T = fits_table('pobjs.fits')

print 'Read', len(T), 'objs'

keep = np.ones(len(T), bool)

I,J,d = match_radec(T.ra, T.dec, T.ra, T.dec, 30. * 0.396 / 3600., notself=True)

keep[I] = False

T.cut(keep)

print 'Cut to', len(T), 'isolated'

T.cut(T.probpsf == 1)

print 'Cut to', len(T), 'point sources'

T.cut(T.psfflux_r * np.sqrt(T.psffluxivar_r) > 80.)

print 'Cut to', len(T), 'high S/N'

F = fits_table('fields.fits')

sdss = DR10(basedir='data/unzip')

print sdss.dasurl

band = 'r'

rerun = '301'

dbs = []

for t in T:

f = F[(F.ramin < t.ra) * (F.ramax > t.ra) * (F.decmin < t.dec) * (F.decmax > t.dec)]

print 'Found', len(f), 'fields'

if len(f) == 0:

continue

ims = []

for r,c,fnum in zip(f.run, f.camcol, f.field):

#url = sdss.get_url('frame', r, c, fnum, rerun=rerun, band=band)

fn = sdss.retrieve('frame', r, c, fnum, band=band, rerun=rerun)

print 'got', fn

wcs = Tan(fn)

ok,x,y = wcs.radec2pixelxy(t.ra, t.dec)

x -= 1

y -= 1

print 'x,y', x,y

ix = int(np.round(x))

iy = int(np.round(y))

S = 25

img = fitsio.read(fn)

print 'img', img.shape, img.dtype

im = img[iy-S:iy+S+1, ix-S:ix+S+1]

if im.shape != (2*S+1, 2*S+1):

continue

H,W = im.shape

ims.append(im)

# plt.clf()

# plt.imshow(im, interpolation='nearest', origin='lower')

# ps.savefig()

imax = np.argmax(im)

my,mx = np.unravel_index(imax, im.shape)

Y = im[my, :]

X = im[:, mx]

Fy = np.fft.fft(Y)

Fx = np.fft.fft(X)

Sy = np.fft.fftshift(Fy)

Sx = np.fft.fftshift(Fx)

Px = Sx.real**2 + Sx.imag**2

Py = Sy.real**2 + Sy.imag**2

freqs1 = np.fft.fftfreq(len(Y))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

print 's0', s0

s0 = s0[0]

dbs.append((Px, Py, np.abs(sfreqs1)))

#break

n = len(ims)

cols = int(np.ceil(np.sqrt(n)))

rows = int(np.ceil(n / float(cols)))

plt.clf()

for i,im in enumerate(ims):

plt.subplot(rows, cols, i+1)

plt.imshow(im, interpolation='nearest', origin='lower')

ps.savefig()

plt.clf()

for Px,Py,sfreq in dbs:

plt.plot(sfreq, 10.*np.log10(Py / Py[s0]), 'b-', alpha=0.2)

plt.plot(sfreq, 10.*np.log10(Px / Px[s0]), 'g-', alpha=0.2)

plt.ylabel('dB')

plt.ylim(-50, 5)

c = 20

W,H = 2*c+1, 2*c+1

xx,yy = np.meshgrid(np.arange(W), np.arange(H))

rsq = (xx-c)**2 + (yy-c)**2

dbs = []

for sigma in np.arange(0.1, 1.01, 0.05):

im = np.exp(-0.5 * rsq / sigma**2)

#F = np.fft.fft2(im)

#mx = max(np.abs(F.real).max(), np.abs(F.imag).max())

# plt.clf()

# plt.subplot(1,2, 1)

# plt.imshow(F.real, interpolation='nearest', origin='lower', vmin=-mx, vmax=mx)

# plt.subplot(1,2, 2)

# plt.imshow(F.imag, interpolation='nearest', origin='lower', vmin=-mx, vmax=mx)

# plt.suptitle('Sigma = %.2f' % sigma)

# ps.savefig()

imax = np.argmax(im)

my,mx = np.unravel_index(imax, im.shape)

Y = im[my, :]

X = im[:, mx]

Fy = np.fft.fft(Y)

Fx = np.fft.fft(X)

Sy = np.fft.fftshift(Fy)

Sx = np.fft.fftshift(Fx)

