Xarray Integration

Integration utilities for Xarray data structures.

This module provides tools to bridge JAX tensors with xarray, allowing symbolic labeling of neural outputs and trained logical weights.

jlnn.utils.xarray_utils.extract_weights_to_xarray(weights: Array, input_labels: List[str], gate_name: str) → DataArray

Extracts trained logic weights into a labeled xarray DataArray.

In LNN, weights w >= 1.0 represent the importance of an antecedent. This function creates a structured view of these weights, mapping each value to its symbolic predicate name.

Parameters:

• weights – JAX array of weights from a logical gate.

• input_labels – Symbolic names of the inputs (e.g., [‘A’, ‘B’]).

• gate_name – Name of the gate these weights belong to.

Returns:

Labeled weights with ‘input’ dimension.

Return type:

xr.DataArray

jlnn.utils.xarray_utils.model_to_xarray(gate_outputs: Dict[str, Array], sample_labels: List[str]) → Dataset

Converts logical model outputs into a labeled xarray Dataset.

By mapping raw (batch, 2) tensors to a Dataset with ‘sample’ and ‘bound’ dimensions, we enable powerful scientific indexing and visualization of truth intervals.

Parameters:

• gate_outputs – Mapping of gate names to their [L, U] output tensors.

• sample_labels – Names for the samples in the batch (e.g., individual IDs).

Returns:

Multi-dimensional dataset containing truth intervals.

Return type:

xr.Dataset

This module allows exporting model outputs to the xarray.Dataset format. This brings advantages such as: * Labelled dimensions: Access to data via predicate names instead of numerical indices. * Advanced indexing: Easy filtering of samples that exhibit high uncertainty. * Serialization: Easy saving of experimental results in NetCDF format.