Skip to main content
Version: 1.1.0

Models

Models is a mapping for Machine Learning models that need to be imported together with the configuration to run inference sessions using ONNX and make predictions for the agents. The agent class and agent instance configurations will then reference these models by their alias to be invoked for each step.

Weights & Biases​

It is also possible to have the simulator download models from Weights & Biases. This prevents having to upload the models each time you would like to run a simulation. To do this you must publish the models as a Weights and Biases Artifact. Once you've done that you can specify the WanDB Artifact name instead.

When defining a wandb_artifact please don't define a location. This is a one or the other option. If defining both an error will be returned.

Format​

nametype
typearray of {
    alias: string,
    location: string,
}
requiredtrue

Example​

models:
  - alias: "bancor_lp_action"
    location: "models/action_user.onnx"
  - alias: "bancor_lp_volume"
    location: "models/amount_user.onnx"
  - alias: "bancor_trader"
    wandb_artifact: "almanak/uncategorized/amount_user.onnx:latest"