def b_ink(x, fnca, k):
"return t, bcoef"
ca = []
for fnc in fnca:
t = [None]*len(x)
for n in range(len(x)):
cf = np.empty_like(fnc)
s = list(fnc.shape)
s[n] = 1
xx = x[n]
xxe = xx[-1]
xxe += (xxe-xx[-2]) * 0.1
kk = k[n]
start = kk // 2
end = -(kk - start)
knots = xx[start:end].copy()
if kk % 2:
knots[:-1] += knots[1:]
knots[-1] += xx[end]
knots *= 0.5
for idx in np.ndindex(tuple(s)):
idx = list(idx)
idx[n] = slice(None)
#ss = InterpolatedUnivariateSpline(x[n], fnc[idx], bbox=(xx[0],xxe), k=k[n]-1)
ss = LSQUnivariateSpline(xx, fnc[idx], knots, bbox=(xx[0],xxe), k=kk-1)
cf[idx] = ss.get_coeffs()
t[n] = np.pad(knots, kk, 'constant', constant_values=(xx[0],xxe))
fnc = cf
ca.append(fnc)
return t, ca
def b_ink(x, fnca, k):
"return t, bcoef"
ca = []
for fnc in fnca:
t = [None] * len(x)
for n in range(len(x)):
cf = np.empty_like(fnc)
s = list(fnc.shape)
s[n] = 1
xx = x[n]
xxe = xx[-1]
xxe += (xxe - xx[-2]) * 0.1
kk = k[n]
start = kk // 2
end = -(kk - start)
knots = xx[start:end].copy()
if kk % 2:
knots[:-1] += knots[1:]
knots[-1] += xx[end]
knots *= 0.5
for idx in np.ndindex(tuple(s)):
idx = list(idx)
idx[n] = slice(None)
#ss = InterpolatedUnivariateSpline(x[n], fnc[idx], bbox=(xx[0],xxe), k=k[n]-1)
ss = LSQUnivariateSpline(xx,
fnc[idx],
knots,
bbox=(xx[0], xxe),
k=kk - 1)
cf[idx] = ss.get_coeffs()
t[n] = np.pad(knots, kk, 'constant', constant_values=(xx[0], xxe))
fnc = cf
ca.append(fnc)
return t, ca
def resid(self, params, **kwargs):
"""Residuals
Return the residuals
Args:
params (lmfit.Parameters): see params in self.model
Returns:
array: model minus data
"""
flux = self.spec.flux.copy()
wav = self.spec.wav
model = self.model(params, wav=wav, **kwargs)
if self.cont_method == 'spline-fm':
model /= self.spline(params, wav)
if self.cont_method == 'spline-dd':
node_wav, node_flux = get_spline_nodes(params)
t = node_wav[1:-1]
spl = LSQUnivariateSpline(wav, flux, t, k=3, ext=0)
flux /= spl(wav)
spl = LSQUnivariateSpline(wav, model, t, k=3, ext=0)
model /= spl(wav)
res = flux - model
return res
def fit(self):
self.knotPvals = np.zeros(len(self.knots))
self.interp = LSQUnivariateSpline(self.x,
self.y,
self.knots,
k=1,
w=self.weights)
def Smooth(x,y):
"""return: smoothed funciton s(x) and estimation of sigma of y for one data point"""
#define segments to do spline smoothing
t=np.linspace(0,len(x),3)[1:-1]
sr = LSQUnivariateSpline(x, y.real, t)
si = LSQUnivariateSpline(x, y.imag, t)
return sr(x)+si(x)*1j, (sr.get_residual()/len(x))**0.5+(si.get_residual()/len(x))**0.5*1j
def spline_coeff(lat, time):
time = time[~np.isnan(lat)]
time_diff = time[1:]-time[:-1]
counter = 0
indeces_remotion = np.array([])
for i in range(0, time_diff.shape[0]):
if time_diff[i] > 60:
if counter < 5:
slack = np.arange(i - counter, i + 1)
indeces_remotion = np.append(indeces_remotion, slack)
counter = 0
else:
counter += 1
time = np.delete(time, indeces_remotion)
lat = lat[~np.isnan(lat)]
lat = np.delete(lat, indeces_remotion)
if lat.shape[0] > 5:
save_data = optimal_knots(lat, time)
t2 = time[save_data[1:-1]]
try2 = LSQUnivariateSpline(time, lat, t2, k=3)
spl_coeff = try2.get_coeffs()
knots = np.r_[(time[save_data[0]],) * (3 + 1), t2, (time[save_data[-1]-1],) * (3 + 1)]
return [spl_coeff, knots] # in this case returns spline coeff and knots
else:
return ["not enough data", lat] # in this case returns original data
def plot_contour_line(ax, x, y, **kwargs):
"""Plot smooth curve from points.
There is some noise in the contour points from MINUIT,
which throws off most interpolation schemes.
The LSQUnivariateSpline as used here gives good results.