Px = Sx.real**2 + Sx.imag**2

Py = Sy.real**2 + Sy.imag**2

freqs1 = np.fft.fftfreq(len(Y))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

s0 = s0[0]

dbs.append((Px, Py, np.abs(sfreqs1), sigma))

plt.clf()

for Px,Py,sfreq,sigma in dbs:

plt.plot(sfreq, 10.*np.log10(Py / Py[s0]), 'b-', alpha=0.2)

plt.plot(sfreq, 10.*np.log10(Px / Px[s0]), 'g-', alpha=0.2)

plt.text(sfreq[-1], 10.*np.log10(Px[-1] / Px[s0]), '%.2f (FWHM %.2f)' % (sigma, sigma * 2.35), ha='right', va='center')

plt.ylabel('dB')

plt.ylim(-40, 3)

#plt.axhline(-3, color='r', alpha=0.5)

# How much to compress the Airy profile so that it is just

# well-sampled (with unit pixels)

scale = np.pi / 2.

x = np.linspace(-5, 5, 500)

y = (2. * scipy.special.j1(x * scale) / (x * scale))**2

# Its half-width at half-max

hwhm = (1./scale * 1.61633)

print 'hwhm', hwhm

# Gaussian with same HWHM

sigma = 2. * hwhm / 2.35

print 'sigma', sigma

g = np.exp(-0.5 * x**2 / sigma**2)

plt.clf()

plt.plot(x, y, 'b-')

plt.plot(x, g, 'r-', lw=2, alpha=0.5)

plt.axhline(0.5, color='r')

plt.axvline(1./scale * 1.61633, color='b', lw=3, alpha=0.5)

plt.axvline(sigma * 2.35 / 2., color='r')

ps.savefig()

c = 200

W,H = 2*c+1, 2*c+1

xx,yy = np.meshgrid(np.arange(W), np.arange(H))

rr = np.hypot(xx-c, yy-c)

dbs = []

for scale in np.arange(1.0, 2.01, 0.1):

im = (2. * scipy.special.j1(rr * scale) / (rr * scale))**2

im[rr == 0] = 1.

#(im,nil,nil,nil) = scipy.special.airy(rr * scale)

F = np.fft.fft2(im)

F = np.fft.fftshift(F)

#F = np.fft.rfft2(im)

mx = max(np.abs(F.real).max(), np.abs(F.imag).max())

# plt.clf()

# plt.subplot(2,2,1)

# plt.imshow(im, interpolation='nearest', origin='lower')

# plt.subplot(2,2,2)

# plt.plot(rr.ravel(), np.hypot(F.real, F.imag).ravel(),

# 'b.', alpha=0.1)

# plt.axvline((W*scale) / np.pi, color='r')

# plt.subplot(2,2,3)

# plt.imshow(F.real, interpolation='nearest', origin='lower',

# vmin=-mx, vmax=mx)

# plt.subplot(2,2,4)

# plt.imshow(F.imag, interpolation='nearest', origin='lower',

# vmin=-mx, vmax=mx)

# ps.savefig()

imax = np.argmax(im)

my,mx = np.unravel_index(imax, im.shape)

Y = im[my, :]

X = im[:, mx]

Fy = np.fft.fft(Y)

Fx = np.fft.fft(X)

Sy = np.fft.fftshift(Fy)

Sx = np.fft.fftshift(Fx)

Px = Sx.real**2 + Sx.imag**2

Py = Sy.real**2 + Sy.imag**2

freqs1 = np.fft.fftfreq(len(Y))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

s0 = s0[0]

dbs.append((X, Y, Px, Py, np.abs(sfreqs1), scale))

plt.clf()

v = np.linspace(0., 2., 500)

plt.plot(v, 2./np.pi * (np.arccos(v) - v*np.sqrt(1-v**2)), 'b-')

ps.savefig()

plt.clf()

for X,Y,Px,Py,sfreq,scale in dbs:

plt.plot(X, 'b-', alpha=0.2)

plt.plot(Y, 'g-', alpha=0.2)

ps.savefig()

plt.clf()

for X,Y,Px,Py,sfreq,scale in dbs:

plt.plot(sfreq, 10.*np.log10(Py / Py[s0]), 'b-', alpha=0.2)

plt.plot(sfreq, 10.*np.log10(Px / Px[s0]), 'g-', alpha=0.2)

plt.text(sfreq[-1], 10.*np.log10(Px[-1] / Px[s0]), '%.2f' % (scale), ha='right', va='center')

