from __future__ import print_function
import numpy as nm
try:
import matplotlib.pyplot as plt
import matplotlib as mpl
except (ImportError, RuntimeError):
plt = mpl = None
#print 'matplotlib import failed!'
from sfepy.base.base import output, pause
[docs]
def spy(mtx, eps=None, color='b', **kwargs):
"""
Show sparsity structure of a `scipy.sparse` matrix.
"""
aux = mtx.tocoo()
ij, val = nm.concatenate((aux.row[:,nm.newaxis],
aux.col[:,nm.newaxis]), 1), aux.data
n_item = aux.getnnz()
n_row, n_col = aux.shape
if eps is not None:
output('using eps =', eps)
ij = nm.compress(nm.absolute(val) > eps, ij, 0)
n_item = ij.shape[0]
else:
output('showing all')
output('n_item:', n_item)
if n_item:
args = {'marker' : '.'}
args.update(kwargs)
plt.plot(ij[:,1], ij[:,0], color, linestyle='None', **args)
plt.axis('image')
plt.axis([-0.5, n_row+0.5, -0.5, n_col+0.5])
plt.xlabel(r'%d x %d: %d nnz, %.2f%% fill'
% (n_row, n_col, n_item, 100. * n_item /
(float(n_row) * float(n_col))))
ax = plt.gca()
ax.set_ylim(ax.get_ylim()[::-1])
[docs]
def spy_and_show(mtx, **kwargs):
spy(mtx, **kwargs)
plt.show()
##
# 13.12.2005, c
[docs]
def print_matrix_diff(title, legend, mtx1, mtx2, mtx_da, mtx_dr, iis):
print('%s: ir, ic, %s, %s, adiff, rdiff' % ((title,) + tuple(legend)))
aux = mtx_da.copy().tocsc() # mtx_da should be CSC, cast for safety anyway.
aux.data = nm.ones(mtx_da.data.shape[0])
ics, irs = aux.nonzero()
for ii in iis:
ir, ic = irs[ii], ics[ii]
print('%5d %5d %11.4e %11.4e %9.2e %9.2e'
% (ir, ic, mtx1[ir,ic], mtx2[ir,ic],
mtx_da[ir,ic], mtx_dr[ir,ic]))
print('total: %d' % len(iis))
##
# 13.12.2005, c
# 14.12.2005
# 15.12.2005
# 18.07.2007
[docs]
def plot_matrix_diff(mtx1, mtx2, delta, legend, mode):
eps = 1e-16
print("min", legend[0] , legend[1], ":", nm.amin(mtx1.data), nm.amin(mtx2.data))
print("max", legend[0] , legend[1], ":", nm.amax(mtx1.data), nm.amax(mtx2.data))
mtx_da = mtx1.copy() # To preserve structure of mtx1.
mtx_da.data[:] = nm.abs(mtx1.data - mtx2.data)
mtx_dr = mtx_da.copy()
mtx_dr.data[:] = -1
iin = nm.where(nm.abs(mtx1.data) > eps)[0]
mtx_dr.data[iin] = mtx_da.data[iin] / nm.abs(mtx1.data[iin])
print("err abs min max:", nm.amin(mtx_da.data), nm.amax(mtx_da.data))
print("err rel min max:", nm.amin(mtx_dr.data), nm.amax(mtx_dr.data))
epsilon = max(1e-5, 10 * delta)
print('epsilon:', epsilon)
pause()
ija = nm.where(mtx_da.data > epsilon)[0]
print_matrix_diff('--- absolute diff', legend,
mtx1, mtx2, mtx_da, mtx_dr, ija)
pause()
iin = nm.where(nm.abs(mtx1.data) > epsilon)[0]
ij = nm.where(nm.abs(mtx_dr.data[iin]) > epsilon)[0]
ij = iin[ij]
print_matrix_diff('--- relative diff', legend,
mtx1, mtx2, mtx_da, mtx_dr, ij)
pause()
ijb = nm.intersect1d(ija, ij)
print_matrix_diff('--- a-r', legend,
mtx1, mtx2, mtx_da, mtx_dr, ijb)
pause()
ii = nm.argsort(mtx_dr.data[ijb])
n_s = min(20, len(ii))
ijbs = ijb[ii[-1:-n_s-1:-1]]
print_matrix_diff('--- a-r 20 biggest (by r)', legend,
mtx1, mtx2, mtx_da, mtx_dr, ijbs)
pause()
if mode < 2: return
h = 100
plt.figure(h); plt.clf()
plt.axes([0.04, 0.6, 0.3, 0.3], frameon=True)
spy(mtx_da, epsilon)
plt.title('absolute diff')
plt.axes([0.68, 0.6, 0.3, 0.3], frameon=True)
iia = nm.where(mtx_dr.data)[0]
mtx_dr.data[nm.setdiff1d(iia, iin)] = 0.0
spy(mtx_dr, epsilon)
plt.title('relative diff')
plt.axes([0.36, 0.6, 0.3, 0.3], frameon=True)
mtx = mtx_dr.copy()
mtx.data[:] = 0.0
ii = nm.intersect1d(nm.where(mtx_dr.data > epsilon)[0],
nm.where(mtx_da.data > epsilon)[0])
mtx.data[ii] = 1.0
spy(mtx, epsilon)
plt.title('a-r intersection')
plt.axes([0.04, 0.08, 0.42, 0.42], frameon=True)
spy(mtx1, epsilon)
plt.title(legend[0])
plt.axes([0.54, 0.08, 0.42, 0.42], frameon=True)
spy(mtx2, epsilon)
plt.title(legend[1])
plt.show()
##
# 02.05.2006, c
[docs]
def set_axes_font_size(ax, size):
labels = ax.get_xticklabels() + ax.get_yticklabels()
for label in labels:
label.set_size(size)
##
# 27.09.2006, c
[docs]
def font_size(size):
return mpl.font_manager.FontProperties(size=size)
##
# 28.08.2007, c
[docs]
def iplot(*args, **kwargs):
plt.ion()
plt.plot(*args, **kwargs)
plt.draw()
plt.ioff()
pause()