DeepCausalMMM

What DeepCausalMMM is

DeepCausalMMM is an open-source Python package for marketing mix modeling that combines deep learning and causal inference to estimate the effects of marketing channels on business KPIs. Its GRU-based temporal model is designed to represent effects that vary over time. The package models relationships between channels with directed acyclic graphs: it uses an upper-triangular mask by default and also offers optional NOTEARS learning.

It supports multi-region analysis and automatic seasonal decomposition for each region. Analytical tools include response curves for saturation analysis, constrained budget optimization and DMA-level contribution calculations. The documentation describes more than 14 interactive visualizations covering performance metrics, channel analysis, economic contributions and DAG networks. For input, media and control data are NumPy arrays organized by region, week and channel, alongside a region-by-week target array.

You can install the package from PyPI with pip or from its GitHub repository. Its documented runtime requirements are Python 3.9 or later, PyTorch 2.0 or later, and NumPy 1.21 or later but below 2.0. The package detects CUDA and uses it when available; the documentation recommends a GPU for large models. It is released under the MIT License, and the project directs users to GitHub issues for bug reports and feature requests. The documentation identifies marketing mix modeling, attribution analysis, budget optimization, causal discovery and multi-touch attribution as use cases.

Who it suits

DeepCausalMMM suits analysts and developers working on marketing mix modeling, attribution, causal discovery or budget optimization in Python. It is a fit for teams with regional and weekly channel data who can work with its documented Python, PyTorch and NumPy requirements.

Getting started

1. Install Python 3.9 or later, PyTorch 2.0 or later, and NumPy 1.21 or later but below 2.0.
2. Install DeepCausalMMM from PyPI with pip or from the project's GitHub repository.
3. Prepare media and control NumPy arrays shaped by region, week and channel, plus a region-by-week target array.
4. Use CUDA-enabled hardware if available; the package detects CUDA automatically.

Verdict

Choose DeepCausalMMM if you need an open-source Python package for causal marketing mix analysis across regions, with optimization and visualization tools included. It requires a compatible Python, PyTorch and NumPy environment, and large models may call for a GPU.