#plt.axvline(scale / np.pi, color='r')

plt.ylabel('dB')

#ps.savefig()

plt.ylim(-50, 5)

S = 25

if pred:

psfs = wise_psf(bands=bands, fftcentral=S*2+1, noplots=True)

print 'psfs:', len(psfs)

(Sx,Sy,Px,Py,sf) = psfs[0]

print 'sx:', len(Sx)

print 'sf', sf

pitch = 0.1

print 'Pitch', pitch

print 'lanczos_test...'

lan3 = lanczos_test(pitch=pitch, srange=(-S-0.5+pitch/2., S+0.5))#, noplots=True)

print 'lanczos_test finished.'

(Slan,Plan,sflan) = lan3[0]

sflan /= pitch

sflan = sflan[:S+1]

Plan = Plan[:S+1]

Slan = Slan[:S+1]

print 'lan3:', len(Slan)

print 'sflan:', sflan

print 'Plan:', Plan

ps.basefn = 'wisestars'

ps.skipto(0)

wfn = 'wbox.fits'

if os.path.exists(wfn):

W = fits_table(wfn)

print 'Read', len(W), 'from', wfn

else:

from wisecat import wise_catalog_radecbox

W = wise_catalog_radecbox(150., 200., 45., 47.)

print 'Got', len(W), 'WISE sources'

W.writeto(wfn)

from unwise_coadd import get_l1b_file, wisedir, zeropointToScale

# plt.clf()

# plt.hist(W.w1snr, 50, histtype='step', color='b')

# plt.hist(W.w1snr[W.w1sat == 0], 50, histtype='step', color='r')

# ps.savefig()

# Find isolated stars...

I,J,d = match_radec(W.ra, W.dec, W.ra, W.dec, 20. * 2.75 / 3600., notself=True)

keep = np.ones(len(W), bool)

keep[J] = False

print np.sum(keep), 'of', len(W), 'have no neighbors'

for iband,band in enumerate(bands):

W.snr = W.get('w%isnr'%band)

W.sat = W.get('w%isat' % band)

Wi = W[keep * (W.sat == 0)]

print len(Wi), 'after cuts'

#(W.snr > 20) *

# plt.clf()

# plt.hist(Wi.snr, 50, histtype='step', color='m')

Wi.cut(Wi.ext_flg == 0)

print 'ext_flg:', len(Wi)

Wi.cut(Wi.na == 0)

print 'na:', len(Wi)

Wi.cut(Wi.nb == 1)

print 'nb:', len(Wi)

# plt.hist(Wi.snr, 50, histtype='step', color='k')

# ps.savefig()

T = Wi[np.argsort(-Wi.snr)]

T.cut(T.snr > 25.)

print 'SN cut:', len(T)

#T = Wi[np.argsort(-Wi.snr)[:25]]

#print 'Coadd S/N:', T.snr

# band = 1

# T = fits_table('w%i.fits' % band)

#

# plt.clf()

# n,b,p = plt.hist(T.get('w%impro' % band), 50, histtype='step', color='b')

# plt.xlabel('w%impro' % band)

#

# print 'blend_ext_flags:', ['0x%x' % x for x in np.unique(T.blend_ext_flags)]

#

# T.cut(T.blend_ext_flags & 0xf == 1)

# print 'Cut to', len(T), 'on blend'

# T.cut(T.get('w%isat' % band) == 0)

# print 'Cut to', len(T), 'on sat'

# T.cut(T.get('w%iflg' % band) == 0)

# print 'Cut to', len(T), 'on flg'

#

# plt.hist(T.get('w%impro' % band), histtype='step', color='r', bins=b)

# ps.savefig()

dbs = []

l1dbs = []

#print 'Coadds', T.coadd_id

for t in T:

print 'coadd', t.coadd_id

#print 'RA,Dec', t.ra, t.dec

coadd = t.coadd_id.replace('_ab41', '').strip()

for base in ['data/unwise', 'data/unwise-nersc']:

dirnm = os.path.join(base, coadd[:3], coadd)

fn = os.path.join(dirnm, 'unwise-%s-w%i-img-m.fits' % (coadd, band))

if os.path.exists(fn):

break

img = fitsio.read(fn)

