 100
img = img[np.newaxis,:,:]
decay = np.exp(-np.linspace(0,0.3,100))[:,np.newaxis,np.newaxis]
data = np.random.normal(size=(100, 200, 200))

View File

@ -13,7 +13,6 @@ import initExample ## Add path to library (just for examples; you do not need th
from pyqtgraph.Qt import QtGui, QtCore, USE_PYSIDE
import numpy as np
import pyqtgraph as pg
import scipy.ndimage as ndi
import pyqtgraph.ptime as ptime
if USE_PYSIDE:
@ -95,10 +94,11 @@ def mkData():
if ui.rgbCheck.isChecked():
data = np.random.normal(size=(frames,width,height,3), loc=loc, scale=scale)
data = ndi.gaussian_filter(data, (0, 6, 6, 0))
data = pg.gaussianFilter(data, (0, 6, 6, 0))
else:
data = np.random.normal(size=(frames,width,height), loc=loc, scale=scale)
data = ndi.gaussian_filter(data, (0, 6, 6))
data = pg.gaussianFilter(data, (0, 6, 6))
pg.image(data)
if dtype[0] != 'float':
data = np.clip(data, 0, mx)
data = data.astype(dt)

View File

@ -14,7 +14,6 @@ import initExample
## This example uses a ViewBox to create a PlotWidget-like interface
#from scipy import random
import numpy as np
from pyqtgraph.Qt import QtGui, QtCore
import pyqtgraph as pg

View File

@ -7,7 +7,6 @@ the mouse.
import initExample ## Add path to library (just for examples; you do not need this)
import numpy as np
import scipy.ndimage as ndi
import pyqtgraph as pg
from pyqtgraph.Qt import QtGui, QtCore
from pyqtgraph.Point import Point
@ -33,8 +32,8 @@ p1.setAutoVisible(y=True)
#create numpy arrays
#make the numbers large to show that the xrange shows data from 10000 to all the way 0
data1 = 10000 + 15000 * ndi.gaussian_filter(np.random.random(size=10000), 10) + 3000 * np.random.random(size=10000)
data2 = 15000 + 15000 * ndi.gaussian_filter(np.random.random(size=10000), 10) + 3000 * np.random.random(size=10000)
data1 = 10000 + 15000 * pg.gaussianFilter(np.random.random(size=10000), 10) + 3000 * np.random.random(size=10000)
data2 = 15000 + 15000 * pg.gaussianFilter(np.random.random(size=10000), 10) + 3000 * np.random.random(size=10000)
p1.plot(data1, pen="r")
p1.plot(data2, pen="g")

View File

@ -10,7 +10,6 @@ import initExample ## Add path to library (just for examples; you do not need th
from pyqtgraph.Qt import QtGui, QtCore
import numpy as np
import pyqtgraph as pg
import scipy.ndimage as ndi
app = QtGui.QApplication([])
@ -18,7 +17,7 @@ app = QtGui.QApplication([])
frames = 200
data = np.random.normal(size=(frames,30,30), loc=0, scale=100)
data = np.concatenate([data, data], axis=0)
data = ndi.gaussian_filter(data, (10, 10, 10))[frames/2:frames + frames/2]
data = pg.gaussianFilter(data, (10, 10, 10))[frames/2:frames + frames/2]
data[:, 15:16, 15:17] += 1
win = pg.GraphicsWindow()

View File

@ -4,7 +4,6 @@ from .Vector import Vector
from .Transform3D import Transform3D
from .Vector import Vector
import numpy as np
import scipy.linalg
class SRTTransform3D(Transform3D):
"""4x4 Transform matrix that can always be represented as a combination of 3 matrices: scale * rotate * translate
@ -118,11 +117,13 @@ class SRTTransform3D(Transform3D):
The input matrix must be affine AND have no shear,
otherwise the conversion will most likely fail.
"""
import numpy.linalg
for i in range(4):
self.setRow(i, m.row(i))
m = self.matrix().reshape(4,4)
## translation is 4th column
self._state['pos'] = m[:3,3]
## scale is vector-length of first three columns
scale = (m[:3,:3]**2).sum(axis=0)**0.5
## see whether there is an inversion
@ -132,9 +133,9 @@ class SRTTransform3D(Transform3D):
self._state['scale'] = scale
## rotation axis is the eigenvector with eigenvalue=1
r = m[:3, :3] / scale[:, np.newaxis]
r = m[:3, :3] / scale[np.newaxis, :]
try:
evals, evecs = scipy.linalg.eig(r)
evals, evecs = numpy.linalg.eig(r)
except:
print("Rotation matrix: %s" % str(r))
print("Scale: %s" % str(scale))