It could probably be simplified, or Bezier curve plotting via
matplotlib could also be tried:
https://matplotlib.org/gallery/api/quad_bezier.html
"""
from scipy.interpolate import LSQUnivariateSpline
x = np.hstack([x, x, x])
y = np.hstack([y, y, y])
t = np.linspace(-1, 2, len(x), endpoint=False)
tp = np.linspace(0, 1, 50)
t_knots = np.linspace(t[1], t[-2], 10)
xs = LSQUnivariateSpline(t, x, t_knots, k=5)(tp)
ys = LSQUnivariateSpline(t, y, t_knots, k=5)(tp)
ax.plot(xs, ys, **kwargs)
def calc_coeffs(self):
coeff = []
for v, (g, kdata) in enumerate(zip(self.basegrid, self.knot_data)):
ranges = [(slice(r[0], r[1], t[0] * 1j) if kd.used() else slice(
(r[1] - r[0]) / 2.0, r[1], 1j))
for r, t, kd in zip(self.ranges, g[0], kdata)]
fnc = self.calc_grid(ranges, g[1], g[2])[v]
for n in range(len(ranges)):
kd = self.knot_data[v, n]
if not kd.used():
continue
s = list(fnc.shape)
s[n] = kd.num_coeffs()
out = np.empty(s)
s[n] = 1
x = np.mgrid[ranges[n]]
for i in np.ndindex(tuple(s)):
i = list(i)
i[n] = slice(None)
ss = LSQUnivariateSpline(x,
fnc[i],
kd.get_knots(),
bbox=kd.get_bbox(),
k=kd.get_order() - 1)
out[i] = ss.get_coeffs()
fnc = out
coeff.append(fnc)
return coeff
def __init__(self, file, knots=[]):
self.filename = file
with open(file, "r") as f:
inData = genfromtxt(f, float, comments='#').T
self.rho = inData[0]
self.orig_data = inData[1]
self.spline = LSQUnivariateSpline(self.rho, self.orig_data, knots, k=5)
def fft_normalize(freq, power, n, bin_sz, cut_peak, knot1, knot2):
""" Takes the values of fft at given frequencies and normalizes the power spectrum using a second order spline
INPUTS:
freq: frequencies where the power of the FFT spectrum has been evaluated
power: power of the FFT spectrum evaluated at freq
n: Number of points in freq
bin_sz: Distance between consecutive frequencies
cut_peak: Maximum relative amplitude for a given peak to be used in second and final spline fit
knot1: Knot distance for the preliminary second order spline with the peaks on it
knot2: Knot distance for the final second order spline where the peaks have been removed
OUTPUT:
power_rel: The FFT spectrum normalized using a second order spline fit and normalized by its median.
"""
knot_w = knot1 * n * bin_sz # difference in freqs (cycles/day) between each knot
first_knot_i = np.round(
knot_w / bin_sz
) #index of the first knot is the first point that is knot_w away from the first value of x
last_knot_i = len(
freq
) - first_knot_i #index of the last knot is the first point that is knot_w away from the last value of x
knots = np.arange(freq[first_knot_i], freq[last_knot_i], knot_w)
spline = LSQUnivariateSpline(
freq, power, knots,
k=2) #the spline, it returns the piecewise function
fit = spline(freq) #the actual y values of the fit
if np.amin(fit) < 0:
fit = np.ones(len(freq))
pre_power_rel = power / fit
# second fit -- by deleting the points higher than cut_peak times the average value of power
pre_indexes = constraint_index_finder(cut_peak, power)
power_fit = np.delete(pre_power_rel, pre_indexes)
freq_fit = np.delete(freq, pre_indexes)
knot_w1 = knot2 * n * bin_sz
first_knot_fit_i = np.round(
knot_w1 / bin_sz
) #index of the first knot is the first point that is knot_w away from the first value of x
last_knot_fit_i = len(
freq_fit
) - first_knot_fit_i #index of the last knot is the first point that is knot_w away from the last value of x
knots_fit = np.arange(freq_fit[first_knot_fit_i],
freq_fit[last_knot_fit_i], knot_w1)
spline = LSQUnivariateSpline(
freq_fit, power_fit, knots_fit,
k=2) #the spline, it returns the piecewise function
fit3 = spline(freq) #the actual y values of the fit applied to freq
if np.amin(fit3) < 0:
fit3 = np.ones(len(freq))
# relative power
power_rel = pre_power_rel / fit3
return power_rel / np.median(power_rel) # so the median of power_rel is 1
def fun_residuals(params, xnor, ynor, w, bbox, k, ext):