#print 'img', img.shape, img.dtype

wcs = Tan(fn)

ok,x,y = wcs.radec2pixelxy(t.ra, t.dec)

x -= 1

y -= 1

#print 'x,y', x,y

ix = int(np.round(x))

iy = int(np.round(y))

im = img[iy-S:iy+S+1, ix-S:ix+S+1]

if im.shape != (2*S+1, 2*S+1):

continue

imH,imW = im.shape

imax = np.argmax(im)

my,mx = np.unravel_index(imax, im.shape)

#print 'max pixel:', my,mx, im[my,mx]

#print 'center pixel:', imH/2,imW/2, im[imH/2,imW/2]

if my != imH/2 or mx != imW/2:

continue

# F = np.fft.fft2(im)

# S = np.fft.fftshift(F)

# P = S.real**2 + S.imag**2

# P /= P.max()

# freqs = np.fft.fftfreq(W)

# sfreqs = np.fft.fftshift(freqs)

# ff = np.hypot(sfreqs.reshape(1,W), sfreqs.reshape(W,1))

# print 'ff', ff.shape

#plt.clf()

#plt.plot(ff.ravel(), P.ravel(), 'b.', alpha=0.1)

#ps.savefig()

Y = im[imH/2, :]

X = im[:, imW/2]

Fy = np.fft.fft(Y)

Fx = np.fft.fft(X)

Sy = np.fft.fftshift(Fy)

Sx = np.fft.fftshift(Fx)

Px = Sx.real**2 + Sx.imag**2

Py = Sy.real**2 + Sy.imag**2

# Grab "blank" lines nearby to assess noise

n = im[imH/4, :]

Fn = np.fft.fft(n)

Sn = np.fft.fftshift(Fn)

Pn = Sn.real**2 + Sn.imag**2

freqs1 = np.fft.fftfreq(len(Y))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

s0 = s0[0]

s1 = s0

dbs.append((Sx[s1:], Sy[s1:], Px[s1:], Py[s1:], sfreqs1[s1:], Pn[s1:]))

print 'sx', len(Sx[s1:])

print 'sf', sfreqs1[s1:]

continue

fn = os.path.join(dirnm, 'unwise-%s-w%i-frames.fits' % (coadd, band))

T = fits_table(fn)

print 'Read', len(T), 'from', fn

T.cut(T.included > 0)

T.cut(T.use > 0)

print 'Cut to', len(T), 'used'

T.cut(T.coextent[:,0] < ix - S)

T.cut(T.coextent[:,1] > ix + S)

T.cut(T.coextent[:,2] < iy - S)

T.cut(T.coextent[:,3] > iy + S)

print 'Cut to', len(T), 'containing target pixels'

#ims = []

#for f in T[:10]:

for f in T:

fn = get_l1b_file(wisedir, f.scan_id, f.frame_num, band)

print 'frame', fn

img,hdr = fitsio.read(fn, header=True)

img -= f.sky1

gain = hdr.get('FEBGAIN')

print 'Gain:', gain

sig1 = np.sqrt(1./f.weight)

zpscale = 1. / zeropointToScale(f.zeropoint)

sig1 /= zpscale

print 'Sigma1:', sig1

print 'Sky:', f.sky1

wcs = Sip(fn)

mask = fitsio.read(fn.replace('-int-', '-msk-') + '.gz')

#print 'RA,Dec', t.ra, t.dec

ok,x,y = wcs.radec2pixelxy(t.ra, t.dec)

#print 'x,y', x,y

x -= 1

y -= 1

fx = int(np.round(x))

fy = int(np.round(y))

im = img[fy-S:fy+S+1, fx-S:fx+S+1]

if im.shape != (2*S+1, 2*S+1):

continue

imH,imW = im.shape

mask = mask[fy-S:fy+S+1, fx-S:fx+S+1]

ok = (mask == 0)

if not patch_image(im, ok):

print 'Patching failed'

continue

Y = im[imH/2, :]

X = im[:, imW/2]

Fy = np.fft.fft(Y)

