Auto ARIMA learner failure

[iTree]

Hello, I got an error with the Auto ARIMA learner component-
“ERROR Python Learner 3:79:0:105 Execute failed: name ‘i_out’ is not defined
Traceback (most recent call last):
File “”, line 47, in
NameError: name ‘i_out’ is not defined”

I had this error message running my workflow as well as the example workflow on Knime hub.

Accessing, Transforming and Modeling Time Series – knime

I can see i_out defined in the Python learner node in the component though. Can anyone help?

[NDekay]

Hello,

What version of KNIME AP and Conda/Python are being used when you are testing this?

Are you able to provide the knime.log file from the attempted execution?

Thank you,
Nickolaus

[iTree]

KNIME AP 4.5.1
see Python setup below, also attached knime log. Thanks!

knime log 03-09.txt (3.5 KB)

[NDekay]

Hello,

After installing
AP 4.5.1
miniconda3 (conda 4.11.0, python 3.9.7, statsmodels packages)
ATMTS workflow
data source file from tableau

and executing the workflow, I have reproduced the error you described. I don’t have any end-user configuration changes to recommend; I have submitted a defect report for this issue and will follow up with you once I get any further updates back on the defect.

Thank you,
Nickolaus

[NDekay]

Hello @iTree,

To follow up on this, our engineers have reproduced the issue and found the following:

• the issue is to do with a change in API of the statsmodels package. Namely, from version 0.11.0, statsmodels.tsa.arima_model.ARIMA was changed to statsmodels.tsa.arima.model.ARIMA, see the discussion here.
• In order to make the worflow work correctly, I would suggest:
  • creating a conda environment via the KNIME Preferences page as described in KNIME Python Integration Guide for Python version 3.8 (this is important, since this is the last Python version supported by statsmodels v0.10.2).
  • installing the statsmodels package into the created env via conda install -c conda-forge statsmodels=0.10.2. Note the version - that is the last version of the package that supports the API used in the Auto ARIMA learner.

With the system configured as above, everything worked.

Regards,
Nickolaus