"""Compute fit residuals"""
spl = LSQUnivariateSpline(x=xnor,
y=ynor,
t=[item.value for item in params.values()],
w=w,
bbox=bbox,
k=k,
ext=ext,
check_finite=False)
return spl.get_residual()
def make_smoothed_err_plot(y, y_hat, title, fname, order=3, num_knots=10):
sort_idxs = np.argsort(y.flatten())
y_sorted = y.flatten()[sort_idxs]
y_hat_sorted = y_hat.flatten()[sort_idxs]
sq_errs = (y_sorted - y_hat_sorted)**2
cum_sq_errs = np.cumsum(sq_errs)
winrad = 100
avg_errs = np.array([
sq_errs[i - winrad:i + winrad].sum() / (2 * winrad)
for i in range(winrad, y.size - winrad)
])
log_avg_errs = np.log10(avg_errs)
fig, ax = plt.subplots()
x = np.arange(avg_errs.size)
ax.plot(x, log_avg_errs, color="red")
knots = np.linspace(x.min() + 1, x.max() - 1, num_knots)
spline = LSQUnivariateSpline(x, log_avg_errs, k=order, t=knots)
log_avg_errs_fit = spline(x)
ax.plot(x, log_avg_errs_fit, "b-")
plt.title(title)
plt.xlabel("Points ordered by true outputs")
plt.ylabel("Log_10 of squared error")
plt.savefig(fname)
plt.close()
def spline_derivatives(landscape, n_bins=100):
"""
:param landscape: original landscape
:param n_bins: number of bins for fitting
:return: tuple of (spline fit to A_z, split fit to A_dot, fit to A_ddot
"""
z = landscape.z
A_z = landscape.A_z
sort_idx = np.argsort(z)
z_sort = z[sort_idx]
A_sort = A_z[sort_idx]
knots = np.linspace(min(z_sort), max(z_sort), endpoint=True, num=n_bins)
spline_A_z = LSQUnivariateSpline(x=z_sort, y=A_sort, k=3, t=knots[1:-1])
A_dot_spline_z = spline_A_z.derivative(1)(z)
A_ddot_spline_z = spline_A_z.derivative(2)(z)
return spline_A_z, A_dot_spline_z, A_ddot_spline_z
def fit_spline(lc, niter=0, reject_thresh=5.0):
'''Fit a smooth spline to the lightcurve.
The number of knots is empirically chosen based on examining a few
lightcurves of relatively bright quasars.
The fit can include iterative rejection, but by default it doesn't.
The chi^2 statistic is returned using the spline model for the fluxes.
'''
if lc['mjd'][0] < 52000:
knots = np.array([52500., 53500., 54000.])
else:
knots = np.array([52700., 53500., 54000.])
iternum = 0
ii = np.where(~lc['mask'])[0]
while True:
knots = knots[(knots > lc['mjd'][ii[0]]) & (knots < lc['mjd'][ii[-1]])]
spfit = LSQUnivariateSpline(lc['mjd'][ii],
lc['flux'][ii],
knots,
np.sqrt(lc['ivar'][ii]),
k=2)
chi2v = (lc['flux'][ii] - spfit(lc['mjd'][ii])) * np.sqrt(
lc['ivar'][ii])
if iternum < niter:
ii = ii[np.abs(chi2v) < reject_thresh]
iternum += 1
else:
break
chi2 = np.sum(chi2v**2)
ndof = len(ii)
return dict(spfit=spfit,
fitvals=spfit(lc['mjd']),
chi2=chi2,
ndof=ndof,
good=ii)
def _fit_method(self, model, x, y, **kwargs):
t = kwargs.pop('t', None)
weights = kwargs.pop('weights', None)
bbox = kwargs.pop('bbox', [None, None])
if t is not None:
if model.user_knots:
warnings.warn(
"The current user specified knots will be "
"overwritten for by knots passed into this function",
AstropyUserWarning)
else:
if model.user_knots:
t = model.t_interior
else:
raise RuntimeError("No knots have been provided")
if bbox != [None, None]:
model.bounding_box = bbox
from scipy.interpolate import LSQUnivariateSpline
spline = LSQUnivariateSpline(x,
y,
t,
w=weights,
bbox=bbox,
k=model.degree)
model.tck = spline._eval_args
return spline
def _spline_filter(x,y,bins=None,num_bins=100,k=3,**kw):
"""
:param x: to filter, independent variable
:param y: to filter
:param bins: x values we want to bin at
:param num_bins: number of bins to use (only used if bins is None)
:param k: degree of spline
:param kw: see LSQUnivariateSpline
:return: spline object
"""
min_x, max_x = min(x), max(x)
if (bins is None):
# fit a spline at the given bins
bins = np.linspace(min_x,max_x,endpoint=True,num=num_bins)
# determine where the bins are in the range of the data for this landscape
good_idx = np.where((bins >= min_x) & (bins <= max_x))
bins_relevant = bins[good_idx]
"""