Fx = np.fft.fft(X)

Sy = np.fft.fftshift(Fy)

Sx = np.fft.fftshift(Fx)

Px = Sx.real**2 + Sx.imag**2

Py = Sy.real**2 + Sy.imag**2

freqs1 = np.fft.fftfreq(len(Y))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

s0 = s0[0]

# Grab "blank" lines nearby to assess noise

n = im[imH/4, :]

Fn = np.fft.fft(n)

Sn = np.fft.fftshift(Fn)

Pn = Sn.real**2 + Sn.imag**2

l1dbs.append((Px[s0+1:], Py[s0+1:], sfreqs1[s0+1:], Pn[s0+1:]))

#ims.append(im)

# plt.clf()

# for i,im in enumerate(ims[:9]):

# plt.subplot(3,3,1+i)

# plt.imshow(im, interpolation='nearest', origin='lower')

# ps.savefig()

print len(dbs), 'coadds'

print len(l1dbs), 'L1bs'

plt.figure(figsize=(4,4))

left = 0.17

right = 0.03

bot = 0.12

top = 0.03

plt.subplots_adjust(left=left, right=1.-right, bottom=bot, top=1.-top)

if pred:

(Sxpsf, Sypsf, Pxpsf, Pypsf, sfpsf) = psfs[iband]

plt.clf()

Spsf = (Sxpsf + Sypsf) / 2.

Ppsf = Spsf.real**2 + Spsf.imag**2

plt.plot(sfpsf, 10.*np.log10(Pxpsf / Pxpsf[0]), 'r-')

plt.plot(sfpsf, 10.*np.log10(Pypsf / Pypsf[0]), 'm-')

plt.plot(sfpsf, 10.*np.log10(Ppsf / Ppsf[0]), '-', color=(1.,0.6,0))

plt.plot(sflan, 10.*np.log10(Plan / Plan[0]), 'k-')

#print 'sflan', sflan

#print 'Plan', Plan

#print 'db', 10.*np.log10(Plan / Plan[0])

for Sx,Sy,Px,Py,sfreq,Pn in dbs:

Sm = (Sx + Sy) / 2.

P = Sm.real**2 + Sm.imag**2

plt.plot(sfreq, 10.*np.log10(P / P[0]), 'b-', alpha=0.1)

Spred = Spsf * Slan

Ppred = Spred.real**2 + Spred.imag**2

plt.plot(sfreq, 10.*np.log10(Ppred / Ppred[0]), 'g-')

plt.ylabel('dB')

plt.ylim(-40, 3)

plt.title('PSF')

ps.savefig()

plt.clf()

for Sx,Sy,Px,Py,sfreq,Pn in dbs:

plt.plot(sfreq, 10.*np.log10(Py / Py[0]), 'b-', alpha=0.2)

plt.plot(sfreq, 10.*np.log10(Px / Px[0]), 'g-', alpha=0.2)

plt.plot(sfreq, 10.*np.log10(Pn / np.sqrt(Px[0]*Py[0])), 'r-', alpha=0.2)

plt.ylabel('dB')

plt.ylim(-40, 3)

plt.title('Isolated sources in coadd: W%i' % band)

ps.savefig()

xx = np.zeros((len(dbs), len(dbs[0][0])))

yy = np.zeros((len(dbs), len(dbs[0][0])))

nn = np.zeros((len(dbs), len(dbs[0][0])))

mm = np.zeros((len(dbs), len(dbs[0][0])))

for i,(Sx,Sy,Px,Py,sfreq,Pn) in enumerate(dbs):

xx[i,:] = 10.*np.log10(Px / Px[0])

yy[i,:] = 10.*np.log10(Py / Py[0])

nn[i,:] = 10.*np.log10(Pn / np.sqrt(Px[0]*Py[0]))

Sm = (Sx + Sy) / 2.

