def __init__(self, xmin, xmax, ymin, ymax, binsize, regfilter, expomap, eventfile):
self.xmin = xmin
self.xmax = xmax
self.ymin = ymin
self.ymax = ymax
assert self.ymax > self.ymin
assert self.xmax > self.xmin
# At least two bins!
assert binsize < (self.xmax - self.xmin) / 2.0
assert binsize < (self.ymax - self.ymin) / 2.0
self.binsize = binsize
self.xbins = numpy.arange(self.xmin, self.xmax, self.binsize)
self.ybins = numpy.arange(self.ymin, self.ymax, self.binsize)
# Prepare mask
self.xcs = (self.xbins[:-1] + self.xbins[1:]) / 2.0
self.ycs = (self.ybins[:-1] + self.ybins[1:]) / 2.0
points = cartesian.cartesian([self.xcs, self.ycs])
# plt.imshow(exposures)
self.mask = numpy.zeros((self.ycs.shape[0], self.xcs.shape[0]), dtype=bool)
for i in range(self.mask.shape[0]):
for j in range(self.mask.shape[1]):
self.mask[i, j] = regfilter.inside1(self.xcs[j], self.ycs[i])
expo = ExposureMap.Exposure(expomap)
wcs = XMMWCS.XMMWCS(eventfile)
skypoints = wcs.xy2sky(points)
ras = skypoints[:, 0]
decs = skypoints[:, 1]
exposures = expo.getExposureAtSky(ras, decs).reshape(self.mask.T.shape).T
idx = exposures <= (exposures.max() / 2.0)
self.mask[idx] = 0
self.empty = ~self.mask
def build_weights_matrix(Y):
(n, m) = [Y.shape[0], Y.shape[1]]
S = std_matrix(Y)
size = n * m
cart = cartesian.cartesian([range(n), range(m)])
cart_r = cart.reshape(n, m, 2)
xy2idx = np.arange(size).reshape(n, m) # [x,y] -> index
W = sparse.lil_matrix((size, size))
for i in range(Y.shape[0]):
for j in range(Y.shape[1]):
idx = xy2idx[i, j]
N = neighborhood(cart_r, [i, j]).reshape(-1, 2)
N = [tuple(neighbor) for neighbor in N]
N.remove((i, j))
p_idx = [xy2idx[xy] for xy in N]
weights = weight((i, j), N, Y, S)
W[idx, p_idx] = -1 * np.asmatrix(weights)
Wn = normalize(W, norm='l1', axis=1)
Wn[np.arange(size), np.arange(size)] = 1
return Wn
def build_weights_matrix(Y):
(n, m) = [Y.shape[0], Y.shape[1]]
S = std_matrix(Y)
size = n * m
cart = cartesian.cartesian([xrange(n), xrange(m)])
cart_r = cart.reshape(n, m, 2)
xy2idx = np.arange(size).reshape(n, m) # [x,y] -> index
W = sparse.lil_matrix((size, size))
for i in xrange(Y.shape[0]):
for j in xrange(Y.shape[1]):
idx = xy2idx[i, j]
N = neighborhood(cart_r, [i, j]).reshape(-1, 2)
N = [tuple(neighbor) for neighbor in N]
N.remove((i, j))
p_idx = [xy2idx[xy] for xy in N]
weights = weight((i, j), N, Y, S)
W[idx, p_idx] = -1 * np.asmatrix(weights)
Wn = normalize(W, norm='l1', axis=1)
Wn[np.arange(size), np.arange(size)] = 1
return Wn
def build_weights_matrix(y):
(n, m) = [y.shape[0], y.shape[1]]
s = std_matrix(y)
size = n * m
cart = cartesian.cartesian([xrange(n), xrange(m)])
cart_r = cart.reshape(n, m, 2)
xy2idx = np.arange(size).reshape(n, m) # [x,y] -> index
w = sparse.lil_matrix((size, size))
for i in xrange(y.shape[0]):
for j in xrange(y.shape[1]):
idx = xy2idx[i, j]
n = neighborhood(cart_r, [i, j]).reshape(-1, 2)
n = [tuple(neighbor) for neighbor in n]
n.remove((i, j))
p_idx = [xy2idx[xy] for xy in n]
weights = weight((i, j), n, y, s)
w[idx, p_idx] = -1 * np.asmatrix(weights)
Wn = normalize(w, norm='l1', axis=1)
Wn[np.arange(size), np.arange(size)] = 1
return Wn
def plot(self, xbins, ybins, **kwargs):
fig, sub = plt.subplots(1, 1)
_ = sub.hist2d(self.X, self.Y, [xbins, ybins])
if len(self.centroids) > 0:
cc = numpy.array(self.centroids)
sub.scatter(cc[:, 0],
cc[:, 1],
s=80,
alpha=0.8,
facecolors='none',
edgecolors='r',
lw=2)
if 'filter' in kwargs.keys():
xcs = (xbins[:-1] + xbins[1:]) / 2.0
ycs = (ybins[:-1] + ybins[1:]) / 2.0
points = cartesian.cartesian([xcs, ycs])