View File

@ -1,5 +1,4 @@
import numpy as np
import scipy.interpolate
from .Qt import QtGui, QtCore
class ColorMap(object):
@ -64,8 +63,8 @@ class ColorMap(object):
ignored. By default, the mode is entirely RGB.
=============== ==============================================================
"""
self.pos = pos
self.color = color
self.pos = np.array(pos)
self.color = np.array(color)
if mode is None:
mode = np.ones(len(pos))
self.mode = mode
@ -92,15 +91,24 @@ class ColorMap(object):
else:
pos, color = self.getStops(mode)
data = np.clip(data, pos.min(), pos.max())
# don't need this--np.interp takes care of it.
#data = np.clip(data, pos.min(), pos.max())
if not isinstance(data, np.ndarray):
interp = scipy.interpolate.griddata(pos, color, np.array([data]))[0]
# Interpolate
# TODO: is griddata faster?
# interp = scipy.interpolate.griddata(pos, color, data)
if np.isscalar(data):
interp = np.empty((color.shape[1],), dtype=color.dtype)
else:
interp = scipy.interpolate.griddata(pos, color, data)
if mode == self.QCOLOR:
if not isinstance(data, np.ndarray):
data = np.array(data)
interp = np.empty(data.shape + (color.shape[1],), dtype=color.dtype)
for i in range(color.shape[1]):
interp[...,i] = np.interp(data, pos, color[:,i])
# Convert to QColor if requested
if mode == self.QCOLOR:
if np.isscalar(data):
return QtGui.QColor(*interp)
else:
return [QtGui.QColor(*x) for x in interp]

View File

@ -399,7 +399,9 @@ class Profiler(object):
only the initial "pyqtgraph." prefix from the module.
"""
_profilers = os.environ.get("PYQTGRAPHPROFILE", "")
_profilers = os.environ.get("PYQTGRAPHPROFILE", None)
_profilers = _profilers.split(",") if _profilers is not None else []
_depth = 0
_msgs = []
@ -415,12 +417,10 @@ class Profiler(object):
_disabledProfiler = DisabledProfiler()
if _profilers:
_profilers = _profilers.split(",")
def __new__(cls, msg=None, disabled='env', delayed=True):
"""Optionally create a new profiler based on caller's qualname.
"""
if disabled is True:
if disabled is True or (disabled=='env' and len(cls._profilers) == 0):
return cls._disabledProfiler
# determine the qualified name of the caller function
@ -432,7 +432,7 @@ class Profiler(object):
else: # we are in a method
qualifier = caller_object_type.__name__
func_qualname = qualifier + "." + caller_frame.f_code.co_name
if func_qualname not in cls._profilers: # don't do anything
if disabled=='env' and func_qualname not in cls._profilers: # don't do anything
return cls._disabledProfiler
# create an actual profiling object
cls._depth += 1
@ -444,9 +444,9 @@ class Profiler(object):
obj._firstTime = obj._lastTime = ptime.time()
obj._newMsg("> Entering " + obj._name)
return obj
else:
def __new__(cls, delayed=True):
return lambda msg=None: None
#else:
#def __new__(cls, delayed=True):
#return lambda msg=None: None
def __call__(self, msg=None):
"""Register or print a new message with timing information.

View File

@ -1,10 +1,8 @@
# -*- coding: utf-8 -*-
from ...Qt import QtCore, QtGui
from ..Node import Node
from scipy.signal import detrend
from scipy.ndimage import median_filter, gaussian_filter
#from ...SignalProxy import SignalProxy
from . import functions
from ... import functions as pgfn
from .common import *
import numpy as np
@ -119,7 +117,11 @@ class Median(CtrlNode):
@metaArrayWrapper
def processData(self, data):
return median_filter(data, self.ctrls['n'].value())
try:
import scipy.ndimage
except ImportError:
raise Exception("MedianFilter node requires the package scipy.ndimage.")
return scipy.ndimage.median_filter(data, self.ctrls['n'].value())
class Mode(CtrlNode):
"""Filters data by taking the mode (histogram-based) of a sliding window"""
@ -156,7 +158,11 @@ class Gaussian(CtrlNode):
@metaArrayWrapper
def processData(self, data):
return gaussian_filter(data, self.ctrls['sigma'].value())
try:
import scipy.ndimage
except ImportError:
raise Exception("GaussianFilter node requires the package scipy.ndimage.")
return pgfn.gaussianFilter(data, self.ctrls['sigma'].value())
class Derivative(CtrlNode):
@ -189,6 +195,10 @@ class Detrend(CtrlNode):
@metaArrayWrapper
def processData(self, data):
try:
from scipy.signal import detrend
except ImportError:
raise Exception("DetrendFilter node requires the package scipy.signal.")
return detrend(data)

