def weight_prime_from_data(corr,
bg1,
bg2,
delta=0.,
scale_factor=1.,
mode="corr",
pre_filter=True):
"""Computes weighting function for weighted normalization.
Parameters
----------
corr : ndarray
Correlation (or difference) data
scale_factor : ndarray
Scaling factor as returned by :func:`.core.scale_factor`. If not provided,
corr data must be computed with scale = True option.
mode : str
Representation mode of the data, either 'corr' (default) or 'diff'
pre_filter : bool
Whether to perform denoising and filtering. If set to False, user has
to perform data filtering.
Returns
-------
out : ndarray
Weight data for weighted sum calculation.
"""
scale_factor = np.asarray(scale_factor)
delta = np.asarray(delta)
bg1, bg2 = np.asarray(bg1), np.asarray(bg2)
if mode == "corr":
#make sure it is decreasing and clipped between 0 and 1
if pre_filter == True:
corr = _avg.denoise(corr)
corr = _avg.decreasing(corr)
corr = np.clip(corr, 0., scale_factor[..., None])
corr = _avg.denoise(corr)
g = np.divide(corr, scale_factor[..., None])
return weight_prime_from_g(g, delta[..., None], bg1[..., None],
bg2[..., None])
elif mode == "diff":
if pre_filter == True:
corr = _avg.denoise(corr)
corr = _avg.increasing(corr)
corr = np.clip(corr, 0., scale_factor[..., None] * 2)
corr = _avg.denoise(corr)
d = np.divide(corr, scale_factor[..., None])
return weight_prime_from_d(d, delta[..., None], bg1[..., None],
bg2[..., None])
else:
raise ValueError("Wrong mode.")
def _data_estimator(data, size=8, n=3, multilevel=False):
from cddm.multitau import log_average, merge_multilevel
from cddm.avg import denoise
if multilevel == False:
x, y = log_average(data, size)
else:
x, y = merge_multilevel(data)
#in case we have nans, remove them before denoising, and return only valid data
mask = np.isnan(y)
mask = np.logical_not(np.all(mask, axis=tuple(range(mask.ndim - 1))))
return x[mask], denoise(y[..., mask], n=n)
ax2.semilogx(x[1:],
std[1:],
marker=MARKERS.get(norm, "o"),
linestyle='',
fillstyle="none",
label="${}$".format(LABELS.get(norm)))
else:
#ax1.semilogx(x[1:],y[0,1:],linestyle = ':',fillstyle = "none")
ax2.semilogx(x[1:], std[1:], linestyle=':', fillstyle="none")
ax1.plot(x[1:], g1(x[1:], i, j), "k", label="$g$")
#: take first run, norm = 3 data for g estimation
x, g = log_average(data[0, 3, i, j, :])
g = denoise(g)
g = decreasing(g)
g = g.clip(0, 1)
#ax1.plot(x[1:],g[1:], "k:",label = "denoised")
x = np.arange(NFRAMES)
ax1.plot(x[1:], w[1:], "k--", label="$w$")
ax1.plot(x[1:], wp[1:], "k:", label="$w'$")
#ax2.set_ylim(ax1.get_ylim())
x, err2 = merge_multilevel(multilevel(err2, binning=0))
x, err3 = merge_multilevel(multilevel(err3, binning=0))
x, err6 = merge_multilevel(multilevel(err6, binning=0))
x, err0 = merge_multilevel(multilevel(err0, binning=0))
lin_5 = np.load(p.join(DATA_PATH,"cross_correlate_multi_raw_fast_norm_5.npy")) multi_5 = np.load(p.join(DATA_PATH,"cross_correlate_multi_raw_slow_norm_5.npy")) #: norm = 3 data normalized with scale = True lin_6 = np.load(p.join(DATA_PATH,"cross_correlate_multi_raw_fast_norm_6.npy")) multi_6 = np.load(p.join(DATA_PATH,"cross_correlate_multi_raw_slow_norm_6.npy")) #: take (i,j) k value (i,j) = (12,2) f, (ax1, ax2) = plt.subplots(2, 1, sharex=True) x,y = log_merge(lin_6, multi_6) #: denoised data, used as a estimator for the weight yd = denoise(decreasing(np.clip(denoise(y),0,1))) x,y = log_merge(lin_5, multi_5) ax2.semilogx(x,y[i,j], label = "norm 5") x,y = log_merge(lin_6, multi_6) ax2.semilogx(x,y[i,j], label = "norm 6") ax1.semilogx(x,yd[i,j], label = "g1") #: now calculate weights and do weighted sum. #: x values for the interpolator x_lin = np.arange(lin_6.shape[-1]) #: iweight for linear part. We have already filtered the data, so pre_filter = False w_lin = weight_from_data(yd, pre_filter = False)