P = Sm.real**2 + Sm.imag**2

mm[i,:] = 10.*np.log10(P / P[0])

if pred:

plt.clf()

med = np.median(mm, axis=0)

plt.plot(sfreq, med, 'g-')

plt.plot(sfpsf, 10.*np.log10(Ppsf / Ppsf[0]), '-', color=(1.,0.6,0))

plt.plot(sflan, 10.*np.log10(Plan / Plan[0]), 'k-')

plt.plot(sfreq, 10.*np.log10(Ppred / Ppred[0]), 'r-')

plt.ylabel('dB')

plt.ylim(-40, 3)

plt.title('PSF')

ps.savefig()

plt.clf()

plt.plot(sfreq, np.median(xx, axis=0), 'g-')

plt.plot(sfreq, np.median(yy, axis=0), 'b-')

plt.plot(sfreq, np.median(nn, axis=0), 'r-')

plt.title('Isolated sources in coadd: W%i median' % band)

#plt.xlabel('frequency (cycles/pixel)')

plt.ylabel('dB')

plt.ylim(-40, 3)

ps.savefig()

print 'Freqs:', sfreq

if len(l1dbs) == 0:

continue

plt.clf()

for Px,Py,sf,noise in l1dbs:

plt.plot(sf, 10.*np.log10(Px / Px[0]), 'g-', alpha=0.2)

plt.plot(sf, 10.*np.log10(Py / Py[0]), 'b-', alpha=0.2)

plt.plot(sf, 10.*np.log10(noise / Py[0]), 'r-', alpha=0.2)

plt.plot(sf, 10.*np.log10(noise / Px[0]), 'r-', alpha=0.2)

plt.ylabel('dB')

plt.ylim(-40, 3)

plt.title('Isolated sources in L1bs: W%i' % band)

ps.savefig()

xx = np.zeros((len(l1dbs), len(l1dbs[0][0])))

yy = np.zeros((len(l1dbs), len(l1dbs[0][0])))

zz = np.zeros((len(l1dbs), len(l1dbs[0][0])))

zz2 = np.zeros((len(l1dbs), len(l1dbs[0][0])))

for i,(Px,Py,sfreq,noise) in enumerate(l1dbs):

xx[i,:] = 10.*np.log10(Px / Px[0])

yy[i,:] = 10.*np.log10(Py / Py[0])

zz[i,:] = 10.*np.log10(noise / Py[0])

zz2[i,:] = 10.*np.log10(noise / Px[0])

plt.clf()

plt.plot(sfreq, np.median(xx, axis=0), 'g-')

plt.plot(sfreq, np.median(yy, axis=0), 'b-')

plt.plot(sfreq, np.median(zz, axis=0), 'r-')

plt.plot(sfreq, np.median(zz2, axis=0), 'r-')

plt.title('Isolated sources in L1bs: W%i median' % band)

#plt.xlabel('frequency (cycles/pixel)')

plt.ylabel('dB')

plt.ylim(-40, 3)

colors = ['b', 'g', 'r', 'm']

dbs = []

slices = []

centrals = []

for band in bands:

if band in [1,2,3]:

# im = fitsio.read('wise-psf-w%i-507.5-507.5.fits' % band)

im = fitsio.read('wise-psf-w%i-507.5-507.5-bright.fits' % band)

else:

# im = fitsio.read('wise-psf-w%i-253.5-253.5.fits' % band)

im = fitsio.read('wise-psf-w%i-253.5-253.5-bright.fits' % band)

im /= im.max()

print 'Image', im.shape, im.dtype

H,W = im.shape

assert(H == W)

print 'center / max:', im[H/2, W/2]

assert(im[H/2, W/2] == 1.0)

c = W/2

# plt.clf()

# plt.imshow(np.log10(im), interpolation='nearest', origin='lower',

# vmin=-8, vmax=0)

# ps.savefig()

nclip = (W - 251) / 2

slc = slice(nclip, -nclip)

central = im[slc,slc]

centrals.append(central)

ch,cw = central.shape

cc = ch/2

slices.append((central[cc,:], central[:,cc]))

if fftcentral is not None:

nclip = (W - fftcentral) / 2

im = im[nclip:-nclip, nclip:-nclip]

H,W = im.shape

Y = im[H/2, :]

X = im[:, W/2]

Fy = np.fft.fft(Y)

Fx = np.fft.fft(X)

Sy = np.fft.fftshift(Fy)

Sx = np.fft.fftshift(Fx)

Px = Sx.real**2 + Sx.imag**2

Py = Sy.real**2 + Sy.imag**2

freqs1 = np.fft.fftfreq(len(Y))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

s0 = s0[0]

Px = Px[s0:]