exclude the first and last bins, to make sure the Schoenberg-Whitney
condition is met for all interior knots (see:
docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.LSQUnivariateSpline
"""
t = bins_relevant[1:-1]
kw = dict(x=x, t=t, k=k, **kw)
return LSQUnivariateSpline(y=y, **kw)
def sfit(ptx_eval, ptx, pty, s=1., k=3, t=None):
""" Fit a bunch of points and return an evaluation vector """
assert len(ptx) == len(pty)
if t is None:
return UnivariateSpline(x=ptx, y=pty, s=s, k=k)(ptx_eval)
else:
return LSQUnivariateSpline(x=ptx, y=pty, t=t, k=k)(ptx_eval)
def iter_spline(time, flux, mask, window_length):
masked_flux = flux[mask==1]
masked_time = time[mask==1]
no_knots = (max(time) - min(time)) / window_length
newflux = masked_flux.copy()
newtime = masked_time.copy()
newtime, newflux = cleaned_array(newtime, newflux)
detrended_flux = masked_flux.copy()
for i in range(constants.PSPLINES_MAXITER):
outliers = 1 - detrended_flux < constants.PSPLINES_STDEV_CUT * np.nanstd(detrended_flux)
mask_outliers = np.ma.where(outliers)
newtime, newflux = cleaned_array(newtime[mask_outliers], newflux[mask_outliers])
# knots must not be at the edges, so we take them as [1:-1]
if len(newtime) < 5:
return np.full(len(time), np.nan)
knots = np.linspace(min(newtime), max(newtime), no_knots)[1:-1]
try:
trend = LSQUnivariateSpline(newtime, newflux, knots)
except:
return np.full(len(time), np.nan)
trend_segment = trend(newtime)
detrended_flux = newflux / trend_segment
outliers = 1 - detrended_flux > constants.PSPLINES_STDEV_CUT * np.nanstd(detrended_flux)
mask_outliers = np.ma.where(mask_outliers)
if len(mask_outliers[0]) == 0:
break
return trend(time) # Final iteration: apply trend to clipped and masked values
def spline_derivatives(landscape,n_bins=100):
"""
:param landscape: original landscape
:param n_bins: number of bins for fitting
:return: tuple of (spline fit to A_z, split fit to A_dot, fit to A_ddot
"""
z = landscape.z
A_z = landscape.A_z
sort_idx = np.argsort(z)
z_sort = z[sort_idx]
A_sort = A_z[sort_idx]
knots = np.linspace(min(z_sort),max(z_sort),endpoint=True,num=n_bins)
spline_A_z = LSQUnivariateSpline(x=z_sort,y=A_sort,k=3,t=knots[1:-1])
A_dot_spline_z = spline_A_z.derivative(1)(z)
A_ddot_spline_z = spline_A_z.derivative(2)(z)
return spline_A_z,A_dot_spline_z,A_ddot_spline_z
def get_order_trace_old(s):
x = np.arange(len(s))
s = np.array(s)
mm = s > max(s) * 0.05
dd1, dd2 = np.nonzero(mm)[0][[0, -1]]
# mask out absorption feature
smooth_size = 20
s0 = ni.median_filter(s, 40)
s_s0 = s - s0
s_s0_std = s_s0[np.abs(s_s0) < 2. * s_s0.std()].std()
mmm = s_s0 > -3. * s_s0_std
s1 = ni.gaussian_filter1d(s0[dd1:dd2], smooth_size, order=1)
#x1 = x[dd1:dd2]
s1r = s1
s1_std = s1r.std()
s1_std = s1r[np.abs(s1r) < 2. * s1_std].std()
s1r[np.abs(s1r) < 2. * s1_std] = np.nan
if np.any(np.isfinite(s1r[:1024])):
i1 = np.nanargmax(s1r[:1024])
i1r = np.where(~np.isfinite(s1r[:1024][i1:]))[0][0]
i1 = dd1 + i1 + i1r #+smooth_size
else:
i1 = dd1
if np.any(np.isfinite(s1r[1024:])):
i2 = np.nanargmin(s1r[1024:])
i2r = np.where(~np.isfinite(s1r[1024:][:i2]))[0][-1]
i2 = dd1 + 1024 + i2r
else:
i2 = dd2
if 0:
p_init = models.Chebyshev1D(degree=6, window=[0, 2047])
fit_p = fitting.LinearLSQFitter()
p = fit_p(p_init, x[i1:i2][mmm[i1:i2]], s[i1:i2][mmm[i1:i2]])
if 1:
# t= np.linspace(x[i1]+10, x[i2-1]-10, 10)
# p = LSQUnivariateSpline(x[i1:i2],
# s[i1:i2],
# t, bbox=[0, 2047])
# t= np.concatenate([[x[1],x[i1-5],x[i1],x[i1+5]],
# np.linspace(x[i1]+10, x[i2-1]-10, 10),
# [x[i2-5], x[i2], x[i2+5],x[-2]]])
t = np.concatenate([[x[1], x[i1]],
np.linspace(x[i1] + 10, x[i2 - 1] - 10, 10),
[x[i2], x[-2]]])
p = LSQUnivariateSpline(x, s0, t, bbox=[0, 2047])
return p
def __init__(self,t,x,knot_spacing=1/30,fit_type='LSQ',**kwargs):
"""
Simple extension of UnivariateSpline to handle multidimensional data.
Params:
-------
t (ndarray)
x (ndarray)
If 2d, then (n_samples,n_dim).
