Last year, we introduced LunchBoxML – a machine learning (ML) plugin for Grasshopper and Dynamo that uses the Accord framework. LunchBoxML introduces several generalized supervised and unsupervised learning tools to visual programming including regression analysis, neural networks, and mixture models.
We have published a few new examples to the Bitbucket repository to demonstrate the application of Gaussian Mixture and Neural Network components.
Neural Network – Space Type Prediction
The Neural Network example demonstrates the use of a simple supervised learning algorithm that uses training data about space attributes (area, length, and width) and their resulting classification (office, conference, etc). The user can test a new space and the neural network will predict how the category of space based on what it has learned from the training data. The training data in the example is very rudimentary for the purposes of demonstration and could to be greatly expanded for practical applications such as expanding the study to include location and adjacency.
Gaussian Mixture – Panel Grouping
The Gaussian Mixture model demonstrates the use of an unsupervised learning algorithm to group 3D panels based on geometry attributes for size and shape. The Grasshopper definition also allows the user to prioritize certain attributes to affect the grouping outcomes. While the application does not attempt to rationalize the shape, the use of the Gaussian Mixture algorithm can be used to unpack the complexity of the panel system to make decisions about its development. The study could be greatly expanded to include classification based on additional attributes for curvature analysis and panel edge angles.