def show_rows_as_images(matfile,
imgheight,
imgwidth,
nrows=10,
ncols=10,
figtitle=""):
"""Will open a .pmat .dmat .amat or .vmat file and consider the beginning of each row a imgheight x imgwidth imagette.
These images will be interactively displayed in a nrows x ncols grid of imagettes."""
data = None
if figtitle == "":
figtitle = matfile
if matfile.endswith(".pmat"):
# Use pure python implementation of pmat (faster loading)
from plearn.vmat.PMat import PMat
data = PMat(matfile)
else:
# Use of VMat through the Python-bridge
from plearn.pyext import AutoVMatrix
data = AutoVMatrix(filename=matfile)
showRowsAsImages(data,
img_height=imgheight,
img_width=imgwidth,
nrows=nrows,
ncols=ncols,
figtitle=figtitle)
def computeAndShowFilters(datapmatfile,
img_height,
img_width,
filtertype='PCA',
lambd=1e-6,
nu=0,
centered=True,
displaytranspose=False):
"""
Input is considered to be the first img_height x img_width columns of datapmatfile.
centered indicates whether we compte centered covariance (default) or uncentered covariance.
Covariance matrix will get lambd*I added to its diagonal, and its off-diagonal terms multiplied by (1-nu).
Filtertype can be 'PCA' or 'denoising' or 'denoising_eig'.
Original version of linear denoising with zeroing noise (with probability of zeroing equal to nu)
is obtained for centered=False and lambda=0
"""
data = load_pmat_as_array(datapmatfile)
inputs = data[:, 0:img_height * img_width]
C = mycov(inputs, centered)
if (filtertype == "PCA"):
filters = computePCAFiltersFromCovariance(C, lambd, nu)
elif (filtertype == "denoising"):
filters = computeDenoisingFiltersFromCovariance(C, lambd, nu)
elif (filtertype == "denoising_eig"):
filters = computeDenoisingEigenFiltersFromCovariance(C, lambd, nu)
else:
raise ValueError("Invalid filtertype " + filtertype)
if displaytranspose:
filters = filters.T
showRowsAsImages(filters, img_height, img_width, figtitle="Filters")
def show_rows_as_images(matfile, imgheight, imgwidth, nrows=10, ncols=10, figtitle=""):
"""Will open a .pmat .dmat .amat or .vmat file and consider the beginning of each row a imgheight x imgwidth imagette.
These images will be interactively displayed in a nrows x ncols grid of imagettes."""
data = None
if figtitle=="":
figtitle = matfile
if matfile.endswith(".pmat"):
# Use pure python implementation of pmat (faster loading)
from plearn.vmat.PMat import PMat
data = PMat(matfile)
else:
# Use of VMat through the Python-bridge
from plearn.pyext import AutoVMatrix
data = AutoVMatrix(filename=matfile)
showRowsAsImages(data, img_height=imgheight, img_width=imgwidth, nrows=nrows, ncols=ncols, figtitle=figtitle)
def computeAndShowFilters(datapmatfile, img_height, img_width, filtertype='PCA', lambd=1e-6, nu=0, centered=True, displaytranspose=False):
"""
Input is considered to be the first img_height x img_width columns of datapmatfile.
centered indicates whether we compte centered covariance (default) or uncentered covariance.
Covariance matrix will get lambd*I added to its diagonal, and its off-diagonal terms multiplied by (1-nu).
Filtertype can be 'PCA' or 'denoising' or 'denoising_eig'.
Original version of linear denoising with zeroing noise (with probability of zeroing equal to nu)
is obtained for centered=False and lambda=0
"""
data = load_pmat_as_array(datapmatfile)
inputs = data[:,0:img_height*img_width]
C = mycov(inputs, centered)
if(filtertype=="PCA"):
filters = computePCAFiltersFromCovariance(C, lambd, nu)
elif(filtertype=="denoising"):
filters = computeDenoisingFiltersFromCovariance(C, lambd, nu)
elif(filtertype=="denoising_eig"):
filters = computeDenoisingEigenFiltersFromCovariance(C, lambd, nu)
else:
raise ValueError("Invalid filtertype "+filtertype)
if displaytranspose:
filters = filters.T
showRowsAsImages(filters, img_height, img_width, figtitle="Filters")