**kwargs
"""
assert x.ndim==2
self.splines = []
self.knot_spacing = knot_spacing
if fit_type=='LSQ':
for i in xrange(x.shape[1]):
self.splines.append(LSQUnivariateSpline(t,x[:,i],
np.linspace(t[1],t[-2],int(np.floor(t[-2]-t[1])/knot_spacing)),
**kwargs))
elif fit_type=='Uni':
for i in xrange(x.shape[1]):
self.splines.append(UnivariateSpline(t,x[:,i],**kwargs))
else:
raise NotImplementedError
def spline_through_data(x,
y,
k=2,
grace_intv=1000.,
smoothing_factor=0.001,
downsample=10,
fixed_control_points=None):
"""Pass a spline through the data
Examples
--------
>>> x = np.arange(1000)
>>> y = np.random.normal(x * 0.1, 0.01)
>>> fun = spline_through_data(x, y, grace_intv=10.)
>>> np.std(y - fun(x)) < 0.01
True
"""
lo_lim, hi_lim = x[0], x[-1]
control_points = \
np.linspace(lo_lim + 2 * grace_intv, hi_lim - 2 * grace_intv,
x.size // downsample)
if fixed_control_points is not None and len(fixed_control_points) > 0:
print(f'Adding control points: {fixed_control_points}')
control_points = np.sort(
np.concatenate((control_points, fixed_control_points)))
detrend_fun = LSQUnivariateSpline(x,
y,
t=control_points,
k=k,
bbox=[lo_lim, hi_lim])
return detrend_fun
def detrend_bin_lightcurve(self,value=3*720+1):
"""
pad the array by 720 bins in each direction
I might need to have multiple layer of detrending
"""
time_bin = np.arange(self.data_pts+2*value)-value
signal_bin = np.concatenate([np.ones(value)*self.signal_bin[0],
self.signal_bin,
np.ones(value)*self.signal_bin[-1]])
tarray = (np.arange(int(len(time_bin)/value))+1)*value+time_bin[0]
tarray = tarray[:-1] # the last know is usually too small and freaks out
exclude = 90
lower_threshold = np.percentile(signal_bin,50-exclude/2.)
upper_threshold = np.percentile(signal_bin,50+exclude/2.)
truncated = signal_bin.copy()
truncated[(signal_bin > upper_threshold)] = upper_threshold
truncated[(signal_bin < lower_threshold)] = lower_threshold
sp = LSQUnivariateSpline(time_bin, truncated,t=tarray)
#sp.set_smoothing_factor(10)
self.lc_filter = sp(time_bin)[value:-value]
self.signal_bin_detrended = self.signal_bin-self.lc_filter+self.Normalization
self.signal_bin_cleaned = self.signal_cleaned-self.lc_filter+self.Normalization
def robust_poly(x,y,polyord,sigreject=3.0,iteration=3,useSpline=False,knots=None):
finitep = np.isfinite(y) & np.isfinite(x)
goodp = finitep ## Start with the finite points
for iter in range(iteration):
if np.sum(goodp) < polyord:
warntext = "Less than "+str(polyord)+"points accepted, returning flat line"
warnings.warn(warntext)
coeff = np.zeros(polyord)
coeff[0] = 1.0
else:
if useSpline == True:
if knots is None:
spl = UnivariateSpline(x[goodp], y[goodp], k=polyord, s=sSpline)
else:
spl = LSQUnivariateSpline(x[goodp], y[goodp], knots, k=polyord)
ymod = spl(x)
else:
coeff = np.polyfit(x[goodp],y[goodp],polyord)
yPoly = np.poly1d(coeff)
ymod = yPoly(x)
resid = np.abs(ymod - y)
madev = np.nanmedian(np.abs(resid - np.nanmedian(resid)))
goodp = (np.abs(resid) < (sigreject * madev))
if useSpline == True:
return spl
else:
return coeff
def IRLSSpline(time, flux, error, Q=400.0, ksep=0.07, numpass=5, order=3):
'''
IRLS = Iterative Re-weight Least Squares
Parameters
----------
time
flux
error
Q
ksep
numpass
order
Returns
-------
'''
weight = 1. / (error**2.0)
knots = np.arange(min(time) + ksep, max(time) - ksep, ksep)
for k in range(numpass):
spl = LSQUnivariateSpline(time, flux, knots, w=weight, k=order)
# spl = UnivariateSpline(time, flux, w=weight, k=order, s=1)
chisq = ((flux - spl(time))**2.) / (error**2.0)
weight = Q / ((error**2.0) * (chisq + Q))
return spl(time)
def fit(x, y, errors=None):
# Sort
ind = np.argsort(x)
x = x[ind]
y = y[ind]
# Fill in errors
if errors is None:
errors = errorEstimate(x, y)
