Skip to content

Instantly share code, notes, and snippets.

@i-trofimtschuk
Forked from andreas-h/r_stl.py
Created December 17, 2013 02:50
Show Gist options
  • Save i-trofimtschuk/7999186 to your computer and use it in GitHub Desktop.
Save i-trofimtschuk/7999186 to your computer and use it in GitHub Desktop.
# -*- coding: utf-8 -*-
import datetime
from numpy import asarray, ceil
import pandas
import rpy2.robjects as robjects
def stl(data, ns, np=None, nt=None, nl=None, isdeg=0, itdeg=1, ildeg=1,
nsjump=None, ntjump=None, nljump=None, ni=2, no=0, fulloutput=False):
"""
Seasonal-Trend decomposition procedure based on LOESS
data : pandas.Series
ns : int
Length of the seasonal smoother.
The value of ns should be an odd integer greater than or equal to 3.
A value ns>6 is recommended. As ns increases the values of the
seasonal component at a given point in the seasonal cycle (e.g., January
values of a monthly series with a yearly cycle) become smoother.
np : int
Period of the seasonal component.
For example, if the time series is monthly with a yearly cycle, then
np=12.
If no value is given, then the period will be determined from the
``data`` timeseries.
nt : int
Length of the trend smoother.
The value of nt should be an odd integer greater than or equal to 3.
A value of nt between 1.5*np and 2*np is recommended. As nt increases,
the values of the trend component become smoother.
If nt is None, it is estimated as the smallest odd integer greater
or equal to ``(1.5*np)/[1-(1.5/ns)]``
nl : int
Length of the low-pass filter.
The value of nl should be an odd integer greater than or equal to 3.
The smallest odd integer greater than or equal to np is used by default.
isdeg : int
Degree of locally-fitted polynomial in seasonal smoothing.
The value is 0 or 1.
itdeg : int
Degree of locally-fitted polynomial in trend smoothing.
The value is 0 or 1.
ildeg : int
Degree of locally-fitted polynomial in low-pass smoothing.
The value is 0 or 1.
nsjump : int
Skipping value for seasonal smoothing.
The seasonal smoother skips ahead nsjump points and then linearly
interpolates in between. The value of nsjump should be a positive
integer; if nsjump=1, a seasonal smooth is calculated at all n points.
To make the procedure run faster, a reasonable choice for nsjump is
10%-20% of ns. By default, nsjump= 0.1*ns.
ntjump : int
Skipping value for trend smoothing. If None, ntjump= 0.1*nt
nljump : int
Skipping value for low-pass smoothing. If None, nljump= 0.1*nl
ni :int
Number of loops for updating the seasonal and trend components.
The value of ni should be a positive integer.
See the next argument for advice on the choice of ni.
If ni is None, ni is set to 2 for robust fitting, to 5 otherwise.
no : int
Number of iterations of robust fitting. The value of no should
be a nonnegative integer. If the data are well behaved without
outliers, then robustness iterations are not needed. In this case
set no=0, and set ni=2 to 5 depending on how much security
you want that the seasonal-trend looping converges.
If outliers are present then no=3 is a very secure value unless
the outliers are radical, in which case no=5 or even 10 might
be better. If no>0 then set ni to 1 or 2.
If None, then no is set to 15 for robust fitting, to 0 otherwise.
fulloutput : bool
If True, a dictionary holding the full output of the original R routine
will be returned.
returns
data : pandas.DataFrame
The seasonal, trend, and remainder components
"""
# make sure that data doesn't start or end with nan
_data = data.copy()
_data = _data.dropna()
# TODO: account for non-monthly series
_idx = pandas.DateRange(start=_data.index[0], end=_data.index[-1],
offset=pandas.datetools.MonthBegin())
data = pandas.Series(index=_idx)
data[_data.index] = _data
# zoo package contains na.approx
zoo_ = robjects.packages.importr("zoo")
ts_ = robjects.r['ts']
stl_ = robjects.r['stl']
naaction_ = robjects.r['na.approx']
# find out the period of the time series
if np is None:
np = 12
# TODO: find out the offset of the Series, and set np accordingly
#if isinstance(data.index.offset, pandas.core.datetools.MonthEnd):
# np = 12
#else:
# raise NotImplementedError()
# fill default values
if nt is None:
nt = ceil((1.5 * np) / (1 - (1.5 / ns)))
nt = nt + 1 if nt % 2 == 0 else nt
if nl is None:
nl = np if np % 2 is 1 else np + 1
if nsjump is None:
nsjump = ceil(ns / 10.)
if ntjump is None:
ntjump = ceil(nt / 10.)
if nljump is None:
nljump = ceil(nl / 10.)
# convert data to R object
if np is 12:
start = robjects.IntVector([data.index[0].year, data.index[0].month])
ts = ts_(robjects.FloatVector(asarray(data)), start=start, frequency=np)
if nt is None:
nt = robjects.rinterface.R_NilValue
result = stl_(ts, ns, isdeg, nt, itdeg, nl, ildeg, nsjump, ntjump, nljump,
False, ni, no, naaction_)
res_ts = asarray(result[0])
try:
res_ts = pandas.DataFrame({"seasonal" : pandas.Series(res_ts[:,0],
index=data.index),
"trend" : pandas.Series(res_ts[:,1],
index=data.index),
"remainder" : pandas.Series(res_ts[:,2],
index=data.index)})
except:
return res_ts, data
raise
# res_ts = pandas.DataFrame({"seasonal" : pandas.Series(index=data.index),
# "trend" : pandas.Series(index=data.index),
# "remainder" : pandas.Series(index=data.index)})
if fulloutput:
return {"time.series" : res_ts,
"weights" : result[1],
"call" : result[2],
"win" : result[3],
"deg" : result[4],
"jump" : result[5],
"ni" : result[6],
"no" : result[7]}
else:
return res_ts
if __name__ == "__main__":
data = np.arange(85.) / 12.
data = np.sin(data * (2*np.pi))
data += np.arange(85.) / 12. * .5
data += .1 * np.random.randn(85)
idx = pandas.DateRange(start=datetime.datetime(1999,1,1), end=datetime.datetime(2006,2,1), offset=pandas.datetools.MonthEnd())
data = pandas.Series(data, index=idx)
res = stl(data, 7, nt=11)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment