The Geoclusterizer is a clustering routine that uses publicly-available data to group like geographical areas. We reduce the dimensionality of the dataset using the LinearCorex method and then fitting a Gaussian Mixture Model on that data to generate clusters of Census tracts.
Marketing professionals often use customer segmentation to serve more differentiated products and services to their customers. Traditionally, marketing analysts have used a set of heuristics and Excel-based analytical methods to segment customers. More recently, unsupervised learning techniques have been employed to segment / cluster customers at scale.
Software dependencies: Install Anaconda and create a virtual environment using the environment.yml file.
Reproduce the results:
- Geoclusterizer uses the pydoit dependency management framework to run tasks; however, the user can also run each task as a standalone py file
- The software is organized into the following tasks, which are listed in the
dodo.pyfile- makedirs: Make directories if they don't exist
- download_acs: Download American Community Survey (ACS) data; script based on this gist
- parse_acs: Parse downloaded ACS data into standalone tables
- scale_and_impute_data: Scale dataset and impute missing data
- select_n_components: Select number of components to use
- train_model: Train Gaussian Mixture model on scaled, imputed data using selected number of components
- To run all the tasks at once, simply cd into the repository and, if all packages have been installed correctly, type
doit - Outputs are saved in the data/processed directory