pycsamt.agents.pinn_agent#

pycsamt.agents.pinn_agent#

PINNInversionAgent wraps PINNInverter1D, PINNInverter2D, and PINNInverter3D.

Physics-informed optimisation requires no labelled training data. Model parameters are updated by gradient descent on the Wait (1954) MT forward- physics loss.

Classes

PINNInversionAgent(*[, dim, n_layers, ...])

PINN-based MT inversion without labelled data.
class pycsamt.agents.pinn_agent.PINNInversionAgent(*, dim=1, n_layers=10, depth_max=2000.0, smoothness_weight=0.01, lateral_weight=0.005, graph_weight=0.005, radius=5000.0, epochs=None, lr=0.01, solver='mt1d', comp='xy', api_key=None, model=None, llm_provider='claude')[source]#

Bases: BaseAgent

PINN-based MT inversion without labelled data.

Optimises a layered Earth by minimising a physics-informed loss via Adam gradient descent. Supports 1-D per-station, joint 2-D profile, and quasi-3-D graph-coupled inversion.

Parameters:

  • dim (overrides) – Dimensionality. Default 1.

  • n_layers (overrides) – Number of layers including the halfspace. Default 10.

  • depth_max (float) – Maximum investigation depth in metres. Default 2000.0.

  • smoothness_weight (float) – Vertical regularisation weight. Default 0.01.

  • lateral_weight (float) – Lateral smoothness weight (2-D only). Default 0.005.

  • graph_weight (float) – Graph-Laplacian spatial weight (3-D only). Default 0.005.

  • radius (float) – Edge radius in metres for the 3-D graph. Default 5000.0.

  • epochs (overrides) – Adam iterations. None uses 500 for 1-D and 300 for 2-D / 3-D.

  • lr (float) – Adam learning rate. Default 1e-2.

  • solver ({"mt1d", "csamt1d"}) – Physics solver. Default "mt1d".

  • comp (str) – Impedance component (1-D only). Default "xy".

  • api_key (str | None) – LLM configuration (optional).

  • model (str | None) – LLM configuration (optional).

  • llm_provider (str) – LLM configuration (optional).

  • keys (Output data)

  • ----------

  • data (sites / path observed)

  • dir (output_dir optional figure/save)

  • dim

  • epochs

  • n_layers

  • keys

  • ----------------

  • object (inverter fitted inverter)

  • (n_layers (section ndarray) – log10-rho section matrix

  • n_stations) – log10-rho section matrix

  • (1-D) (models list of LayeredModel)

  • int (n_stations)

  • {station (rms_per_station dict)

  • float (rms_global)

  • None (residuals_df pandas.DataFrame or)

  • None

  • dict (figure_paths)

  • dict

Examples

>>> agent = PINNInversionAgent(
...     dim=1, n_layers=10, epochs=200
... )
>>> res = agent.execute(
...     {"path": "/data/L22PLT"}
... )
>>> res["rms_global"]
0.18
SYSTEM_PROMPT: str = 'You are an expert in physics-informed neural-network\ninversion for MT/CSAMT geophysics.\nGiven a PINN inversion result write 4-5 sentences:\n1. State dimensionality (1-D/2-D/3-D) and convergence.\n2. Report final RMS (log10 rho-ohm-m) and data fit.\n3. Describe the resistivity structure recovered.\n4. Flag stations or regions with high residuals.\n5. Recommend adjustments to epochs, regularisation,\n   or whether to switch to a hybrid approach.\nReply in plain scientific English.\n'#

Override in subclasses to give the LLM its domain expertise.

execute(input_data)[source]#

Run this agent on input_data and return an AgentResult.

Subclasses must implement this method. The contract:

  • Reset self._last_cost = 0.0 at the top.

  • Record wall-clock time with t0 = time.time().

  • Return AgentResult(elapsed_seconds=time.time()-t0, cost_estimate_usd=self._last_cost, ...).

Parameters:

input_data (dict[str, Any])

Return type:

AgentResult