def form_data(self):
''' Form time series data used in lm model.
'''
assert self.t is not None, "t is not defined."
assert self.y is not None, "y is not defined."
globalenv['t'] = FloatVector(self.t)
globalenv['y'] = FloatVector(self.y)
if self.ysd is None:
globalenv['ysd'] = FloatVector(ones_like(self.y))
else:
globalenv['ysd'] = FloatVector(self.ysd)
_r('the_data <- data.frame(t=t,y=y,ysd=ysd)')
def predict(self, t, level=0.95, interval='none'):
''' Interface of R predict function.
'''
globalenv['t_pred'] = FloatVector(t)
y=array(_r('predict(fit,data.frame(t=t_pred),level=%f,interval="%s")'%\
(level,interval)))
return y
def get_res(self):
''' Return residual of the part of time series that is used in linear regression.
Nan is there is no residula available.
'''
tp = asarray(_r('resid(fit)'))
res = zeros_like(self.t) + nan
res[self.subset] = tp
return res
def __init__(self):
# initialize the r process
_r('''
# define a heaviside function:
hvsd <-
function(x, a = 0){
result = (sign(x-a) + 1.)/2.
result
}
T<-365.
Omega<-2*pi/T
''')
## @var if_sea
# If consider seasonal signal?
self.if_sea = None
## @var if_semi
# if consider semisesonal signal?
self.if_semi = None
## jumps list
self.jumps = None
## data_par : t
self.t = None
## y
self.y = None
## ysd
self.y = None
## linear sections used for linear regression.
self.linsecs = None
## record the outliers
self.if_outlier = None
## outlier criteri (times of st):
self.outlier_cri = 3.0
def lm(self, verbose=True):
''' Do linear regression.
'''
self.form_model()
self.form_data()
self.form_subset()
_r('fit <- lm(fml,the_data,subset)')
nth = 1
if verbose:
print('%dth iter: std(mm)=%.3f vel(mm/yr)=%.3f outliers #:%d'%\
(nth,self.get_res_std()*1000.,self.get_vel()*365.*1000.,
sum(self.if_outlier)))
nth += 1
while self.mark_outliers() > 0:
self.form_subset()
#_r('fit <- lm(fml,the_data,subset,weights=1/ysd^2)')
_r('fit <- update(fit,.~.,subset=subset,weights=1/ysd^2)')
if verbose:
print('%dth iter: std(mm)=%.3f vel(mm/yr)=%.3f outliers #:%d'%\
(nth,self.get_res_std()*1000.,self.get_vel()*365.*1000.,
sum(self.if_outlier)))
nth += 1
def get_magnitude_semiseasonal(self):
S = _r("fit$coe[['sin(2 * Omega * t)']]")[0]
C = _r("fit$coe[['cos(2 * Omega * t)']]")[0]
return sqrt(S**2 + C**2)
def get_jumps(self):
''' Reture jumps and their value.
'''
for jump in self.jumps:
yield (jump, _r("fit$coe[['hvsd(t, %d)']]" % jump)[0])
def get_vel_sd(self):
'''
'''
return _r("summary(fit)$coef['t','Std. Error']")[0]
def get_vel(self):
'''
'''
return _r("fit$coe[['t']]")[0]
def get_res_time_series(self):
res = asarray(_r('resid(fit)'))
return res, self.t[self.subset]
def get_res_std(self):
''' Return standard error of residuls.
'''
return float(_r("summary(fit)$sigma")[0])
if __name__ == '__main__':
# local testing code
mod = PreRModel()
# model:
mod.if_sea = True
mod.if_semi = True
mod.jumps = []
# data:
tp = loadtxt('../J550.IGS08')
t = tp[:, 0]
e = tp[:, 1] * 1000.
t_eq = 55631
mod.t = t
mod.y = e
mod.linsecs = [[-inf, 55631]]
mod.lm()
print(_r('summary(fit)'))
print(_r('coef(summary(fit))'))
plot(mod.t, mod.get_res(), 'x')
res = mod.predict_res(mod.t[mod.if_outlier], mod.y[mod.if_outlier])
plot(mod.t[mod.if_outlier], res, 'r.')
show()
