def shear_fit(x, y, fig_ax, label):
coeff = tool_box.fit_1d(x, y, 2, "scipy")
# y = a1 + a2*x a3*x^2 = a2(x+a1/2/a2)^2 +...
# gh = - a1/2/a2, gh_sig = \sqrt(1/2/a2)
g_h = -coeff[1] / 2. / coeff[2]
g_sig = 0.70710678118 / numpy.sqrt(coeff[2])
if fig_ax:
fig_ax.scatter(x, y, alpha=0.7, s=5)
fig_ax.plot(x,
coeff[0] + coeff[1] * x + coeff[2] * x**2,
alpha=0.7,
label=label)
return g_h, g_sig, coeff
gal_c.drawImage(image=gal_img, scale=pixel_scale)
gal_img_arr = gal_img.array
gal_pow = fq.pow_spec(gal_img_arr)
mg1, mg2, mn = fq.shear_est(gal_pow, psf_pow, F=True)[:3]
GN[0] += mg1
GN[1] += mg2
GN[2] += mn
mgs[0, k] = GN[0] / GN[2]
mgs[1, k] = GN[1] / GN[2]
# if rank == 0:
# print(g1-mgs[0])
# print(g2-mgs[1])
mc1 = tool_box.fit_1d(g1, mgs[0], 1, method="scipy")
mc2 = tool_box.fit_1d(g2, mgs[1], 1, method="scipy")
mcs[0, i, j] = mc1[1] -1 # m1
mcs[1, i, j] = mc1[0] # c1
mcs[2, i, j] = mc2[1]-1 # m2
mcs[3, i, j] = mc2[0] # c2
if rank == 0:
if ellip > 0.8:
numpy.savez("./result/galsim/cache_%s_ellip_mix.npz"%psf_tag, mcs)
pic_nm = "./result/galsim/mc_%s_ellip_mix.png"%psf_tag
pic_nm_pdf = "./result/galsim/mc_%s_ellip_mix.pdf"%psf_tag
else:
numpy.savez("./result/galsim/cache_%s_ellip_%.2f.npz"%(psf_tag,ellip), mcs)
idx_1 = cfht_mag > 0 idx_2 = cfht_mag < 50 mag = cfht_mag[idx_1&idx_2] bin_num = 50 nums, bins = numpy.histogram(mag, bin_num) print(nums, nums.shape) print(bins, bins.shape) fit_x = bins[:bin_num] print(fit_x, fit_x.shape) idx_1 = fit_x <= 24 idx_2 = fit_x >= 18.5 fit_nums = numpy.log10(nums[idx_1&idx_2]) print(fit_nums.shape) paras = tool_box.fit_1d(fit_x[idx_1&idx_2],fit_nums, 1, "lsq") print(paras) fmt='%2.f%%' fig_x = 8 fig_y = fig_x*4/6 figs = (fig_x, fig_y) fonts = 20 xy_lb_size = 18 legend_size = fonts - 4 axis_linewidth = 1.2 plt_line_width = 2 cap_size = 5 xticks = mtick.FormatStrFormatter(fmt) fig = plt.figure(figsize=figs)
def find_shear_grid(G,
NU,
G_PDF_bin,
G_hist_bin,
NU_hist_bin,
chisq_gap=100,
dg=0.01,
fit_num=10,
ax=False):
data_num = G.shape[0]
bin_num = G_PDF_bin.shape[0] - 1
bin_num2 = int(bin_num * 0.5)
inverse = range(int(bin_num / 2 - 1), -1, -1)
num_g = G_hist_bin.shape[0] - 1
num_nu = NU_hist_bin.shape[0] - 1
grid_x, grid_y = numpy.zeros(
(num_nu, num_g), dtype=numpy.float64), numpy.zeros((num_nu, num_g),
dtype=numpy.float64)
hist_num2d = numpy.zeros((num_nu, num_g), dtype=numpy.intc)
hist2d_fast(G, NU, data_num, G_hist_bin, NU_hist_bin, num_g, num_nu,
hist_num2d, grid_x, grid_y)
iters = 0
change = 1
left, right = -0.1, 0.11
while change == 1:
change = 0
mc = (left + right) / 2.
