search

chengqianghuang / exp-anomaly-detector-AIOps

Using DNN for univariate time series anomaly detection over AIOps Competition dataset
44 stars 29 forks source link

ValueError: Non-stationary starting autoregressive parameters found with `enforce_stationarity` set to True. #1

Open PuHaoran opened 5 years ago

PuHaoran commented 5 years ago

I am getting a "ValueError: Non-stationary starting autoregressive parameters found with enforce_stationarity set to True." error when doing the feature extraction step（get_feature_SARIMA_residuals function）. The version of my statsmodels is 0.9.0. I set enforce_stationarity = False and enforce_invertibility = False, but get some Nan values. Excuse me, do you know the solution?

chengqianghuang commented 5 years ago

I am getting a "ValueError: Non-stationary starting autoregressive parameters found with enforce_stationarity set to True." error when doing the feature extraction step（get_feature_SARIMA_residuals function）. The version of my statsmodels is 0.9.0. I set enforce_stationarity = False and enforce_invertibility = False, but get some Nan values. Excuse me, do you know the solution?

Hi Haoran,

The statsmodels is not really stable when I try to do the experiments. My suggestion would be just to get rid of the feature that is created using statsmodels, e.g., get_feature_SARIMA_residuals, and modify corresponding parts so as to train the model.

In my experiments, successfully using statsmodels only increase the accuracy by 1-4 percents.

Good luck. @PuHaoran