View File

@ -1,4 +1,3 @@
import scipy
import numpy as np
from ...metaarray import MetaArray
@ -47,6 +46,11 @@ def downsample(data, n, axis=0, xvals='subsample'):
def applyFilter(data, b, a, padding=100, bidir=True):
"""Apply a linear filter with coefficients a, b. Optionally pad the data before filtering
and/or run the filter in both directions."""
try:
import scipy.signal
except ImportError:
raise Exception("applyFilter() requires the package scipy.signal.")
d1 = data.view(np.ndarray)
if padding > 0:
@ -67,6 +71,11 @@ def applyFilter(data, b, a, padding=100, bidir=True):
def besselFilter(data, cutoff, order=1, dt=None, btype='low', bidir=True):
"""return data passed through bessel filter"""
try:
import scipy.signal
except ImportError:
raise Exception("besselFilter() requires the package scipy.signal.")
if dt is None:
try:
tvals = data.xvals('Time')
@ -85,6 +94,11 @@ def besselFilter(data, cutoff, order=1, dt=None, btype='low', bidir=True):
def butterworthFilter(data, wPass, wStop=None, gPass=2.0, gStop=20.0, order=1, dt=None, btype='low', bidir=True):
"""return data passed through bessel filter"""
try:
import scipy.signal
except ImportError:
raise Exception("butterworthFilter() requires the package scipy.signal.")
if dt is None:
try:
tvals = data.xvals('Time')
@ -175,6 +189,11 @@ def denoise(data, radius=2, threshold=4):
def adaptiveDetrend(data, x=None, threshold=3.0):
"""Return the signal with baseline removed. Discards outliers from baseline measurement."""
try:
import scipy.signal
except ImportError:
raise Exception("adaptiveDetrend() requires the package scipy.signal.")
if x is None:
x = data.xvals(0)

