Machinelearningnotebooks: AMLConfig target_lags param ambiguity?

[swanderz]

Why does the description for target_lags, but only allow int ? I would love to pass multiple lags

This is represented as a list or single integer

> When forecasting, this parameter represents the number of rows to lag the target values based on the frequency of the data. This is represented as a list or single integer. Lag should be used when the relationship between the independent variables and dependent variable do not match up or correlate by default. For example, when trying to forecast demand for a product, the demand in any month may depend on the price of specific commodities 3 months prior. In this example, you may want to lag the target (demand) negatively by 3 months so that the model is training on the correct relationship.

Document Details

⚠ Do not edit this section. It is required for docs.microsoft.com ➟ GitHub issue linking.

• ID: 0bc2b21e-6b1a-cb94-2857-147177a29d7c
• Version Independent ID: d14620a6-a2f6-49f1-632e-73903d41de8c
• Content: azureml.train.automl.automlconfig.AutoMLConfig class - Azure Machine Learning Python
• Content Source: AzureML-Docset/stable/docs-ref-autogen/azureml-train-automl-client/azureml.train.automl.automlconfig.AutoMLConfig.yml
• Service: machine-learning
• Sub-service: core
• GitHub Login: @j-martens
• Microsoft Alias: jmartens

[RohitMungi-MSFT]

@Aniththa The documentation for target_lagsindicates it is of type int but it looks like it should be list(int) according to the description of the field. Could you please review and correct the same?

[swanderz]

There's a few other forecasting params that should also be list(int) I believe. Plus the UI doesn't even let you enter a list and tells you it must be an integer. @nibaccam

[cartacioS]

You are both correct, we have work items opened now to update the documentation to say that target_lags accept a list as well as an integer. As of today UI only supports the single integer as it's directed towards a different user persona. Would be happy to sync and understand if this is not meeting yours or other UI customers needs.

Could you point out the other forecasting params that should be list but aren't documented correctly?

[swanderz]

Well now I'm just not sure... maybe it is more about clarifying fixing some mislabeled things.

lag_length?

in the function def at the top of AutoMLConfig, there's mention of lag_length: int = 0, but I assume that its been renamed to target_lags, right?

Can I add multiple lags to lag_length? In the blog post , it mentions how passing lag_length=2 creates a 1-day lag and a 2-day lag? So is lag_length

The table below illustrates how a lag length of two would be treated. Green columns are engineered features with lags of sales by one day and two day. The blue arrows indicate how each of the lags are generated by the training data. Not a number (Nan) are created when sample data does not exist for that lag period.

target_rolling_window_size

Can I create multiple lags? like I can with target_lags?

[vmagelo]

re: update of target_lags parameter in AutoMLConfig class: this is included in a PR in progress for updating docs.

[cartacioS]

Lag_length is an outdated parameter before target_lags were introduced. We will be removing this parameter from our documentation. Though the blog (due to the nature of blogs) will remain as is.

Target_lags does accept a list of integers and those are the doc updates we have created work items for and will address.

[vmagelo]

Updated docs for target_lags. Changes will appear in next release.

[nibaccam]

Thank you for reporting this @swanderz . The docs update has been checked in. We will now proceed to close this thread. #please-close