Py = Py[s0:]

sf = sfreqs1[s0:]

Sx = Sx[s0:]

Sy = Sy[s0:]

dbs.append((Sx, Sy, Px, Py, sf))

# plt.clf()

# plt.plot(sf, 10.*np.log10(Px / Px[0]), 'g-')

# plt.plot(sf, 10.*np.log10(Py / Py[0]), 'b-')

# plt.ylabel('dB')

# plt.ylim(-40, 3)

# plt.title('PSF models: W%i' % band)

# ps.savefig()

if noplots:

return dbs

cmap = 'jet'

ps.basefn = 'wisepsf'

if bw:

ps.pattern = 'wisepsf-%s-bw.%s'

#cmap = 'gray'

cmap = antigray

colors = ['k']*4

plt.figure(figsize=(4,4))

if True:

vmin=-7

vmax=0

# bot = 0.05

# top = 0.02

# left = 0.1

# right = 0.02

bot = 0.12

top = 0.03

left = 0.12

right = 0.03

# colorbar cbar

plt.figure(figsize=(0.75,4))

plt.clf()

ax = plt.gca()

print 'ax pos', ax.get_position()

ax.set_position([0.01, bot, 0.35, 1.-bot-top])

#cax,kw = matplotlib.colorbar.make_axes(ax, fraction=0.9)

#fraction 0.15; fraction of original axes to use for colorbar

#pad 0.05 if vertical, 0.15 if horizontal; fraction of original axes between colorbar and new image axes

#shrink 1.0; fraction by which to shrink the colorbar

#aspect 20; ratio of long to short dimensions

cb = matplotlib.colorbar.ColorbarBase(ax, cmap=cmap,

norm=matplotlib.colors.Normalize(vmin=vmin, vmax=vmax))

tv = np.arange(vmin, vmax+1)

cb.set_ticks(tv, update_ticks=False)

tt = ['$10^{%i}$' % t for t in tv]

tt[-1] = '$1$'

cb.set_ticklabels(tt, update_ticks=False)

cb.update_ticks()

ps.savefig()

plt.figure(figsize=(4,4))

plt.subplots_adjust(left=left, right=1.-right, bottom=bot, top=1.-top)

for central in centrals:

ch,cw = central.shape

cc = ch/2

plt.clf()

plt.imshow(np.log10(np.maximum(10.**(vmin-1), central)),

interpolation='nearest', origin='lower',

vmin=vmin, vmax=vmax, cmap=cmap,

extent=[-cc,cc,-cc,cc])

ps.savefig()

if True:

left = 0.17

right = 0.03

bot = 0.12

top = 0.03

plt.subplots_adjust(left=left, right=1.-right, bottom=bot, top=1.-top)

for i,(sx,sy) in enumerate(slices):

band = bands[i]

plt.clf()

S = len(sx)

mid = S/2

#vmin,vmax = -6,0

#vmin,vmax = -3,0

x = np.arange(len(sx)) - mid

pa = plt.semilogy(x, sx, '-', color=colors[i])

pb = plt.plot(x, sy, '-', color=colors[i], lw=3, alpha=0.5)

xlo,xhi = -10, 10

xx = np.linspace(xlo, xhi, 500)

scale = np.pi / 2.

yy = (2. * scipy.special.j1(xx * scale) / (xx * scale))**2

yy[xx == 0] = 1.

pc = plt.plot(xx, yy, '--', color='k')

plt.figlegend((pb[0], pa[0], pc[0]),

('W%i(y)' % band, 'W%i(x)' % band, 'Airy*'),

loc='upper right', fontsize=fontsize)

plt.xlim(xlo, xhi)

plt.ylim(1e-3/1.1, 1.1)

plt.xlabel('Pixel')

plt.ylabel('PSF profile')

ps.savefig()

if True:

left = 0.17

right = 0.03

bot = 0.12

top = 0.03

plt.subplots_adjust(left=left, right=1.-right, bottom=bot, top=1.-top)

# How much to compress the Airy profile so that it is just

# well-sampled (with unit pixels)

scale = np.pi / 2.