View File

@ -34,17 +34,6 @@ import decimal, re
import ctypes
import sys, struct
try:
import scipy.ndimage
HAVE_SCIPY = True
if getConfigOption('useWeave'):
try:
import scipy.weave
except ImportError:
setConfigOptions(useWeave=False)
except ImportError:
HAVE_SCIPY = False
from . import debug
def siScale(x, minVal=1e-25, allowUnicode=True):
@ -383,7 +372,7 @@ def affineSlice(data, shape, origin, vectors, axes, order=1, returnCoords=False,
"""
Take a slice of any orientation through an array. This is useful for extracting sections of multi-dimensional arrays such as MRI images for viewing as 1D or 2D data.
The slicing axes are aribtrary; they do not need to be orthogonal to the original data or even to each other. It is possible to use this function to extract arbitrary linear, rectangular, or parallelepiped shapes from within larger datasets. The original data is interpolated onto a new array of coordinates using scipy.ndimage.map_coordinates (see the scipy documentation for more information about this).
The slicing axes are aribtrary; they do not need to be orthogonal to the original data or even to each other. It is possible to use this function to extract arbitrary linear, rectangular, or parallelepiped shapes from within larger datasets. The original data is interpolated onto a new array of coordinates using scipy.ndimage.map_coordinates if it is available (see the scipy documentation for more information about this). If scipy is not available, then a slower implementation of map_coordinates is used.
For a graphical interface to this function, see :func:`ROI.getArrayRegion <pyqtgraph.ROI.getArrayRegion>`
@ -422,8 +411,12 @@ def affineSlice(data, shape, origin, vectors, axes, order=1, returnCoords=False,
affineSlice(data, shape=(20,20), origin=(40,0,0), vectors=((-1, 1, 0), (-1, 0, 1)), axes=(1,2,3))
"""
if not HAVE_SCIPY:
raise Exception("This function requires the scipy library, but it does not appear to be importable.")
try:
import scipy.ndimage
have_scipy = True
except ImportError:
have_scipy = False
have_scipy = False
# sanity check
if len(shape) != len(vectors):
@ -445,7 +438,6 @@ def affineSlice(data, shape, origin, vectors, axes, order=1, returnCoords=False,
#print "tr1:", tr1
## dims are now [(slice axes), (other axes)]
## make sure vectors are arrays
if not isinstance(vectors, np.ndarray):
vectors = np.array(vectors)
@ -461,12 +453,18 @@ def affineSlice(data, shape, origin, vectors, axes, order=1, returnCoords=False,
#print "X values:"
#print x
## iterate manually over unused axes since map_coordinates won't do it for us
if have_scipy:
extraShape = data.shape[len(axes):]
output = np.empty(tuple(shape) + extraShape, dtype=data.dtype)
for inds in np.ndindex(*extraShape):
ind = (Ellipsis,) + inds
#print data[ind].shape, x.shape, output[ind].shape, output.shape
output[ind] = scipy.ndimage.map_coordinates(data[ind], x, order=order, **kargs)
else:
# map_coordinates expects the indexes as the first axis, whereas
# interpolateArray expects indexes at the last axis.
tr = tuple(range(1,x.ndim)) + (0,)
output = interpolateArray(data, x.transpose(tr))
tr = list(range(output.ndim))
trb = []
@ -483,6 +481,117 @@ def affineSlice(data, shape, origin, vectors, axes, order=1, returnCoords=False,
else:
return output
def interpolateArray(data, x, default=0.0):
"""
N-dimensional interpolation similar scipy.ndimage.map_coordinates.
This function returns linearly-interpolated values sampled from a regular
grid of data.
*data* is an array of any shape containing the values to be interpolated.
*x* is an array with (shape[-1] <= data.ndim) containing the locations
within *data* to interpolate.
Returns array of shape (x.shape[:-1] + data.shape)
For example, assume we have the following 2D image data::
>>> data = np.array([[1, 2, 4 ],
[10, 20, 40 ],
[100, 200, 400]])
To compute a single interpolated point from this data::
>>> x = np.array([(0.5, 0.5)])
>>> interpolateArray(data, x)
array([ 8.25])
To compute a 1D list of interpolated locations::
>>> x = np.array([(0.5, 0.5),
(1.0, 1.0),
(1.0, 2.0),
(1.5, 0.0)])
>>> interpolateArray(data, x)
array([ 8.25, 20. , 40. , 55. ])
To compute a 2D array of interpolated locations::
>>> x = np.array([[(0.5, 0.5), (1.0, 2.0)],
[(1.0, 1.0), (1.5, 0.0)]])
>>> interpolateArray(data, x)
array([[ 8.25, 40. ],
[ 20. , 55. ]])
..and so on. The *x* argument may have any shape as long as
```x.shape[-1] <= data.ndim```. In the case that
```x.shape[-1] < data.ndim```, then the remaining axes are simply
broadcasted as usual. For example, we can interpolate one location
from an entire row of the data::
>>> x = np.array([[0.5]])
>>> interpolateArray(data, x)
