GeomAIModels#

To build a GeomAI model, use GeomAIProject.build_model.

Directory#

class GeomAIModelDirectory#

Provides a collection of methods related to building models.

build(project: GeomAIProject | str, configuration: dict | GeomAIModelConfiguration) → GeomAIModel#

Launches a GeomAI build with the given configuration.

Parameters:

Examples

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

simai_client = asc.from_config()
project = simai_client.geomai.projects.get("new_secret_project")
configuration = GeomAIModelConfiguration(build_preset="default")
model = simai_client.geomai.models.build(project, configuration)

Use a previous configuration for a new build in the same project:

a_project = simai.geomai.projects.get("project_A")
build_conf = a_project.last_model_configuration
new_model = simai.geomai.models.build(build_conf)

Use a previous configuration for a new build in another project:

a_project = simai.geomai.projects.get("project_A")
build_conf = a_project.last_model_configuration
b_project = simai.geomai.projects.get("project_B")
new_model = simai.geomai.models.build(build_conf)

Model#

class GeomAIModel#

GeomAI model representation.

reload() → None#

Refresh the object with its representation from the server.

wait(timeout: float | None = None) → bool#

Wait for all jobs concerning the object to either finish or fail.

Parameters:

timeout (float | None) – Maximum amount of time in seconds to wait. The default is None, it means that there is no maximum on the time to wait.

Returns:

True if the computation has finished, False if the operation timed out.

Return type:

bool

property configuration: GeomAIModelConfiguration#

Build configuration of a model.

property failure_reason#

Optional message giving the causes for why the creation of the object failed.

See also

property fields: dict#

Dictionary containing the raw object representation.

property has_failed#

Boolean indicating if the creation of the object failed.

See also

property id: str#

ID of the object on the server.

property is_pending#

Boolean indicating if the object is still in creation. The value becomes False once object creation is either successful or has failed.

See also

property is_ready#

Boolean indicating if the object has finished creating without error.

See also

property project_id: str#

The ID of the GeomAI project where the model exists.

GeomAIModelConfiguration#

pydantic model GeomAIModelConfiguration#

Raises InvalidArguments if the input data cannot be validated to from a valid model.

field build_preset: Literal['debug', 'short', 'default', 'long'] | None = None#

The preset to use for the model training duration. One of debug, short, default, long.

  • debug: 4 minutes + 15 sec per geometry.

  • short: 45 minutes + 15 sec per geometry.

  • default: 3 hours + 15 sec per geometry.

  • long: 15 hours + 15 sec per geometry.

Mutually exclusive with nb_epochs.

field nb_epochs: int | None = None#

The number of times each training data is seen by the model during the training, between 1 and 1000.

Mutually exclusive with build_preset.

field nb_latent_param: int = 512#

An optional integer that defines the dimension of the latent space between 2 and 1024.

Warning

The default value of 512 is highly recommended if you are a non-expert user.

If you are an expert user, ready to experiment and assess correctly your models performance for different latent parameter value, and your use case has specific constraints, you can change this parameter, keeping the following in mind:

  • If the number is too low, geometries might become too coarse, and the reconstruction performance might degrade.

  • If the number is too high, the model will not be able to generate new geometries correctly.

The default value of 512 is used if nb_latent_param is not set.