mcl = left
mcr = right
fmc = get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, mc,
bin_num, bin_num2, inverse)
fmcl = get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, mcl,
bin_num, bin_num2, inverse)
fmcr = get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, mcr,
bin_num, bin_num2, inverse)
# print("%d. %.4f %.2f %.4f %.2f %.4f %.2f"%(iters, left, fmcl, mc,fmc,right,fmcr))
temp = fmc + chisq_gap
if fmcl > temp:
# left = (mc + mcl) / 2.
left = mcl + (mc - mcl) / 3
change = 1
if fmcr > temp:
# right = (mc + mcr) / 2.
right = mcr - (mcr - mc) / 3
change = 1
iters += 1
if right - left < dg:
break
if iters > 40:
break
ghs = numpy.linspace(left, right, fit_num)
xi2 = numpy.array([
get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, gh, bin_num,
bin_num2, inverse) for gh in ghs
])
# print(iters, ghs)
if ax:
ax.plot(ghs, xi2)
coeff = tool_box.fit_1d(ghs, xi2, 2, "scipy")
gh = -coeff[1] / 2. / coeff[2]
gh_sig = 0.70710678118 / numpy.sqrt(coeff[2])
return gh, gh_sig, grid_x, grid_y, hist_num2d
num = int(argv[1])
order = int(argv[2])
percent = float(argv[3])
x = numpy.linspace(-5, 5, num)
fx = 0
coeff = numpy.random.uniform(-2, 2, order + 1)
for i in range(order + 1):
fx += coeff[i] * (x**i)
plt.plot(x, fx, label='origin')
print("Input coeff: ", coeff)
for i in range(num):
fx[i] = fx[i] + numpy.random.normal(0, numpy.abs(percent * fx[i]), 1)[0]
coeff, cov, fxy = tool_box.fit_1d(x, fx, order, "lsq")
coeff_s = tool_box.fit_1d(x, fx, order, "scipy")
# print(cov)
# print(fxy)
print("Lsq: ", coeff)
print("Scipy: ", coeff_s)
pic_title = "coeff: "
for i in range(order + 1):
pic_title += "%.4f, " % coeff[i]
fx_fit = 0
for i in range(order + 1):
fx_fit += coeff[i] * (x**i)
plt.scatter(x, fx, label="noise")
plt.plot(x, fx_fit, label="fit")
plt.title(pic_title)
limited=40)
img.axs[0][0].plot(g1_hat,
g1_chisq,
marker="o",
c="C%d" % data_tag,
label=labels[data_tag])
img.axs[1][0].plot(g2_hat,
g2_chisq,
marker="o",
c="C%d" % data_tag,
label=labels[data_tag])
if data_tag != 2:
# estimate shear signal
coeff = tool_box.fit_1d(g1_hat, g1_chisq, 2, "scipy")
g1_estimate = -coeff[1] / 2. / coeff[2]
g1_estimate_sig = 0.70710678118 / numpy.sqrt(coeff[2])
# estimate shear signal
coeff = tool_box.fit_1d(g2_hat, g2_chisq, 2, "scipy")
g2_estimate = -coeff[1] / 2. / coeff[2]
g2_estimate_sig = 0.70710678118 / numpy.sqrt(coeff[2])
# plot chi squared
img.axs[0][1].plot(g1_hat,
g1_chisq,
marker="o",
c="C%d" % data_tag,
label=labels[data_tag])
ys = img.axs[0][1].set_ylim()
# 1 sigma range
print("From CPP")
print("%.5f (%.5f), %.5f (%.5f)" % (mc1[0] - 1, mc1[1], mc1[2], mc1[3]))
# print("%.5f (%.5f), %.5f (%.5f)"%(mc1_n[0],mc1_n[1],mc1_n[2],mc1_n[3]))
print("%.5f (%.5f), %.5f (%.5f)" % (mc2[0] - 1, mc2[1], mc2[2], mc2[3]))
# print("%.5f (%.5f), %.5f (%.5f)"%(mc2_n[0],mc2_n[1],mc2_n[2],mc2_n[3]))
check_result = numpy.zeros_like(mg_l)
new_result = numpy.zeros_like(mg_l)
for i in range(shear_num):
fit_x1 = chisq_g1[i, 20:]
fit_y1 = chisq_g1[i, :20]
fit_x2 = chisq_g2[i, 20:]
fit_y2 = chisq_g2[i, :20]
coeff = tool_box.fit_1d(fit_x1, fit_y1, 2, "scipy")
fit_curve_g1 = coeff[0] + coeff[1] * fit_x1 + coeff[2] * fit_x1**2
print(-coeff[1] / 2. / coeff[2])
fit_slope_g1 = fit_y1 - fit_curve_g1
coeff = tool_box.fit_1d(fit_x1, fit_y1 - fit_slope_g1, 2, "scipy")
print(-coeff[1] / 2. / coeff[2])
fit_curve_g1 = coeff[0] + coeff[1] * fit_x1 + coeff[2] * fit_x1**2
coeff_new, slope_coeff = new_fit(fit_x1, fit_y1)
print(coeff_new) #,slope_coeff)
print(coeff)
new_fit_curve_g1 = coeff_new[