array([[ 5.5, 11. , 22. ]])
This is useful for interpolating from arrays of colors, vertexes, etc.
"""
prof = debug.Profiler()
result = np.empty(x.shape[:-1] + data.shape, dtype=data.dtype)
nd = data.ndim
md = x.shape[-1]
# First we generate arrays of indexes that are needed to
# extract the data surrounding each point
fields = np.mgrid[(slice(0,2),) * md]
xmin = np.floor(x).astype(int)
xmax = xmin + 1
indexes = np.concatenate([xmin[np.newaxis, ...], xmax[np.newaxis, ...]])
fieldInds = []
totalMask = np.ones(x.shape[:-1], dtype=bool) # keep track of out-of-bound indexes
for ax in range(md):
mask = (xmin[...,ax] >= 0) & (x[...,ax] <= data.shape[ax]-1)
# keep track of points that need to be set to default
totalMask &= mask
# ..and keep track of indexes that are out of bounds
# (note that when x[...,ax] == data.shape[ax], then xmax[...,ax] will be out
# of bounds, but the interpolation will work anyway)
mask &= (xmax[...,ax] < data.shape[ax])
axisIndex = indexes[...,ax][fields[ax]]
#axisMask = mask.astype(np.ubyte).reshape((1,)*(fields.ndim-1) + mask.shape)
axisIndex[axisIndex < 0] = 0
axisIndex[axisIndex >= data.shape[ax]] = 0
fieldInds.append(axisIndex)
prof()
# Get data values surrounding each requested point
# fieldData[..., i] contains all 2**nd values needed to interpolate x[i]
fieldData = data[tuple(fieldInds)]
prof()
## Interpolate
s = np.empty((md,) + fieldData.shape, dtype=float)
dx = x - xmin
# reshape fields for arithmetic against dx
for ax in range(md):
f1 = fields[ax].reshape(fields[ax].shape + (1,)*(dx.ndim-1))
sax = f1 * dx[...,ax] + (1-f1) * (1-dx[...,ax])
sax = sax.reshape(sax.shape + (1,) * (s.ndim-1-sax.ndim))
s[ax] = sax
s = np.product(s, axis=0)
result = fieldData * s
for i in range(md):
result = result.sum(axis=0)
prof()
totalMask.shape = totalMask.shape + (1,) * (nd - md)
result[~totalMask] = default
prof()
return result
def transformToArray(tr):
"""
Given a QTransform, return a 3x3 numpy array.
@ -577,17 +686,25 @@ def transformCoordinates(tr, coords, transpose=False):
def solve3DTransform(points1, points2):
"""
Find a 3D transformation matrix that maps points1 onto points2.
Points must be specified as a list of 4 Vectors.
Points must be specified as either lists of 4 Vectors or
(4, 3) arrays.
"""
if not HAVE_SCIPY:
raise Exception("This function depends on the scipy library, but it does not appear to be importable.")
A = np.array([[points1[i].x(), points1[i].y(), points1[i].z(), 1] for i in range(4)])
B = np.array([[points2[i].x(), points2[i].y(), points2[i].z(), 1] for i in range(4)])
import numpy.linalg
pts = []
for inp in (points1, points2):
if isinstance(inp, np.ndarray):
A = np.empty((4,4), dtype=float)
A[:,:3] = inp[:,:3]
A[:,3] = 1.0
else:
A = np.array([[inp[i].x(), inp[i].y(), inp[i].z(), 1] for i in range(4)])
pts.append(A)
## solve 3 sets of linear equations to determine transformation matrix elements
matrix = np.zeros((4,4))
for i in range(3):
matrix[i] = scipy.linalg.solve(A, B[:,i]) ## solve Ax = B; x is one row of the desired transformation matrix
## solve Ax = B; x is one row of the desired transformation matrix
matrix[i] = numpy.linalg.solve(pts[0], pts[1][:,i])
return matrix
@ -600,8 +717,7 @@ def solveBilinearTransform(points1, points2):
mapped = np.dot(matrix, [x*y, x, y, 1])
"""
if not HAVE_SCIPY:
raise Exception("This function depends on the scipy library, but it does not appear to be importable.")
import numpy.linalg
## A is 4 rows (points) x 4 columns (xy, x, y, 1)
## B is 4 rows (points) x 2 columns (x, y)
A = np.array([[points1[i].x()*points1[i].y(), points1[i].x(), points1[i].y(), 1] for i in range(4)])
@ -610,7 +726,7 @@ def solveBilinearTransform(points1, points2):
## solve 2 sets of linear equations to determine transformation matrix elements
matrix = np.zeros((2,4))
for i in range(2):
matrix[i] = scipy.linalg.solve(A, B[:,i]) ## solve Ax = B; x is one row of the desired transformation matrix
matrix[i] = numpy.linalg.solve(A, B[:,i]) ## solve Ax = B; x is one row of the desired transformation matrix
return matrix
@ -629,6 +745,10 @@ def rescaleData(data, scale, offset, dtype=None):
try:
if not getConfigOption('useWeave'):
raise Exception('Weave is disabled; falling back to slower version.')
try:
import scipy.weave
except ImportError:
raise Exception('scipy.weave is not importable; falling back to slower version.')
## require native dtype when using weave
if not data.dtype.isnative:
@ -671,68 +791,13 @@ def applyLookupTable(data, lut):
Uses values in *data* as indexes to select values from *lut*.
The returned data has shape data.shape + lut.shape[1:]
Uses scipy.weave to improve performance if it is available.
