def stl_decompose(df, column, freq=52):
dfd = pd.DataFrame(index=df.index)
series = list(df[column].values)
length = len(series)
rts = r.ts(series, frequency=freq)
decomposed = list(r.stl(rts, 'periodic', robust=True).rx2('time.series'))
dfd['trend'] = decomposed[length:2 * length]
dfd['seasonal'] = decomposed[0:length]
dfd['residuals'] = decomposed[2 * length:3 * length]
return dfd
def decompose(series, frequency, s_window, **kwargs):
df = pd.DataFrame(index=series.index)
#df['date'] = series.index
series.interpolate(inplace=True)
s = [x for x in series.values]
length = len(series)
s = r.ts(s, frequency=frequency)
decomposed = [x for x in r.stl(s, s_window, **kwargs).rx2('time.series')]
df['observed'] = series.values
df['trend'] = decomposed[length:2*length]
df['seasonal'] = decomposed[0:length]
df['residual'] = decomposed[2*length:3*length]
return df
def decompose(series,
frequency,
s_window='periodic',
log=False,
theme=False,
**kwargs):
'''
Decompose a time series into seasonal, trend and irregular components using loess,
acronym STL.
https://www.rdocumentation.org/packages/stats/versions/3.4.3/topics/stl
params:
series: a time series
frequency: the number of observations per “cycle”
(normally a year, but sometimes a week, a day or an hour)
https://robjhyndman.com/hyndsight/seasonal-periods/
s_window: either the character string "periodic" or the span
(in lags) of the loess window for seasonal extraction,
which should be odd and at least 7, according to Cleveland
et al.
log: boolean. take log of series
theme: a bokeh theme
**kwargs: See other params for stl at
https://www.rdocumentation.org/packages/stats/versions/3.4.3/topics/stl
'''
df = pd.DataFrame()
df['date'] = series.index
if log: series = series.pipe(np.log)
s = [x for x in series.values]
length = len(series)
s = r.ts(s, frequency=frequency)
decomposed = [x for x in r.stl(s, s_window).rx2('time.series')]
df['observed'] = series.values
df['trend'] = decomposed[length:2 * length]
df['seasonal'] = decomposed[0:length]
df['residuals'] = decomposed[2 * length:3 * length]
return df
def auto_arima(endog, exog=None, freq=None):
if freq is None:
freq = 1
# endog_r = r.ts(pandas2ri.py2ri(endog), freq=freq)
# if using more recent version of rpy2, py2ri was renamed to py2rpy
# see reference: https://stackoverflow.com/questions/55990529/module-rpy2-robjects-pandas2ri-has-no-attribute-ri2py
endog_r = r.ts(pandas2ri.py2rpy(endog), freq=freq)
autoarima_args = {
"seasonal": True,
"stationary": False,
"trace": True,
"max.order": 20,
"max.p": 20,
"max.q": 20,
"max.P": 20,
"max.Q": 20,
"max.D": 20,
"max.d": 20,
"start.p": 1,
"start.q": 1,
"start.P": 1,
"start.Q": 1
}
if exog is not None:
# add noise to avoid rank-deficient error for exog
scale = np.std(exog.values)
z = scale * 1e-4 * np.random.randn(*exog.shape)
exog_r = r.matrix(exog.values + z,
nrow=exog.shape[0],
ncol=exog.shape[1],
dimnames=[[], exog.columns.tolist()])
fit_r = forecast.auto_arima(y=endog_r, xreg=exog_r, **autoarima_args)
else:
fit_r = forecast.auto_arima(y=endog_r, **autoarima_args)
fit_dict = dict(fit_r.items())
# for proof of this order see last comment:
# https://stats.stackexchange.com/questions/178577/how-to-read-p-d-and-q-of-auto-arima
p, q, P, Q, s, d, D = list(fit_dict["arma"])
return (p, d, q), (P, D, Q, s)
def mtm_from_R(x, K=3, NW=6, nFFT='default', plot=0, deltat=0.001, jkCIProb = 0.95, maxf=10, plotftest=True):
from rpy2 import robjects
from rpy2.robjects.packages import importr
from rpy2.robjects import r
MTM = importr('multitaper')
#creating the R-type serie
serie = robjects.FloatVector(x)
serie = r.ts(serie, deltat = deltat)
#executing the mtm function
results = MTM.spec_mtm(serie,k=K, nw=NW, nFFT=nFFT, plot=0, Ftest=1, jackknife=1, jkCIProb = jkCIProb)
#extracting variables from results
freqs = np.array(results.rx2('freq'))
spec = np.array(results.rx2('spec'))
upperCI = np.array(results.rx2('mtm').rx2('jk').rx2('upperCI'))
lowerCI = np.array(results.rx2('mtm').rx2('jk').rx2('lowerCI'))
Ftest = np.array(results.rx2('mtm').rx2('Ftest'))
if plot == 1:
fig = plt.figure()
plt.hold(1)
ax1 = fig.add_subplot(111)
ax1.plot(freqs, spec, 'k-', linewidth=2)
ax1.plot(freqs, upperCI, 'r:')
ax1.plot(freqs, lowerCI, 'g:')
ax1.set_ylabel('Power Spectral Density')
ax1.set_xlabel('Frequency [cycles/m]')
ax2 = ax1.twinx()
if plotftest == True:
ax2.plot(freqs, Ftest, 'y')
ax2.set_ylabel('Ftest')
ax2.set_xlim(0, maxf)
else:
ax1.set_xlim(0, maxf)
plt.show()
return freqs, spec, upperCI, lowerCI, Ftest