How to construct response curves in original scale? How to get sigmoid params?

[Snowcatcat]

Constructing the response curves used to be 1 line of code:
sigmoid_params = mmm.compute_channel_curve_optimization_parameters_original_scale(method='sigmoid')

Does anyone know how to do it now? In the original scale. I have not updated my package for a while, looks like this function has retried and not in the instructions any more.

Thanks

[williambdean]

Is this the function that you are looking for?

pymc-marketing/pymc_marketing/mmm/mmm.py

Lines 1801 to 1834 in a87f1fa

	def plot_direct_contribution_curves(
	self,
	show_fit: bool = False,
	same_axes: bool = False,
	xlim_max: int | None = None,
	channels: list[str] | None = None,
	quantile_lower: float = 0.05,
	quantile_upper: float = 0.95,
	method: str | None = None,
	) -> plt.Figure:
	"""Plot the direct contribution curves for each marketing channel.
	The term "direct" refers to the fact that we plot costs vs immediate returns and
	we do not take into account the lagged effects of the channels e.g. adstock transformations.
	Parameters
	----------
	show_fit : bool, optional
	If True, the function will also plot the curve fit based on the specified method. Defaults to False.
	xlim_max : int, optional
	The maximum value to be plot on the X-axis. If not provided, the maximum value in the data will be used.
	channels : List[str], optional
	A list of channels to plot. If not provided, all channels will be plotted.
	same_axes : bool, optional
	If True, all channels will be plotted on the same axes. Defaults to False.
	method : str | None, optional
	Deprecated.
	Returns
	-------
	plt.Figure
	A matplotlib Figure object with the direct contribution curves.
	"""

You can see that it is in the original scale by default:

pymc-marketing/pymc_marketing/mmm/mmm.py

Lines 1853 to 1855 in a87f1fa

	channel_contributions = self.compute_channel_contribution_original_scale().mean(
	["chain", "draw"]
	)

The method doesn't need to be passed since that is derived from the model definition.

mmm = ...
mmm.fit(X, y)
mmm.plot_direct_contribution_curves(show_fit=True)

[Snowcatcat]

Hi @williambdean ! Thanks for your response! Seeing the curves is good, but I would like to know how to construct the curves in the excel. the previous version offers this code: sigmoid_params = mmm.compute_channel_curve_optimization_parameters_original_scale(method='sigmoid') and gives us something like this:
{'audio_S': array([2.52077197e+02, 1.42054675e-05]),
'ctv_S': array([2.80807420e+02, 5.95599636e-06]),
'display_S': array([7.21954207e+02, 2.08661774e-06]),
'native_S': array([1.48818804e+03, 1.07180699e-05]),
'ooh_S': array([3.14424279e+02, 7.22348481e-06]),
'tv_S': array([5.63911138e+03, 1.51360258e-06]),}

Using these sigmoid parameters, we can easily construct the response curves in original scale anywhere. I'm wondering how to do something like this in pymc marketing 0.12?

[AdimDrewnik]

Hi. Did you solve your question? What is actually a direct contribution? It is not obvious from the documentation.