Building a Generative Design Model

This example demonstrates how to configure a Generative Design model, start the model training process, and monitor the build progress.

Before you begin

• Complete “Creating a GeomAI Project and Uploading Training Data” to create a project with training data.

• Ensure all training data in your project are ready (processed successfully).

Import necessary libraries

import ansys.simai.core as asc
from ansys.simai.core.data.geomai.models import GeomAIModelConfiguration

Configure your settings

Update these variables with your specific settings:

ORGANIZATION = "my_organization"  # Replace with your organization name
PROJECT_NAME = "new-bracket-project"  # Replace with your project name
NB_LATENT_PARAMS = 15  # Number of latent parameters (2-256)
BUILD_PRESET = "default"  # Options: "debug", "short", "default", "long"

The BUILD_PRESET options correspond to:

• "debug": Fast training for testing (very few epochs).

• "short": Quick training with reduced accuracy.

• "default": Balanced training time and quality.

• "long": Longer training for best quality.

Initialize the client and get the project

Connect to the instance:

simai_client = asc.SimAIClient(organization=ORGANIZATION)
geomai_client = simai_client.geomai

Retrieve the project by name:

project = geomai_client.projects.get(name=PROJECT_NAME)
print(f"Using project: {project.name}")

Verify project data are ready

Before building a model, ensure all training data are processed:

project_data = project.data()
ready_data = [data for data in project_data if data.is_ready]
print(f"Project has {len(ready_data)}/{len(project_data)} ready training data")

if len(ready_data) < len(project_data):
    print("Warning: Some training data are not ready. Model may fail to build.")

Configure the model

To define the configuration for your model, you can either specify the build preset or the number of epochs. To do so, instead of build_preset, you can specify the number of epochs directly. Example: nb_epochs=100.

The number of latent parameters defines the complexity of the model’s latent space; start with a small number (e.g., 10) and adjust based on your needs.

configuration = GeomAIModelConfiguration(
    build_preset=BUILD_PRESET,
    nb_latent_param=NB_LATENT_PARAMS,  # Must be between 2 and 256
)

Build the model

Start the model training process:

model = geomai_client.models.build(project, configuration)
print(f"Started build for model {model.id}")

Wait for model training to complete

Monitor the training process and handle completion:

print("Waiting for model training to complete (this may take a while)...")
while not model.wait(timeout=60):  # Check every 60 seconds
    print(f"Model {model.id} is still training...")

if model.has_failed:
    print(f"Model training failed: {model.failure_reason}")
else:
    print(f"Model {model.id} trained successfully and is ready to use.")

Next steps

Once your model is trained, you can:

• Generate random geometries: Generating Random Geometries

• Interpolate between geometries: Interpolating Between Geometries