Source code for jlnn.utils.helpers

#!/usr/bin/env python3

"""
General utility functions for Logical Neural Networks.
"""

# Imports
import jax.numpy as jnp



def scalar_to_interval(x: jnp.ndarray) -> jnp.ndarray:   
    """
    Converts standard [0, 1] probability scalars into precise [L, U] intervals.
    
    This is used to ground the JLNN model with data from classical datasets 
    where truth values are known exactly (L = U = x).
    
    Args:
        x (jnp.ndarray): Tensor of scalar truth values.
        
    Returns:
        jnp.ndarray: Tensor of intervals with shape ``(*x.shape, 2)``.
    """
    return jnp.stack([x, x], axis=-1)




def is_precise(interval: jnp.ndarray, epsilon: float = 1e-5) -> bool:
    """
    Checks if a truth interval has collapsed into a single point (L ≈ U).
    
    Args:
        interval: Truth interval tensor [L, U].
        epsilon: Maximum allowed difference between bounds.
        
    Returns:
        True if the uncertainty is within epsilon, False otherwise.
    """
    return float(jnp.abs(interval[..., 0] - interval[..., 1])) < epsilon