Note: color gradient lookup tables can be generated using GradientWidget.
"""
if data.dtype.kind not in ('i', 'u'):
data = data.astype(int)
## using np.take appears to be faster than even the scipy.weave method and takes care of clipping as well.
return np.take(lut, data, axis=0, mode='clip')
### old methods:
#data = np.clip(data, 0, lut.shape[0]-1)
#try:
#if not USE_WEAVE:
#raise Exception('Weave is disabled; falling back to slower version.')
### number of values to copy for each LUT lookup
#if lut.ndim == 1:
#ncol = 1
#else:
#ncol = sum(lut.shape[1:])
### output array
#newData = np.empty((data.size, ncol), dtype=lut.dtype)
### flattened input arrays
#flatData = data.flatten()
#flatLut = lut.reshape((lut.shape[0], ncol))
#dataSize = data.size
### strides for accessing each item
#newStride = newData.strides[0] / newData.dtype.itemsize
#lutStride = flatLut.strides[0] / flatLut.dtype.itemsize
#dataStride = flatData.strides[0] / flatData.dtype.itemsize
### strides for accessing individual values within a single LUT lookup
#newColStride = newData.strides[1] / newData.dtype.itemsize
#lutColStride = flatLut.strides[1] / flatLut.dtype.itemsize
#code = """
#for( int i=0; i<dataSize; i++ ) {
#for( int j=0; j<ncol; j++ ) {
#newData[i*newStride + j*newColStride] = flatLut[flatData[i*dataStride]*lutStride + j*lutColStride];
#}
#}
#"""
#scipy.weave.inline(code, ['flatData', 'flatLut', 'newData', 'dataSize', 'ncol', 'newStride', 'lutStride', 'dataStride', 'newColStride', 'lutColStride'])
#newData = newData.reshape(data.shape + lut.shape[1:])
##if np.any(newData != lut[data]):
##print "mismatch!"
#data = newData
#except:
#if USE_WEAVE:
#debug.printExc("Error; disabling weave.")
#USE_WEAVE = False
#data = lut[data]
#return data
def makeRGBA(*args, **kwds):
"""Equivalent to makeARGB(..., useRGBA=True)"""
@ -1056,6 +1121,45 @@ def colorToAlpha(data, color):
#raise Exception()
return np.clip(output, 0, 255).astype(np.ubyte)
def gaussianFilter(data, sigma):
"""
Drop-in replacement for scipy.ndimage.gaussian_filter.
(note: results are only approximately equal to the output of
gaussian_filter)
"""
if np.isscalar(sigma):
sigma = (sigma,) * data.ndim
baseline = data.mean()
filtered = data - baseline
for ax in range(data.ndim):
s = sigma[ax]
if s == 0:
continue
# generate 1D gaussian kernel
ksize = int(s * 6)
x = np.arange(-ksize, ksize)
kernel = np.exp(-x**2 / (2*s**2))
kshape = [1,] * data.ndim
kshape[ax] = len(kernel)
kernel = kernel.reshape(kshape)
# convolve as product of FFTs
shape = data.shape[ax] + ksize
scale = 1.0 / (abs(s) * (2*np.pi)**0.5)
filtered = scale * np.fft.irfft(np.fft.rfft(filtered, shape, axis=ax) *
np.fft.rfft(kernel, shape, axis=ax),
axis=ax)
# clip off extra data
sl = [slice(None)] * data.ndim
sl[ax] = slice(filtered.shape[ax]-data.shape[ax],None,None)
filtered = filtered[sl]
return filtered + baseline
def downsample(data, n, axis=0, xvals='subsample'):
"""Downsample by averaging points together across axis.
If multiple axes are specified, runs once per axis.
@ -1473,7 +1577,11 @@ def traceImage(image, values, smooth=0.5):
If image is RGB or RGBA, then the shape of values should be (nvals, 3/4)
The parameter *smooth* is expressed in pixels.
"""
try:
import scipy.ndimage as ndi
except ImportError:
raise Exception("traceImage() requires the package scipy.ndimage, but it is not importable.")
if values.ndim == 2:
values = values.T
values = values[np.newaxis, np.newaxis, ...].astype(float)
@ -1487,7 +1595,7 @@ def traceImage(image, values, smooth=0.5):
paths = []
for i in range(diff.shape[-1]):
d = (labels==i).astype(float)
d = ndi.gaussian_filter(d, (smooth, smooth))
d = gaussianFilter(d, (smooth, smooth))
lines = isocurve(d, 0.5, connected=True, extendToEdge=True)
path = QtGui.QPainterPath()
for line in lines:
@ -1967,14 +2075,16 @@ def invertQTransform(tr):
bugs in that method. (specifically, Qt has floating-point precision issues
when determining whether a matrix is invertible)
"""
if not HAVE_SCIPY:
try:
import numpy.linalg
arr = np.array([[tr.m11(), tr.m12(), tr.m13()], [tr.m21(), tr.m22(), tr.m23()], [tr.m31(), tr.m32(), tr.m33()]])
inv = numpy.linalg.inv(arr)
return QtGui.QTransform(inv[0,0], inv[0,1], inv[0,2], inv[1,0], inv[1,1], inv[1,2], inv[2,0], inv[2,1])
except ImportError:
inv = tr.inverted()
if inv[1] is False:
raise Exception("Transform is not invertible.")
return inv[0]
arr = np.array([[tr.m11(), tr.m12(), tr.m13()], [tr.m21(), tr.m22(), tr.m23()], [tr.m31(), tr.m32(), tr.m33()]])
inv = scipy.linalg.inv(arr)
return QtGui.QTransform(inv[0,0], inv[0,1], inv[0,2], inv[1,0], inv[1,1], inv[1,2], inv[2,0], inv[2,1])
def pseudoScatter(data, spacing=None, shuffle=True, bidir=False):

