Question about predicted values from Arima model parameter optimization loop start node

Hi there. Dear all

I am currently analyzing an Arima model and have a question about the Forecast table, which is the 1st port of the Arima Predictor.

In the workflow below, I use the “Parameter Optimization Loop Start” node to pick the optimal parameters “AR”, “I”, and “MA” with the minimum mse value.

I have a question here.

  1. Is Forecast, the 1st port of the ARIMA Predictor node, the result of predicting with the optimal parameters derived through the Parameter Optimization Loop node above?

  2. If it is not the 1st, does the Arima model not need to use the Loop statement to select the optimal model and make a forecast, but simply use the ARIMA Predictor node’s forecast obtained with the default values?

I would be grateful for your answers.

Hello @JaeHwanChoi,

Thank you so much for posting here. The ARIMA nodes that you have used are under development and are not out of labs yet. I would recommend you to check out the verified components instead.

Thanks,

Best,
Ali

1 Like

Thank you for your response. @aliasghar_marvi !

When you say Verified components, do you mean the Arima node corresponding to “ARIMA Learner – KNIME Community Hub”?

I checked inside the component and it was developed using a Python script, so it seems that it cannot be used unless the environment is configured.

Thanks,

Best,
Choi

Hi @JaeHwanChoi,

You can choose to use bundle packages that way you do not have to worry about additional configuration of the environment. In preferences you can choose the following:

Thanks

Best,
Ali

1 Like

Hi. @aliasghar_marvi

I have a question.

When a user uses KNIME’s Python Script on their personal Local PC,

1). Anaconda must be installed on the personal PC to set the path to File → Preferences → Conda, and 2). File → Preferences → Python, they can configure a new virtual environment with conda and use it, right?

I would be grateful if you could answer these two questions.

Best,
Choi

Hi @JaeHwanChoi ,

Apologies for being late. I was away for several days.

  1. So if your script is using a package that comes bundled with KNIME (eg. pandas, pyarrow, sci-kit, statsmodel, etc…) then you do not need to worry about conda environments. Otherwise even a miniconda installation on your system shall suffice as well.

  2. Precisely, but then you do not need a dedicated conda node. Now lets say you have to share that node with someone, you would need a Conda Environment Propagation with all the required packages. That way you can select that specific environment through Conda Environment Propagation node and pre configure it with Python script and you should be good.

I am certain you will have more questions around it still. Therefore, we have a dedicated KNIME space which explains use of Python scripting with KNIME. Here:

Best,
Ali

1 Like

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.