0] + coeff_new[1] * fit_x1 + coeff_new[2] * fit_x1**2
check_result[0, i] = -coeff[1] / 2. / coeff[2]
check_result[1, i] = 0.70710678118 / numpy.sqrt(coeff[2])
for i in range(2):
# shear
chi_left = chi_data[i][:, :half_bin][:, inverse]
chi_right = chi_data[i][:, half_bin:]
dn = (chi_right - chi_left)**2
dm = chi_right + chi_left
chi = numpy.sum(dn / dm, axis=1)
idx = chi == numpy.nan
chi[idx] = -10
chisq[:, i] = chi
idx = chi >= 0
chi_min = chi[idx].min()
idx = chi <= chi_min + 40
coeff = tool_box.fit_1d(gh[idx], chisq[:, i][idx], 2, "scipy")
chisq_fit.append((chisq[:, i][idx], gh[idx], coeff))
g_h = -coeff[1] / 2. / coeff[2]
g_sig = 0.70710678118 / numpy.sqrt(coeff[2])
final_signal[rank, i * 2] = g_h
final_signal[rank, i * 2 + 1] = g_sig
# shear*critical_surface_density
chi_left = chi_crit_data[i][:, :half_bin][:, inverse]
chi_right = chi_crit_data[i][:, half_bin:]
dn = (chi_right - chi_left)**2
dm = chi_right + chi_left
chi = numpy.sum(dn / dm, axis=1)
idx = chi == numpy.nan
chi[idx] = -10
chisq_crit[:, i] = chi
img = Image_Plot(fig_x=14, fig_y=5, xpad=0.15, ypad=0.15)
img.subplots(1, 1)
for i in range(10, 7, -1):
# for i in range(chi_guess_num-1,-1,-1):
tag_1 = int(2 * i)
tag_2 = tag_1 + 1
npz = numpy.load(data_path + "/chisq_%d_%d_%s.npz" %
(tag_1, tag_2, data_type[j]))
chi_guess_bin = npz["arr_0"]
chisq_arr = npz["arr_1"][0]
idx = chisq_arr <= 150
ls = img.line_styles[i]
coeff = tool_box.fit_1d(chi_guess_bin[idx], chisq_arr[idx], 2, "scipy")
xi_err = numpy.sqrt(1 / 2. / coeff[2])
xi = -coeff[1] / 2. / coeff[2]
chi_measures[j][0, i] = xi
chi_measures[j][1, i] = xi_err
print(tag_1, tag_2, i, chi_guess[i], xi, xi_err)
img.axs[0][0].plot(chi_guess_bin[idx],
chisq_arr[idx],
linewidth=2.2,
label="$\\xi^t=%.2e, \\xi^m=%.3e(%.2e)$" %
(chi_guess[i], xi, xi_err),
ls=ls[0],
c=ls[1])
ys = img.axs[0][0].set_ylim()
img.axs[0][0].plot([chi_guess[i], chi_guess[i]], [ys[0], ys[1]],
def find_shear_grid(G,
NU,
G_PDF_bin,
G_hist_bin,
NU_hist_bin,
chisq_gap=100,
dg=0.01,
fit_num=10):
data_num = G.shape[0]
bin_num = G_PDF_bin.shape[0] - 1
bin_num2 = int(bin_num * 0.5)
inverse = range(int(bin_num / 2 - 1), -1, -1)
num_g = G_hist_bin.shape[0] - 1
num_nu = NU_hist_bin.shape[0] - 1
grid_x, grid_y = numpy.zeros(
(num_nu, num_g), dtype=numpy.float64), numpy.zeros((num_nu, num_g),
dtype=numpy.float64)
hist_num2d = numpy.zeros((num_nu, num_g), dtype=numpy.intc)
hist2d_fast(G, NU, data_num, G_hist_bin, NU_hist_bin, num_g, num_nu,
hist_num2d, grid_x, grid_y)
iters = 0
change = 1
left, right = -0.1, 0.099
while change == 1:
change = 0
mc = (left + right) / 2.
mcl = left
mcr = right
fmc = get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, mc,
bin_num, bin_num2, inverse)
fmcl = get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, mcl,
bin_num, bin_num2, inverse)
fmcr = get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, mcr,
bin_num, bin_num2, inverse)
temp = fmc + chisq_gap
if fmcl > temp:
left = (mc + mcl) / 2.
change = 1
if fmcr > temp:
right = (mc + mcr) / 2.
change = 1
# print(iters, left,right)
iters += 1
if right - left < dg:
break
if iters > 12:
break
ghs = numpy.linspace(left, right, fit_num)
xi2 = numpy.array([
get_chisq_grid(hist_num2d, grid_x, grid_y, G_PDF_bin, gh, bin_num,
bin_num2, inverse) for gh in ghs
])
# print(iters, ghs)
# img = Image_Plot()
# img.subplots(1, 1)
# img.axs[0][0].plot(ghs, xi2)
# img.show_img()
coeff = tool_box.fit_1d(ghs, xi2, 2, "scipy")
gh = -coeff[1] / 2. / coeff[2]
gh_sig = 0.70710678118 / numpy.sqrt(coeff[2])
return gh, gh_sig, grid_x, grid_y, hist_num2d