View File

@ -651,13 +651,12 @@ class PlotDataItem(GraphicsObject):
def _fourierTransform(self, x, y):
## Perform fourier transform. If x values are not sampled uniformly,
## then use interpolate.griddata to resample before taking fft.
## then use np.interp to resample before taking fft.
dx = np.diff(x)
uniform = not np.any(np.abs(dx-dx[0]) > (abs(dx[0]) / 1000.))
if not uniform:
import scipy.interpolate as interp
x2 = np.linspace(x[0], x[-1], len(x))
y = interp.griddata(x, y, x2, method='linear')
y = np.interp(x2, x, y)
x = x2
f = np.fft.fft(y) / len(y)
y = abs(f[1:len(f)/2])

View File

@ -13,11 +13,8 @@ of how to build an ROI at the bottom of the file.
"""
from ..Qt import QtCore, QtGui
#if not hasattr(QtCore, 'Signal'):
#QtCore.Signal = QtCore.pyqtSignal
import numpy as np
from numpy.linalg import norm
import scipy.ndimage as ndimage
#from numpy.linalg import norm
from ..Point import *
from ..SRTTransform import SRTTransform
from math import cos, sin
@ -1086,105 +1083,6 @@ class ROI(GraphicsObject):
mapped = fn.transformCoordinates(img.transform(), coords)
return result, mapped
### transpose data so x and y are the first 2 axes
#trAx = range(0, data.ndim)
#trAx.remove(axes[0])
#trAx.remove(axes[1])
#tr1 = tuple(axes) + tuple(trAx)
#arr = data.transpose(tr1)
### Determine the minimal area of the data we will need
#(dataBounds, roiDataTransform) = self.getArraySlice(data, img, returnSlice=False, axes=axes)
### Pad data boundaries by 1px if possible
#dataBounds = (
#(max(dataBounds[0][0]-1, 0), min(dataBounds[0][1]+1, arr.shape[0])),
#(max(dataBounds[1][0]-1, 0), min(dataBounds[1][1]+1, arr.shape[1]))
#)
### Extract minimal data from array
#arr1 = arr[dataBounds[0][0]:dataBounds[0][1], dataBounds[1][0]:dataBounds[1][1]]
### Update roiDataTransform to reflect this extraction
#roiDataTransform *= QtGui.QTransform().translate(-dataBounds[0][0], -dataBounds[1][0])
#### (roiDataTransform now maps from ROI coords to extracted data coords)
### Rotate array
#if abs(self.state['angle']) > 1e-5:
#arr2 = ndimage.rotate(arr1, self.state['angle'] * 180 / np.pi, order=1)
### update data transforms to reflect this rotation
#rot = QtGui.QTransform().rotate(self.state['angle'] * 180 / np.pi)
#roiDataTransform *= rot
### The rotation also causes a shift which must be accounted for:
#dataBound = QtCore.QRectF(0, 0, arr1.shape[0], arr1.shape[1])
#rotBound = rot.mapRect(dataBound)
#roiDataTransform *= QtGui.QTransform().translate(-rotBound.left(), -rotBound.top())
#else:
#arr2 = arr1
#### Shift off partial pixels
## 1. map ROI into current data space
#roiBounds = roiDataTransform.mapRect(self.boundingRect())
## 2. Determine amount to shift data
#shift = (int(roiBounds.left()) - roiBounds.left(), int(roiBounds.bottom()) - roiBounds.bottom())
#if abs(shift[0]) > 1e-6 or abs(shift[1]) > 1e-6:
## 3. pad array with 0s before shifting
#arr2a = np.zeros((arr2.shape[0]+2, arr2.shape[1]+2) + arr2.shape[2:], dtype=arr2.dtype)
#arr2a[1:-1, 1:-1] = arr2
## 4. shift array and udpate transforms
#arr3 = ndimage.shift(arr2a, shift + (0,)*(arr2.ndim-2), order=1)
#roiDataTransform *= QtGui.QTransform().translate(1+shift[0], 1+shift[1])
#else:
#arr3 = arr2
#### Extract needed region from rotated/shifted array
## 1. map ROI into current data space (round these values off--they should be exact integer values at this point)
#roiBounds = roiDataTransform.mapRect(self.boundingRect())
##print self, roiBounds.height()
##import traceback
##traceback.print_stack()
#roiBounds = QtCore.QRect(round(roiBounds.left()), round(roiBounds.top()), round(roiBounds.width()), round(roiBounds.height()))
##2. intersect ROI with data bounds
#dataBounds = roiBounds.intersect(QtCore.QRect(0, 0, arr3.shape[0], arr3.shape[1]))
##3. Extract data from array
#db = dataBounds
#bounds = (
#(db.left(), db.right()+1),
#(db.top(), db.bottom()+1)
#)
#arr4 = arr3[bounds[0][0]:bounds[0][1], bounds[1][0]:bounds[1][1]]
#### Create zero array in size of ROI
#arr5 = np.zeros((roiBounds.width(), roiBounds.height()) + arr4.shape[2:], dtype=arr4.dtype)
### Fill array with ROI data
#orig = Point(dataBounds.topLeft() - roiBounds.topLeft())
#subArr = arr5[orig[0]:orig[0]+arr4.shape[0], orig[1]:orig[1]+arr4.shape[1]]
#subArr[:] = arr4[:subArr.shape[0], :subArr.shape[1]]
### figure out the reverse transpose order
#tr2 = np.array(tr1)
#for i in range(0, len(tr2)):
#tr2[tr1[i]] = i
#tr2 = tuple(tr2)
### Untranspose array before returning
#return arr5.transpose(tr2)
def getAffineSliceParams(self, data, img, axes=(0,1)):
"""
Returns the parameters needed to use :func:`affineSlice <pyqtgraph.affineSlice>` to