idx = kwargs['filter'].inside(points[:, 0], points[:, 1])
sub.plot(points[~idx, 0],
points[~idx, 1],
'x',
markersize=8,
color='white')
#for cell in self.cells.flatten():
# sub.add_patch( Rectangle((cell.x, cell.y), cell.w, cell.h, color='red',alpha=0.5, fill=False) )
return fig
def neighbourDirections(dimensions): arrays = [np.asarray([-1,0,1]) for i in range(dimensions)] return np.delete(cartesian(arrays), (3**dimensions)/2,0)
def regrid(sourceimage,targetimage,fillval = NAN,theader = 0,tpix = []):
"""
This takes sourceimage and puts it onto targetimage grid, using wcs
solutions for both. Grid points with no info from sourceimage are
set to fillval. Requires scipy0.14.0
sourceimage: path to image to regrid - should already be convolved
to appropriate resolution using resconvolve
targetimage: path to image whose grid is to be used in regridding
fillval: value to give to empty grid positions
(kwarg, default = NAN)
theader: specify a header containing WCS solution to use
(kwarg, default = 0)
Returns array with targetimage dimensions.
"""
# Start timer
start = time.time()
# Load in source data and header information
sdata,sheader = fits.getdata(sourceimage,header = True)
# Create WCS object for the source image
sourcewcs = wcs.WCS(sheader)
# Create array of pixel indices in source image
x = arange(sdata.shape[1])
y = arange(sdata.shape[0])
# Interpolate the image data over pixel indices
interp = scipy.interpolate.RectBivariateSpline(y,x,sdata)
# Load in target grid data
if theader == 0 or tpix == []:
tdata,theader = fits.getdata(targetimage,header=True)
# Create all possible pairs of pixel coordinates in target grid
coords = cartesian([arange(tdata.shape[1]),arange(tdata.shape[0])])
elif theader != 0 and tpix != []:
assert len(tpix) == 4
dx,ux,dy,uy = tpix
# Create all possible pairs of pixel coordinates in target grid
tdata = zeros((uy-dy,ux-dx))
coords = cartesian([arange(dx,ux),arange(dy,uy)])
# Extract x and y columns of pixel pairs
xpixs = coords[:,0]
ypixs= coords[:,1]
# Create WCS object for target grid
targetwcs = wcs.WCS(theader)
# Convert target grid pixels to ra/dec
world = targetwcs.wcs_pix2world(coords,0)
# Convert target grid ra/dec to source pixel coordinates
dpix = sourcewcs.wcs_world2pix(world,0)
# Extract x and y columns of converted pixel pairs
xdpixs = dpix[:,0]
ydpixs = dpix[:,1]
# Find where target grid corresponds to actual source image data
good = where((xdpixs >= min(x)) & (xdpixs <= max(x)) &
(ydpixs >= min(y)) & (ydpixs <= max(y)))
# Pick out only pixels with relevant image data
xpixs = xpixs[good]
ypixs = ypixs[good]
xdpixs = xdpixs[good]
ydpixs = ydpixs[good]
# Create grid to fill up with source image regrid
tofill = copy(tdata)
tofill[:] = fillval
# Choose indices of array positions to be changed
inds = (ypixs,xpixs)
tofill[inds] = interp(ydpixs,xdpixs,grid=False) # needs scipy 0.14.0
# End timer
end = time.time()
# Print time to run
print 'Regridded in ',(end-start)/60.,' min'
return tofill,tdata
def longest_path(self, spur_node_position=0, spur_node=None):
"""
Find the longest path in a multi-partite graph.
This is an implementation of algorithm (1).
spur_node_position : int
Position of the spur node in the multi-partite graph,
0 means the true source node of the graph.
The spur node is used in the k-longest path algorithm
were we find the longest path starting at a specific node
not only from the source node of the graph.
spur_node : np.array
Identifier of the spur node. ex: np.array([11,4,19])
for self.substring_length = 3.
