Morpheus
      Back to home
      1. Knowledge base
      2. Morpheus

      Understanding Hyperparameters in Morpheus for Marketing Mix Modeling (MMM)

      Learn how hyperparameters in Morpheus improve your Marketing Mix Modeling. Understand each available hyperparameter, its function, and why they are beneficial.

      What Are Hyperparameters in Marketing Mix Modeling?

      Hyperparameters in MMM are predefined settings that shape how a model learns from data. Unlike model parameters, which are learned from the data itself (e.g., the impact of TV ads on sales), hyperparameters are set before training and influence the model's behavior.

      For example, in MMM, hyperparameters can control how much past marketing spend influences future performance (adstock effect), how quickly returns diminish as spending increases (saturation), or how external factors like seasonality are handled.

      Why Are Hyperparameters Important?

      • Improve Model Accuracy – Fine-tune the model to better reflect real-world marketing dynamics.
      • Prevent Overfitting – Ensure the model captures meaningful trends without being too sensitive to noise.
      • Enhance Interpretability – Help marketers and analysts understand how media channels contribute to business outcomes.
      • Adapt to Business-Specific Insights – Allow customization based on industry knowledge or historical marketing behavior.

      While adjusting hyperparameters can enhance model performance, modern MMM tools like Morpheus can estimate them automatically, making their use optional.

      Hyperparameters in Morpheus

      Important: Morpheus auto-estimates hyperparameters using Bayesian techniques—no manual setup needed. Fine-tuning is available for experts with prior marketing insights.

      Morpheus provides several hyperparameters that can be adjusted for better control over MMM modeling. Below is a list of the key hyperparameters, their distributions, and their roles in the model.

      If you edit any hyperparameter settings, you can revert to the default model configuration at any time.

      1. Intercept

      • Distribution: Normal
      • Key values: Mean (μ = 0.2), Standard Deviation (σ = 2)
      • Function: Represents the baseline level of the dependent variable (e.g., sales) in the absence of any marketing input.

      2. Saturation Beta

      • Distribution: HalfNormal
      • Function: Controls how media saturation is modeled, ensuring that higher spends exhibit diminishing returns.

      3. Saturation Lambda

      • Distribution: Gamma
      • Key values: Shape (α = 3), Rate (β = 1)
      • Function: Defines the saturation effect in response to marketing spend.

      4. Gamma Control

      • Distribution: Laplace
      • Key values: Mean (μ = 2), Scale (b = 0.2)
      • Function: Regularizes control variables, such as seasonality or external factors.

      5. Gamma Fourier

      • Distribution: Normal
      • Key values: Mean (μ = 0), Standard Deviation (σ = 0.3)
      • Function: Manages Fourier series components for capturing periodic trends in data.

      6. Likelihood

      • Distribution: Normal
      • Function: Defines the probability distribution used for model fitting and validation.

      7. Adstock Alpha

      • Distribution: Beta
      • Key values: Shape (α = 2), Shape (β = 2)
      • Function: Controls the adstock effect, which determines how past marketing spend carries over into future periods.

      8. Peak Effect Delay

      • Distribution: Beta
      • Key values: Shape (α = 1), Shape (β = 4)
      • Function: Defines the lag between ad exposure and its maximum impact.

      9. Coef Trend

      • Distribution: Normal
      • Key values: Mean (μ = 0), Standard Deviation (σ = 1.5)
      • Function: Captures long-term trends independent of marketing efforts.

      Do You Need to Adjust These Hyperparameters?

      If you are new to MMM or Morpheus, you do not need to manually set hyperparameters. Morpheus will estimate these values using Bayesian techniques, ensuring the model adapts to your data. However, if you have prior knowledge or historical insights about marketing effects, fine-tuning hyperparameters can improve model accuracy.

      When to Adjust Hyperparameters?

      • If your model is overfitting or underfitting – Adjusting distributions can help balance bias and variance.
      • If you have prior knowledge of media behavior – Use empirical data to define realistic values.
      • If certain variables exhibit extreme volatility – Regularization through hyperparameters can improve stability.

      Conclusion

      Hyperparameters in Morpheus offer advanced users a way to refine their Marketing Mix Modeling approach. While completely optional, they help ensure a more accurate and interpretable model. Understanding each hyperparameter allows you to take full control of your MMM process, making your insights more actionable and precise.