CODEWITHSUNDEEP
pandasmissing-datastatsmodels
eternity1
eternity1

Reputation: 681

Statsmodels: requires arrays without NaN or Infs - but test shows there are no NaNs or Infs

I am trying to run an ADF-test from the statsmodels' adfuller module. It gives me an error: 

ValueError: array must not contain infs or NaNs

Via another question I could replace my NaNs (NaN in data frame: when first observation of time series is NaN, frontfill with first available, otherwise carry over last / previous observation.

Although I checked for NaNs and inf:

df[pd.isnull(df).any(axis=1)]
np.isinf(df).any()
np.isnan(df).any()

which gives me "no results" for the pandas statement and both "false" for the numpy statements, my function still tells me the same error.

Is there a bug?

ValueError: array must not contain infs or NaNs during Biclustering

Upvotes: 1

Views: 2923

Answers (1)

eternity1
eternity1

Reputation: 681

I have solved it now via:

 x = pd.DataFrame(x.replace([np.inf, -np.inf], np.nan))
 x = x.fillna(method='ffill')
 x = x.fillna(method='bfill')
 x = x.iloc[:, 0]

which gives me a x series without any nan of inf. Problem was the switch from series to dataframe and the back to series

Upvotes: 1

Related Questions