SARIMA learner & predictor

[2ulaikha]

Hi All,

I have a seasonal time series data that I wanted to forecast but I can’t find SARIMA learner & SARIMA predictor in node repository. Is there any extension that I need to install first to make the nodes appear in the node repository?

Thanks

Siti

[ScottF]

Hi there @2ulaikha -

We have recently released verified components for SARIMA on the KNIME Hub, with an associated example workflow built by @corey. Check the links below for more:

SARIMA – KNIME Hub

SARIMA_Example – corey

[2ulaikha]

Hi @ScottF , thanks for your reply.

I tried to execute the example workflow built by @Corey but the execution failed on SARIMA learner, SARIMA predictor and Visualize Forecast nodes. The console box stated that error python predictor : no module named ‘statsmodels’. What should I do?

[ScottF]

Hi @2ulaikha -

Assuming you already have Python setup to work with him as described in the documentation (KNIME Python Integration Guide) - and it sounds like you do - then you just need to add the statsmodels package to your Python environment.

You can do this by opening an Anaconda command prompt, activating the Python environment used by KNIME, and using conda install statsmodels.

[2ulaikha]

Hi @ScottF

I have installed statsmodels package through Anaconda command prompt as you suggested, but the errors still appear. The errors as per below:

WARN Python Learner 0:0:0:105 C:\Users\siti.zulaikha\Anaconda3\envs\SARIM\lib\site-packages\statsmodels\tsa\base\tsa_model.py:214: ValueWarning: An unsupported index was provided and will be ignored when e.g. forecasting.
WARN Python Learner 0:0:0:105 C:\Users\siti.zulaikha\Anaconda3\envs\SARIM\lib\site-packages\statsmodels\base\model.py:568: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
ERROR Python Predictor 0:0:0:331 Execute failed: No module named 'statsmodels

[Daniel_Weikert]

you have installed it in your base environment. You seem to use py3_knime environment in KNIME so the package is still missing there I suppose
br

[ScottF]

I was just going to suggest what @Daniel_Weikert said - he’s too fast!

In the Anaconda command prompt, try activate py3_knime first before adding the statsmodels package.

[2ulaikha]

Finally it works! Thank you very much @ScottF for your help. Great to have someone in the forum offering a quick solution like you

[genchak]

Hi @ScottF . I have a different error message for SARIMA. I have installed all the packages with exactly the way you listed above. Here is the message I got : "ERROR Python Predictor 0:0:0:331 Execute failed: ‘initialized_diffuse’ " . I searched for this error and I couldn’t find a solution for this. Would you check, if you have time please? Thank you.

[genchak]

Here are some more details from Console:

WARN Python Learner 0:0:0:105 C:\Users\udnha2\AppData\Local\Continuum\anaconda3\envs\SARIM\lib\site-packages\statsmodels\tsa\base\tsa_model.py:214: ValueWarning: An unsupported index was provided and will be ignored when e.g. forecasting.
WARN Python Learner 0:0:0:105 C:\Users\udnha2\AppData\Local\Continuum\anaconda3\envs\SARIM\lib\site-packages\statsmodels\tsa\base\tsa_model.py:214: ValueWarning: An unsupported index was provided and will be ignored when e.g. forecasting.
WARN Python Learner 0:0:0:105 C:\Users\udnha2\AppData\Local\Continuum\anaconda3\envs\SARIM\lib\site-packages\statsmodels\base\model.py:568: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals

[Corey]

Hi @genchak,

I’ve been investigating this error and finally figured it out.
Due to a small change in the Conda environment propagation node with KNIME v4.4 some of the python scripts where not using the proper python environment inside the component.

I’m working to push the corrected component to our example server but in the fix is super simple:

1. Disconnect the component, you can do this by right clicking and going into the component sub menu, this enables you to edit the component:
   

   

2. Open the component, you can do this from the same Menu you see above but select open. Or hold control and double click. You’ll need to make a small edit to the nodes I’ve circled in red here:
   

   

3. Open the python script nodes inside the component, navigate to the flow variables tab, and verify that the bottom line, python3Command, is populated with the Conda.environment variable.
   

   

After that everything should run fine. Apologies for the inconvenience. I’ll also try to post here again when the fix has been applied to the component on the example server.

[genchak]

Hi @Corey Yes, perfect. It worked. Thank you very much. Hakan

[Corey]

Wanted to circle back to this post to let everyone know that the fix has been pushed to the Published version of the SARIMA components.

If you download the SARIMA components from the KNIME Hub, or from the example server they should work just fine now.

Thanks everyone!