Recall that each node are identified using by a
substring of length self.substring_length
"""
# The "n" of the n-partite graph
n = self.peptide_length - self.substring_length + 1
# The spur node is just before the sink node: spur_node -> t
# The longest path is trivial
if spur_node_position == n:
return spur_node[1:], self.Wt_weight(spur_node[1:], n)
# Arrays for the dynamic programming
length_to = np.zeros(
(2, self.aminoacid_count**self.substring_length),
dtype=PATH_LENGTH_DTYPE) - PATH_LENGTH_DTYPE(np.inf)
predecessor = np.zeros(
(n, self.aminoacid_count**self.substring_length),
dtype=INT_AMINO_ACID_DTYPE)
edge_index = 0
# Edges leaving the source node
if spur_node_position == 0:
a_index = 0
for a in cartesian(
repeat(np.arange(self.aminoacid_count,
dtype=INT_AMINO_ACID_DTYPE),
times=self.substring_length)):
if not self.is_blacklisted(edge_index):
length_to[0, a_index] = self.W_weight(a, 0)
a_index += 1
edge_index += 1
# The source is a spur node
else:
# Keep the count
edge_index += self.aminoacid_count**self.substring_length
# Visit each edges of the n-partite graph using a pre-defined topological order
# Edges are always of the form: a -> s
for i in xrange(1, n):
# Initialize the length_to vector
length_to[i % 2] = length_to[i % 2] - PATH_LENGTH_DTYPE(np.inf)
a_index = 0 # Do not need anymore
s_index = 0
# Edges are not attainable from the spur node.
# i.e. edges on the left side of the spur node
if i < spur_node_position:
# Keep track of the edge index
edge_index += self.aminoacid_count**(self.substring_length + 1)
# Edges at the spur node position
elif i == spur_node_position:
for s in cartesian(
repeat(np.arange(self.aminoacid_count,
dtype=INT_AMINO_ACID_DTYPE),
times=self.substring_length)):
# Weight on the edge a -> s
edge_weight = self.W_weight(s, i)
a_index = (s_index - s[-1]) / self.aminoacid_count
for a_prime in xrange(self.aminoacid_count):
# a is obtained partly from s
a = np.roll(s, 1)
a[0] = a_prime
# Only edges of the form: spur_node -> s
if np.alltrue(
a == spur_node
) and not self.is_blacklisted(edge_index):
# Weight on the edge spur_node -> s
length_to[i % 2, s_index] = edge_weight
predecessor[i, s_index] = a_prime
edge_index += 1
a_index += self.aminoacid_count**(
self.substring_length - 1)
s_index += 1
# Edges on the right side of the spur node
else:
for s in cartesian(
repeat(np.arange(self.aminoacid_count,
dtype=INT_AMINO_ACID_DTYPE),
times=self.substring_length)):
# Weight on the edge a -> s
edge_weight = self.W_weight(s, i)
a_index = (s_index - s[-1]) / self.aminoacid_count
for a_prime in xrange(self.aminoacid_count):
if not self.is_blacklisted(edge_index):
# Edge a -> s
if length_to[i % 2, s_index] < length_to[
(i - 1) % 2, a_index] + edge_weight:
length_to[i % 2, s_index] = length_to[
(i - 1) % 2, a_index] + edge_weight
predecessor[i, s_index] = a_prime
edge_index += 1
a_index += self.aminoacid_count**(
self.substring_length - 1)
s_index += 1
length_to_t = PATH_LENGTH_DTYPE(-np.inf)
predecessor_of_t = None
a_index = 0
# Edges heading to the sink node
for a in cartesian(
repeat(np.arange(self.aminoacid_count,
dtype=INT_AMINO_ACID_DTYPE),
times=self.substring_length)):