# print(np.vstack((x,y,errors)).T)
# We learn a global multiplicative factor on top of the errors
# using cross-validation. This lets us set the cutoff to the most
# computationally efficient value and removes a degree of freedom.
cutoff = np.log(0.5)
alphas = 10**np.linspace(-3., 3., num=21, endpoint=True)
best = [1e100, alphas[len(alphas) // 2]]
for a in alphas:
# sub-sample for cross-validation
inds = np.random.choice(len(x), size=len(x) // 5, replace=False)
xS = x[inds]
yS = y[inds]
errorsA = a * errors[inds]
inds = np.argsort(xS)
xS = xS[inds]
yS = yS[inds]
errorsA = errorsA[inds]
knots, knotLocs = rawFit(xS, yS, errorsA)
interp = LSQUnivariateSpline(xS, yS, knotLocs, k=1, w=1 / errorsA**2)
resid = np.sum((interp(x) - y)**2)
# print(best,a,resid)
if resid < best[0]:
best[0] = resid
best[1] = a
knots, knotLocs = rawFit(x, y, errors * best[1])
interp = LSQUnivariateSpline(x, y, knotLocs, k=1, w=1 / errors**2)
print('Final Answer:', len(knots))
return interp(x), (x, y, knots, knotLocs, best), range(len(x)), interp
def detrend_bin_lightcurve(self,value=3*48+1):
#front,back = self.sector_seperator
front = int((self.time_bin[-1]-self.time_bin[0])/2)
back = front
print(front)
# detrend first orbit
orbit1_time_bin = np.arange(front+2*value)-value
orbit1_signal_bin = self.signal_bin[:front]
signal_bin = np.concatenate([np.ones(value)*orbit1_signal_bin[0],
orbit1_signal_bin,
np.ones(value)*orbit1_signal_bin[-1]])
tarray = (np.arange(int(len(orbit1_time_bin)/value))+1)*value+orbit1_time_bin[0]
tarray = tarray[:-1] # the last knot is usually too small and freaks out so remove
lower_threshold = np.percentile(signal_bin,5)
upper_threshold = np.percentile(signal_bin,95)
truncated = signal_bin.copy()
truncated[(signal_bin > upper_threshold)] = upper_threshold
truncated[(signal_bin < lower_threshold)] = lower_threshold
sp = LSQUnivariateSpline(orbit1_time_bin, truncated,t=tarray)
lc_filter1 = sp(orbit1_time_bin)[value:-value]
# detrend second orbit
orbit2_time_bin = np.arange((self.data_pts-back)+2*value)-value
orbit2_signal_bin = self.signal_bin[back:]
signal_bin = np.concatenate([np.ones(value)*orbit2_signal_bin[0],
orbit2_signal_bin,
np.ones(value)*orbit2_signal_bin[-1]])
tarray = (np.arange(int(len(orbit2_time_bin)/value))+1)*value+orbit2_time_bin[0]
tarray = tarray[:-1] # the last knot is usually too small and freaks out so remove
lower_threshold = np.percentile(signal_bin,5)
upper_threshold = np.percentile(signal_bin,95)
truncated = signal_bin.copy()
truncated[(signal_bin > upper_threshold)] = upper_threshold
truncated[(signal_bin < lower_threshold)] = lower_threshold
sp = LSQUnivariateSpline(orbit2_time_bin, truncated,t=tarray)
lc_filter2 = sp(orbit2_time_bin)[value:-value]
self.lc_filter = np.concatenate([lc_filter1,np.zeros(back-front),lc_filter2])
self.signal_bin_detrended = self.signal_bin-self.lc_filter+self.Normalization
self.signal_bin_cleaned = self.signal_cleaned-self.lc_filter+self.Normalization
class dataSmoother:
def __init__(self, file, knots=[]):
self.filename = file
with open(file, "r") as f:
inData = genfromtxt(f, float, comments='#').T
self.rho = inData[0]
self.orig_data = inData[1]
self.spline = LSQUnivariateSpline(self.rho, self.orig_data, knots, k=5)
def plot_orig(self):
fig = self._plot_base()
fig.scatter(self.rho, self.orig_data, color='red')
plt.show()
return fig
def plot(self):
fig = self._plot_base()
fig.scatter(self.rho, self.orig_data, color='red')
fig.plot(self.rho, self.spline(self.rho), color='black')
fig.scatter(self.rho, self.spline(self.rho), color='green', marker="x")
fig.legend(["Spline", "Original", "Cleaned"])
