Missing Method: get_paid_media_channels_argument_builder() Not Found in meridian.input_data

[HoussemBouazizi]

🔍 Issue Summary

The function get_paid_media_channels_argument_builder() is referenced in the Meridian documentation, but it does not exist in meridian.input_data. This causes an AttributeError when trying to call it as shown in the official documentation. Additionally, when attempting to use the InputData class as a workaround, it forces a geographic hierarchy model, which is not suitable for my use case as I am working on a national model without a geo dimension.

📌 Expected Behavior

• The method meridian.input_data.get_paid_media_channels_argument_builder() should exist and return an argument builder for paid media channels.
• National models should have a way to access this method without requiring InputData and geographic dimensions.

🐛 Observed Behavior

Calling:

build_media_channel_args = meridian.input_data.get_paid_media_channels_argument_builder()

Results in the error:

AttributeError: module 'meridian.input_data' has no attribute 'get_paid_media_channels_argument_builder'

Attempting to use the InputData class to access the method forces me to provide geo, which does not apply to my national model setup.

🔗 References

• [[Meridian Documentation](https://developers.google.com/meridian/docs) ](https://developers.google.com/meridian/docs/user-guide/configure-model#default-spec)
• [Meridian GitHub Repository](https://github.com/google/meridian)

🙏 Request

Could you confirm whether get_paid_media_channels_argument_builder() is supposed to work without requiring InputData, or if an alternative exists for national models that do not include a geo dimension? If this is a bug, please clarify how national models should handle media channel arguments. Thanks for your help! 🙌

[santoso-wijaya]

Hi,

We updated the documentation in tandem with commit 4521db1 that we pushed yesterday.

However, it has not been published to PyPI yet. If you had installed Meridian with pip install google-meridian, it will pull the latest Meridian package (v.1.0.3) without this recent change.

Some workarounds:

• You can access get_paid_media_channels_argument_builder() if you installed Meridian directly from GitHub at HEAD commit with pip install --upgrade git+https://github.com/google/meridian.git
• Pass arrays directly to the parameters that expect channel-indexed values without the facilitation of get_paid_media_channels_argument_builder(). get_paid_media_channels_argument_builder() makes this easier, but you can still pass the optimizer naked arrays (take care of channel ordering, though).
• Wait until we publish v1.0.4 to PyPI later today with these changes included.

---

I'm not sure what you meant by using "InputData class as a workaround, it forces a geographic hierarchy model, which is not suitable for my use case as I am working on a national model without a geo dimension." You should be able to create an InputData whose data array constituents do not include a geo dimension--effectively making it a national model. Can you share an example stack trace that you came upon?

[santoso-wijaya]

Could you confirm whether get_paid_media_channels_argument_builder() is supposed to work without requiring InputData

You can bypass the requirement of InputData by creating the argument builder directly:

from meridian.data import arg_builder

build_media_channel_args = arg_builder.OrderedListArgumentBuilder([
    'Channel1',
    'Channel2',
    'Channel3',
    'Channel4',
    'Channel5',
    'Channel6'
])

# The rest is the same as the code snippet in https://developers.devsite.corp.google.com/meridian/docs/user-guide/configure-model

my_input_data.get_paid_media_channels_argument_builder() just facilitates the seeding of the argument builder with your media channel names.

However, this also requires the latest Meridian code at HEAD. If you're still working with [email protected] or earlier, you will have to create, for instance, roi_m_mu and roi_m_sigma manually. For example:

roi_m_mu    = [0.2, 0.3, 0.4, 0.3, 0.3, 0.4]  # must match media channel ordering in InputData
roi_m_sigma = [0.7, 0.9, 0.6, 0.7, 0.6, 0.5]  # must match media channel ordering in InputData

prior = prior_distribution.PriorDistribution(
    roi_m=tfp.distributions.LogNormal(
        roi_m_mu, roi_m_sigma, name=constants.ROI_M
    )
)