View File

@ -0,0 +1,64 @@
import pyqtgraph as pg
import numpy as np
from numpy.testing import assert_array_almost_equal, assert_almost_equal
np.random.seed(12345)
def testSolve3D():
p1 = np.array([[0,0,0,1],
[1,0,0,1],
[0,1,0,1],
[0,0,1,1]], dtype=float)
# transform points through random matrix
tr = np.random.normal(size=(4, 4))
tr[3] = (0,0,0,1)
p2 = np.dot(tr, p1.T).T[:,:3]
# solve to see if we can recover the transformation matrix.
tr2 = pg.solve3DTransform(p1, p2)
assert_array_almost_equal(tr[:3], tr2[:3])
def test_interpolateArray():
data = np.array([[ 1., 2., 4. ],
[ 10., 20., 40. ],
[ 100., 200., 400.]])
x = np.array([[ 0.3, 0.6],
[ 1. , 1. ],
[ 0.5, 1. ],
[ 0.5, 2.5],
[ 10. , 10. ]])
result = pg.interpolateArray(data, x)
import scipy.ndimage
spresult = scipy.ndimage.map_coordinates(data, x.T, order=1)
assert_array_almost_equal(result, spresult)
# test mapping when x.shape[-1] < data.ndim
x = np.array([[ 0.3, 0],
[ 0.3, 1],
[ 0.3, 2]])
r1 = pg.interpolateArray(data, x)
r2 = pg.interpolateArray(data, x[0,:1])
assert_array_almost_equal(r1, r2)
# test mapping 2D array of locations
x = np.array([[[0.5, 0.5], [0.5, 1.0], [0.5, 1.5]],
[[1.5, 0.5], [1.5, 1.0], [1.5, 1.5]]])
r1 = pg.interpolateArray(data, x)
r2 = scipy.ndimage.map_coordinates(data, x.transpose(2,0,1), order=1)
assert_array_almost_equal(r1, r2)
if __name__ == '__main__':
test_interpolateArray()

View File

@ -0,0 +1,39 @@
import pyqtgraph as pg
from pyqtgraph.Qt import QtCore, QtGui
import numpy as np
from numpy.testing import assert_array_almost_equal, assert_almost_equal
testPoints = np.array([
[0, 0, 0],
[1, 0, 0],
[0, 1, 0],
[0, 0, 1],
[-1, -1, 0],
[0, -1, -1]])
def testMatrix():
"""
SRTTransform3D => Transform3D => SRTTransform3D
"""
tr = pg.SRTTransform3D()
tr.setRotate(45, (0, 0, 1))
tr.setScale(0.2, 0.4, 1)
tr.setTranslate(10, 20, 40)
assert tr.getRotation() == (45, QtGui.QVector3D(0, 0, 1))
assert tr.getScale() == QtGui.QVector3D(0.2, 0.4, 1)
assert tr.getTranslation() == QtGui.QVector3D(10, 20, 40)
tr2 = pg.Transform3D(tr)
assert np.all(tr.matrix() == tr2.matrix())
# This is the most important test:
# The transition from Transform3D to SRTTransform3D is a tricky one.
tr3 = pg.SRTTransform3D(tr2)
assert_array_almost_equal(tr.matrix(), tr3.matrix())
assert_almost_equal(tr3.getRotation()[0], tr.getRotation()[0])
assert_array_almost_equal(tr3.getRotation()[1], tr.getRotation()[1])
assert_array_almost_equal(tr3.getScale(), tr.getScale())
assert_array_almost_equal(tr3.getTranslation(), tr.getTranslation())