hypar-io/ml

Collecting ideas and building components for ML in AEC

Hypar ML utilities

Tools for training and deploying models on Hypar.

Glossary

• Function - A Hypar function. Locally, this is a directory containing a hypar.json file and a function's code. On Hypar's web application, functions are added to workflows.
• Workflow - A set of functions that execute on Hypar's web application.
• Input file - A JSON file containing one or more sets of input arguments for a function.
• Input model - A JSON Hypar model, usually exported from Hypar, that contains input data required by a function being executed locally.

Getting Started

• Follow the Things You Need section in the Getting Started guide to install the prerequisites to build and run a Hypar function.
• Install the Hypar command line interface (CLI). In Installing and Using the Command Line Interface you may need to install an alpha release of the CLI as new features are added. To do so, you can do:

  dotnet tool install -g hypar.cli --version [VERSION]

  Where:
  • [VERSION] is the latest alpha version available here.
• Create a workflow using Hypar's web application that will generate an input model that provides the data required by your function.
  • For example, if you are using Hypar's Facade function to generate a facade, you'll need a workflow that generates an envelope and some levels. You will not need to add the facade function to that workflow. The facade function will be run locally.
• Export the workflow from Hypar using More -> Export -> glTF.
• Follow the instructions in https://github.com/hypar-io/ML/tree/master/input_generation to generate an input file for the function and execute the function using the input file.
• Follow the instructions in https://github.com/hypar-io/ML/tree/master/image_generation to generate images from the resulting models.

ml's Issues

Support input_schema in input_generation script

The input_generation tool only supports Hypar functions that use the legacy "input" field in their hypar.json. It should be updated to support the new JSONSchema-based "input_schema" field.