# Their is no point at blacklisting an edge heading
# to the sink node but it's possible!
if not self.is_blacklisted(edge_index):
edge_weight = self.Wt_weight(a[1:], n)
if length_to_t < length_to[(n - 1) % 2, a_index] + edge_weight:
length_to_t = length_to[(n - 1) % 2, a_index] + edge_weight
predecessor_of_t = np.array(a, dtype=INT_AMINO_ACID_DTYPE)
a_index += 1
edge_index += 1
# There is no path between the source and the sink
if predecessor_of_t == None:
return np.zeros(0, dtype=INT_AMINO_ACID_DTYPE), -np.Inf
# Initialise the peptide sequence
peptide = np.zeros(self.peptide_length, dtype=INT_AMINO_ACID_DTYPE)
peptide[-self.substring_length:] = predecessor_of_t
# Revert the path using the predecessors to find the peptide
for i in range(self.peptide_length - self.substring_length)[::-1]:
# Compute the prececessor index
predecessor_index = 0
for j in xrange(self.substring_length):
predecessor_index += peptide[i + j +
1] * self.aminoacid_count**(
self.substring_length - j - 1)
# Keep track of the peptide sequence
peptide[i] = predecessor[i + 1, predecessor_index]
# Return the encoded peptide sequence and the binding affinity (path length)
return peptide[spur_node_position:], length_to_t
import cartesian
import argparse
import itertools
import numpy as np
parser = argparse.ArgumentParser()
parser.add_argument('-m', '--min',
help='List of minimum investment amounts for each asset in the portfolio.',
nargs='+',
type=float)
parser.add_argument('-s', '--step',
help='List of step sizes for each asset in the portfolio.',
nargs='+',
type=float)
parser.add_argument('-x', '--max',
help='List of maximum investment amounts for each asset in the portfolio.',
nargs='+',
type=float)
args = parser.parse_args()
num_assets = len(args.min)
possible_portfolio_values = [None]*num_assets
for i in range(num_assets):
possible_portfolio_values[i] = np.arange(args.min[i], args.max[i]+args.step[i], args.step[i], dtype=float)
possible_portfolios = cartesian.cartesian(possible_portfolio_values)
np.savetxt(sys.stdout,possible_portfolios, delimiter=' ')
def regrid(sourceimage, targetimage, fillval=NAN, theader=0, tpix=[]):
"""
This takes sourceimage and puts it onto targetimage grid, using wcs
solutions for both. Grid points with no info from sourceimage are
set to fillval. Requires scipy0.14.0
sourceimage: path to image to regrid - should already be convolved
to appropriate resolution using resconvolve
targetimage: path to image whose grid is to be used in regridding
fillval: value to give to empty grid positions
(kwarg, default = NAN)
theader: specify a header containing WCS solution to use
(kwarg, default = 0)
Returns array with targetimage dimensions.
"""
# Start timer
start = time.time()
# Load in source data and header information
sdata, sheader = fits.getdata(sourceimage, header=True)
# Create WCS object for the source image
sourcewcs = wcs.WCS(sheader)
# Create array of pixel indices in source image
x = arange(sdata.shape[1])
y = arange(sdata.shape[0])
# Interpolate the image data over pixel indices
interp = scipy.interpolate.RectBivariateSpline(y, x, sdata)
# Load in target grid data
if theader == 0 or tpix == []:
tdata, theader = fits.getdata(targetimage, header=True)
# Create all possible pairs of pixel coordinates in target grid
coords = cartesian([arange(tdata.shape[1]), arange(tdata.shape[0])])
elif theader != 0 and tpix != []:
assert len(tpix) == 4
dx, ux, dy, uy = tpix
# Create all possible pairs of pixel coordinates in target grid
tdata = zeros((uy - dy, ux - dx))
coords = cartesian([arange(dx, ux), arange(dy, uy)])
# Extract x and y columns of pixel pairs
xpixs = coords[:, 0]
ypixs = coords[:, 1]
# Create WCS object for target grid
targetwcs = wcs.WCS(theader)
# Convert target grid pixels to ra/dec
world = targetwcs.wcs_pix2world(coords, 0)
# Convert target grid ra/dec to source pixel coordinates
dpix = sourcewcs.wcs_world2pix(world, 0)
# Extract x and y columns of converted pixel pairs
xdpixs = dpix[:, 0]
ydpixs = dpix[:, 1]
# Find where target grid corresponds to actual source image data
good = where((xdpixs >= min(x)) & (xdpixs <= max(x)) & (ydpixs >= min(y))
& (ydpixs <= max(y)))
# Pick out only pixels with relevant image data
xpixs = xpixs[good]
ypixs = ypixs[good]
xdpixs = xdpixs[good]
ydpixs = ydpixs[good]
# Create grid to fill up with source image regrid
tofill = copy(tdata)
tofill[:] = fillval
# Choose indices of array positions to be changed
inds = (ypixs, xpixs)
tofill[inds] = interp(ydpixs, xdpixs, grid=False) # needs scipy 0.14.0
# End timer
end = time.time()
# Print time to run
print 'Regridded in ', (end - start) / 60., ' min'
return tofill, tdata
