AI And Model-Zoo Agents

These agents wrap neural inversion, uncertainty, anomaly detection, and pre-trained model access. They may require optional AI dependencies and model checkpoints.

AIInversionAgent

AIInversionAgent runs end-to-end 1-D AI inversion. It can be instantiated directly or created from a model-zoo checkpoint when available.

from pycsamt.agents import AIInversionAgent

agent = AIInversionAgent.from_pretrained("mt1d-resnet-5layer-v1")
result = agent.execute({
    "path": "/data/WILLY_EDIs",
    "output_dir": "/out/willy_ai1d",
})

print(result.get("rms_global"))

Inv2DAgent

Inv2DAgent performs 2-D profile inversion using U-Net style models. Use it when lateral continuity along a profile is important and a compatible model or training setup is available.

result = Inv2DAgent().execute({
    "path": "/data/WILLY_profile",
    "output_dir": "/out/willy_inv2d",
})

Inv3DAgent

Inv3DAgent performs 3-D spatial AI inversion with graph-based models. Use it when inter-station geometry and spatial relationships are part of the inversion target.

result = Inv3DAgent().execute({
    "path": "/data/WILLY_grid",
    "output_dir": "/out/willy_inv3d",
})

EnsembleAgent

EnsembleAgent runs ensemble inversion workflows and summarizes uncertainty from multiple predictions. Use it when uncertainty bands or coverage metrics are required alongside a predicted resistivity model.

result = EnsembleAgent(n_estimators=5, epochs=50).execute({
    "path": "/data/WILLY_EDIs",
    "output_dir": "/out/willy_ensemble",
})

print(result.get("coverage"))

JointInversionAgent

JointInversionAgent runs multi-modal inversion, for example MT with TEM or CSAMT with supporting data. Use it when multiple geophysical modalities constrain the same subsurface target.

result = JointInversionAgent(modalities=["mt", "tem"]).execute({
    "path": "/data/WILLY_MT",
    "secondary_path": "/data/WILLY_TEM",
    "output_dir": "/out/willy_joint",
})

AnomalyDetectionAgent

AnomalyDetectionAgent flags anomalous data patterns. Use it for station-frequency anomaly screening, survey triage, or finding regions that need manual review before inversion.

anomalies = AnomalyDetectionAgent().execute({
    "path": "/data/WILLY_EDIs",
    "output_dir": "/out/willy_anomalies",
})

ModelZooAgent

ModelZooAgent lists available pre-trained models, downloads checkpoints, and runs predictions where supported.

from pycsamt.agents import ModelZooAgent

zoo = ModelZooAgent()

models = zoo.execute({"action": "list"})
print(models["models"])

checkpoint = zoo.execute({
    "action": "download",
    "model_name": "mt1d-resnet-5layer-v1",
})
print(checkpoint.get("checkpoint_path"))

AI Workflow Pattern

MTLoaderAgent -> DataQCAgent -> DenoisingAgent
-> AIInversionAgent or Inv2DAgent or Inv3DAgent
-> InterpretationAgent -> ReportAgent

Add EnsembleAgent when uncertainty is part of the objective. Add JointInversionAgent when a secondary modality is available.