plt.show()
return fig
def save(self):
answer = ""
while answer not in ["y", "n"]:
answer = input("OK to continue [Y/N]? ").lower()
print("Backing up old file")
os.rename(self.filename, self.filename + '.bak')
contents = np.array([self.rho, self.spline(self.rho)]).T
savetxt(self.filename + "_clean", contents, delimiter=' ')
def get_spline(self):
return self.spline
def set_knots(self, knots):
self.spline = LSQUnivariateSpline(self.rho, self.orig_data, knots, k=5)
print("knots set")
def _plot_base(self):
plot = plt.figure()
fig = plot.add_subplot(111)
fig.set_xlabel(r'$\rho', fontsize=30)
fig.set_ylabel(r'Data', fontsize=30)
plt.xticks(fontsize=20)
plt.yticks(fontsize=20)
return fig
def plot_derivative(self):
fig = self._plot_base()
fig.scatter(self.rho, self.spline.derivative()(self.rho), color="blue")
return fig
def spline_fit(q, G0, k=3, knots=None,num=None):
if num is None:
num = min(150,q.size//(k)-1)
if (knots is None):
step = q.size//num
assert step > 0 and step
knots = q[1:-1:step]
spline_G0 = LSQUnivariateSpline(x=q, y=G0,t=knots, k=k)
return spline_G0
def IRLSSpline(time,
flux,
error,
Q=400.0,
ksep=0.07,
numpass=5,
order=3,
debug=False):
'''
IRLS = Iterative Re-weight Least Squares
Parameters
----------
time
flux
error
Q
ksep
numpass
order
Returns
-------
'''
weight = 1. / (error**2.0)
knots = np.arange(np.nanmin(time) + ksep, np.nanmax(time) - ksep, ksep)
if debug is True:
print('IRLSSpline: knots: ', np.shape(knots))
print('IRLSSpline: time: ', np.shape(time), np.nanmin(time), time[0],
np.nanmax(time), time[-1])
print('IRLSSpline: <weight> = ', np.mean(weight))
print(np.where((time[1:] - time[:-1] < 0))[0])
print(flux)
# plt.figure()
# plt.errorbar(time, flux, error)
# plt.scatter(knots, knots*0. + np.median(flux))
# plt.show()
for k in range(numpass):
spl = LSQUnivariateSpline(time,
flux,
knots,
k=order,
check_finite=True,
w=weight)
# spl = UnivariateSpline(time, flux, w=weight, k=order, s=1)
chisq = ((flux - spl(time))**2.) / (error**2.0)
weight = Q / ((error**2.0) * (chisq + Q))
return spl(time)
def update_normal_vector(self):
"""
update the constraint by approximating the
loglikelihood hypersurface as a spline in
each dimension.
This is an approximation which
improves as the algorithm proceeds
"""
n = self.ensemble[0].dimension
tracers_array = np.zeros((len(self.ensemble), n))
for i, samp in enumerate(self.ensemble):
tracers_array[i, :] = samp.values
V_vals = np.atleast_1d([p.logL for p in self.ensemble])
self.normal = []
for i, x in enumerate(tracers_array.T):
# sort the values
# self.normal.append(lambda x: -x)
idx = x.argsort()
xs = x[idx]
Vs = V_vals[idx]
# remove potential duplicate entries
xs, ids = np.unique(xs, return_index=True)
Vs = Vs[ids]
# pick only finite values
idx = np.isfinite(Vs)
Vs = Vs[idx]
xs = xs[idx]
# filter to within the 90% range of the Pvals
Vl, Vh = np.percentile(Vs, [5, 95])
(idx, ) = np.where(np.logical_and(Vs > Vl, Vs < Vh))
Vs = Vs[idx]
xs = xs[idx]
# Pick knots for this parameters: Choose 5 knots between
# the 1st and 99th percentiles (heuristic tuning WDP)
knots = np.percentile(xs, np.linspace(1, 99, 5))
# Guesstimate the length scale for numerical derivatives
dimwidth = knots[-1] - knots[0]
delta = 0.1 * dimwidth / len(idx)
# Apply a Savtzky-Golay filter to the likelihoods (low-pass filter)
window_length = len(
idx) // 2 + 1 # Window for Savtzky-Golay filter
if window_length % 2 == 0: window_length += 1
f = savgol_filter(
Vs,
window_length,
5, # Order of polynominal filter
deriv=1, # Take first derivative
delta=delta, # delta for numerical deriv
mode='mirror' # Reflective boundary conds.
