numerax 0.3.0

No project description provided

numerax

Statistical and numerical computation functions for JAX, focusing on tools not available in the main JAX API.

📖 Documentation

Installation

pip install numerax

Features

Special Functions

Inverse functions for statistical distributions with differentiability support:

import jax.numpy as jnp
import numerax

# Gamma distribution quantiles
x = numerax.special.gammap_inverse(p, a)

# Chi-squared distribution quantiles  
x = numerax.special.chi2.ppf(q, df, loc=0, scale=1)

Key features:

• Halley's method for fast convergence
• Custom JVP implementation for exact gradients
• Numerical stability with adaptive precision
• Full JAX transformation compatibility

Profile Likelihood

Efficient profile likelihood computation for statistical inference with nuisance parameters:

import jax.numpy as jnp
import numerax

# Example: Normal distribution with mean inference, variance profiling
def normal_llh(params, data):
    mu, log_sigma = params
    sigma = jnp.exp(log_sigma)
    return jnp.sum(-0.5 * jnp.log(2 * jnp.pi) - log_sigma 
                   - 0.5 * ((data - mu) / sigma) ** 2)

# Profile over log_sigma, infer mu
is_nuisance = [False, True]  # mu=inference, log_sigma=nuisance

def get_initial_log_sigma(data):
    return jnp.array([jnp.log(jnp.std(data))])

profile_llh = numerax.stats.make_profile_llh(
    normal_llh, is_nuisance, get_initial_log_sigma
)

# Evaluate profile likelihood
data = jnp.array([1.2, 0.8, 1.5, 0.9, 1.1])
llh_val, opt_nuisance, diff, n_iter = profile_llh(jnp.array([1.0]), data)

Key features:

• JIT-compiled for performance
• L-BFGS optimization with convergence diagnostics
• Configurable tolerance and initial values
• Handles parameter masking automatically

Utilities

Development utilities for creating JAX functions with custom derivatives while ensuring proper documentation support. Includes decorators for preserving function metadata when using JAX's advanced features.

Requirements

• Python ≥ 3.12
• JAX
• jaxtyping
• optax

Acknowledgements

This work is supported by the Department of Energy AI4HEP program.