pastas.stats.tests.diagnostics#

diagnostics(series, alpha=0.05, nparam=0, lags=15, stats=(), float_fmt='{0:.2f}')[source]#

Methods to compute various diagnostics checks for a time series.

Parameters:

  • series (pandas.Series) – Time series to compute the diagnostics for.

  • alpha (float, optional) – significance level to use for the hypothesis testing.

  • nparam (int, optional) – Number of parameters of the noisemodel.

  • lags (int, optional) – Maximum number of lags (in days) to compute the autocorrelation tests for.

  • stats (tuple, optional) – Tuple with the diagnostic checks to perform. Not implemented yet.

  • float_fmt (str) – String to use for formatting the floats in the returned DataFrame.

Returns:

df – DataFrame with the information for the diagnostics checks. The final column in this DataFrame report if the Null-Hypothesis is rejected. If H0 is not rejected (=False) the data is in agreement with one of the properties of white noise (e.g., normally distributed).

Return type:

Pandas.DataFrame

Notes

Different tests are computed depending on the regularity of the time step of the provided time series. pd.infer_freq is used to determined if the time steps are regular.

Examples

>>> data = pd.Series(index=pd.date_range(start=0, periods=1000, freq="D"),
>>>                 data=np.random.rand(1000))
>>> ps.stats.diagnostics(data)
Out[0]:
                  Checks Statistic P-value  Reject H0
Shapiroo       Normality      1.00    0.86      False
D'Agostino     Normality      1.18    0.56      False
Runs test      Autocorr.     -0.76    0.45      False
Durbin-Watson  Autocorr.      2.02     nan      False
Ljung-Box      Autocorr.      5.67    1.00      False

In this example, the Null-hypothesis is not rejected and the data may be assumed to be white noise.