Preparing Data and Training Custom HOOPS Embeddings

These notebooks demonstrate how to prepare a CAD dataset and train a custom HOOPS Embeddings model using the SIGNAL architecture.

Follow Install HOOPS AI for environment and license setup. Custom embedding training is optional: CAD Indexing and Retrieval Using HOOPS Embeddings can use the pre-trained model supplied with HOOPS AI.

Work through the notebooks in this order:

1. demo_HOOPS_EMBEDDINGS_DataPrep.ipynb gathers and encodes raw CAD files into a reusable .dataset store and an .infoset metadata file.

2. demo_HOOPS_EMBEDDINGS_training.ipynb loads the encoded data, creates dataset splits, trains the embedding model, and saves the model checkpoint.

Data preparation only needs to run once for a given dataset. Reuse its outputs when repeating training with different settings, then use the saved model in the indexing and retrieval workflow.

The code and resources for this tutorial can be found in the HOOPS-AI-Tutorials GitHub repository.

Info

From the repository root, run jupyter lab embeddings_pipeline/demo_HOOPS_EMBEDDINGS_DataPrep.ipynb to prepare the data before opening the training notebook.

• Preparing CAD Data for HOOPS Embeddings Training
• Environment setup
• Prerequisite: encoded dataset
• Step 2: Split the dataset
• Step 3: Train and save the model (SIGNAL architecture)
• Next steps