)
# construct a LSQ spline interpolant
self.normal.append(LSQUnivariateSpline(xs, f, knots, ext=3, k=3))
if self.DEBUG:
np.savetxt('dlogL_spline_%d.txt' % i,
np.column_stack((xs, Vs, self.normal[-1](xs), f)))
def open_coordinate_figure(request):
selectedVal = request.GET['selectedVal']
byGeoLat = True if request.GET['byGeoLat'] == 'true' else False
if (byGeoLat):
df = pd.read_sql(
'select "StationName", "Latitude", "' + selectedVal +
'" from "world_station_supermag" order by "Latitude"', connection)
indexNames = df[df[selectedVal] > 0.7].index
df.drop(indexNames, inplace=True)
df.sort_values('Latitude')
x = df["Latitude"]
else:
df = pd.read_sql(
'select "StationName", "mlatitude", "' + selectedVal +
'" from "world_station_supermag" order by "mlatitude"', connection)
indexNames = df[df[selectedVal] > 0.7].index
df.drop(indexNames, inplace=True)
df.sort_values('mlatitude')
x = df["mlatitude"]
y = df[selectedVal]
# f2 = interp1d(x, y, kind='cubic')
# plt.grid(True,linestyle='dashed')
# interpolatedOutputList, lnspc = InterpolateData(df, selectedVal, 0)
plt.xticks(np.arange(-90, 90, 10.0))
# plt.plot(x, y, 'ro')
# t = np.linspace(-70, 70, 7)
# t = np.linspace(-70, 70, 5)
t = np.linspace(-70, 70, 5)
spl = LSQUnivariateSpline(np.array(x), np.array(y), t[1:-1])
xs = np.linspace(-90, 90, 1000)
plt.plot(x, y, 'ro', ms=5)
plt.plot(xs, spl(xs), 'g-', lw=3)
# xnew = np.linspace(-90, 90, num=115, endpoint=True)
# for k in (1,2,3): # line parabola cubicspline
# extrapolator = UnivariateSpline( x, y, k=k )
# y = extrapolator(xnew)
# label = "k=%d" % k
# # print(label, y)
# plt.plot( xnew, y, label=label) # pylab
fig = plt.gcf()
# fig.set_size_inches(18.5, 10.5)
FigureCanvasAgg(fig)
buf = io.BytesIO()
plt.savefig(buf, format='png')
plt.close(fig)
response = HttpResponse(buf.getvalue(), content_type="image/png")
response[
'Content-Disposition'] = 'attachment; filename="Result' + selectedVal + '.png"'
return response
def _gufunc_unispline_noerr_upscale(x,
y,
num_knots,
ix,
out=None): # pragma: no cover
xi = x.min()
xf = x.max()
t = np.linspace(xi, xf, num_knots)[1:-1]
fn_interp = LSQUnivariateSpline(x, y, t=t)
out[:] = fn_interp(ix)
def from_knots_coeffs(knots, coeffs, k=3):
n = len(knots) + 2*k
t = np.empty(n, dtype="d")
t[:k] = knots[0]
t[k:-k] = knots
t[-k:] = knots[-1]
c = np.zeros(n, dtype="d")
c[:len(coeffs)] = coeffs
return LSQUnivariateSpline._from_tck((t, c, k))
def calc_coeffs(self):
coeff = []
for v, (g, kdata) in enumerate(zip(self.basegrid, self.knot_data)):
ranges = [(slice(r[0], r[1], t[0]*1j) if kd.used() else slice((r[1]-r[0])/2.0,r[1],1j)) for r, t, kd in zip(self.ranges,g[0],kdata)]
fnc = self.calc_grid(ranges, g[1], g[2])[v]
for n in range(len(ranges)):
kd = self.knot_data[v,n]
if not kd.used():
continue
s = list(fnc.shape)
s[n] = kd.num_coeffs()
out = np.empty(s)
s[n] = 1
x = np.mgrid[ranges[n]]
for i in np.ndindex(tuple(s)):
i = list(i)
i[n] = slice(None)
ss = LSQUnivariateSpline(x, fnc[i], kd.get_knots(), bbox=kd.get_bbox(), k=kd.get_order()-1)
out[i] = ss.get_coeffs()
fnc = out
coeff.append(fnc)
return coeff
def fit_cubic(self):
"""Fit (x,y) arrays using a cubic Spline approximation.
It uses LSQUnivariateSpline from scipy.interpolate
which uses order-1 knots.
"""
x = self.x
y = self.y
xmin,xmax = x.min(),x.max()
npieces = self.order
rn = (xmax - xmin)/npieces
tn = [(xmin+rn*k) for k in range(1,npieces)]
zu = LSQUnivariateSpline(x,y,tn)
self.coeff = zu.get_coeffs()
self.fx = lambda x: x
self.evfunc = lambda p,x: zu(x)
self.ier = None
return
def deserialize(data):
"""Robert's deserialization code."""
self = LSQUnivariateSpline.__new__(LSQUnivariateSpline)
self._data = data
self._reset_class()
return self
coeffs_list = []
for x, y in r["cent_bottom_list"]:
msk = ~y.mask & np.isfinite(y.data)
x1, y1 = x[msk], y.data[msk]
xmin = x1.min()
xmax = x1.max()
# i1 = max(np.searchsorted(t, xmin)-1, 0)
# i2 = np.searchsorted(t, xmax)
m = (xmin <= t) & (t <= xmax)
m1 = ni.maximum_filter1d(m, 3)
t1 = t[m1]
spl = LSQUnivariateSpline(x1, y1,
t1[1:-1], bbox = [t1[0], t1[-1]])
spl = LSQUnivariateSpline(x1, y1,
t2[1:-3], bbox = [t2[0], t2[-3]])
coeffs = np.empty(len(t)+2, dtype="f")
coeffs.fill(np.nan)
if np.all(m1):
coeffs[:] = spl.get_coeffs()
else:
coeffs[1:-1][m1] = spl.get_coeffs()[1:-1]
coeffs_list.append([coeffs])
print len(spl.get_knots()), len(spl.get_coeffs())
plot(x1, y1)