# Its half-width at half-max

hwhm = (1./scale * 1.61633)

# Gaussian with same HWHM

sigma = 2. * hwhm / 2.35

print 'sigma', sigma

x = np.linspace(-10, 10, 1000)

dx = x[1]-x[0]

g = np.exp(-0.5 * x**2 / sigma**2)

airy = (2. * scipy.special.j1(x * scale) / (x * scale))**2

Fg = np.fft.fft(g)

Sg = np.fft.fftshift(Fg)

Pg = Sg.real**2 + Sg.imag**2

freqs1 = np.fft.fftfreq(len(g))

sfg = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfg == 0)

Fa = np.fft.fft(airy)

Sa = np.fft.fftshift(Fa)

Pa = Sa.real**2 + Sa.imag**2

Pg = Pg[s0:]

Pa = Pa[s0:]

sfg = sfg[s0:]

print 'dx', dx

sfg /= dx

for i,(nil,nil,Px,Py,sf) in enumerate(dbs):

band = bands[i]

plt.clf()

pa = plt.plot(sf, 10.*np.log10(Px / Px[0]), '-', color=colors[i])

pb = plt.plot(sf, 10.*np.log10(Py / Py[0]), '-', color=colors[i],

lw=3, alpha=0.5)

pc = plt.plot(sfg, 10.*np.log10(Pg / Pg[0]), 'k--')

pd = plt.plot(sfg, 10.*np.log10(Pa / Pa[0]), 'k:')

lp = (pb[0], pa[0], pc[0], pd[0])

lt = ('W%i (y)' % band, 'W%i (x)' % band,

'Gaussian*', 'Airy*')

if band == 1:

pitch = 0.001

lan3 = lanczos_test(noplots=True, pitch=pitch)

(Slan,Plan,sflan) = lan3[0]

sflan /= pitch

pe = plt.plot(sflan, 10.*np.log10(Plan / Plan[0]), 'k-.')

lp = (pe[0],) + lp

lt = ('Lanczos-3',) + lt

plt.ylabel('Power in Fourier component (dB)')

plt.ylim(-40, 3)

plt.xlabel('Frequency (cycles/pixel)')

plt.legend(lp, lt, loc='lower left', fontsize=fontsize)

plt.xlim(0, 0.5)

pows = []

#for subx in np.linspace(-0.5, 0.5, 11):

for subx in [0]:

lo,hi = srange

dx = np.arange(lo, hi, pitch).astype(np.float32) + 3 + subx

ix = np.round(dx).astype(np.int32)

ddx = (dx - ix).astype(np.float32)

outimg = np.zeros_like(dx)

inimg = np.zeros((1,7), np.float32)

inimg[0,3] = 1.

lanczos3_interpolate(ix, np.zeros_like(ix),

ddx, np.zeros_like(ddx),

[outimg], [inimg])

# plt.clf()

# plt.plot(outimg)

# ps.savefig()

F = np.fft.fft(outimg)

S = np.fft.fftshift(F)

P = S.real**2 + S.imag**2

freqs1 = np.fft.fftfreq(len(outimg))

sfreqs1 = np.fft.fftshift(freqs1)

s0 = np.flatnonzero(sfreqs1 == 0)

s0 = s0[0]

s1 = s0# + 1

pows.append((S[s1:], P[s1:], sfreqs1[s1:]))

if noplots:

return pows

plt.figure(figsize=(4,4))

left = 0.17

right = 0.03

bot = 0.12

top = 0.03

plt.subplots_adjust(left=left, right=1.-right, bottom=bot, top=1.-top)

plt.clf()

for nil,P,sfreq in pows:

plt.plot(sfreq / pitch, 10.*np.log10(P / P[0]), 'b-')

#plt.plot(sfreq / pitch, 10.*np.log10(P / P[0]), 'b-')

plt.ylabel('Power in Fourier component (dB)')

plt.ylim(-40, 3)

#plt.title('Lanczos3')

plt.xlim(0, 0.5)

ps.savefig()

# plt.clf()

# for S,P,sfreq in pows:

# pp = int(np.round(1./pitch))

# plt.plot(sfreq[::pp], 10.*np.log10(P[::pp] / P[0]), 'b-')

# print 'freqs', sfreq[::pp]

# plt.ylabel('Power in Fourier component (dB)')

# plt.ylim(-40, 3)

# plt.xlim(0, 0.5)

# ps.savefig()