Fix GMM.condition() ValueError with single conditioning value

[jkitchin]

Summary

GMM.condition() raises ValueError: setting an array element with a sequence when conditioning on a single value. This affects all input formats ([[val]], [val], np.array(...), scalar).

Root Cause

The to_norm_factor_and_exponents() method returns exponents as an array of shape (n_samples,). When conditioning on a single value, this is a 1-element array, not a scalar. The original code attempted to assign this array directly to marginal_prior_exponents[k] (a scalar slot), causing the error.

Fix

Extract the scalar value with exponents[0] since condition() is designed to handle a single conditioning value:

# Before (broken):
marginal_norm_factors[k], marginal_prior_exponents[k] =     mvn.marginalize(indices).to_norm_factor_and_exponents(x)

# After (fixed):
norm_factor, exponents =     mvn.marginalize(indices).to_norm_factor_and_exponents(x)
marginal_norm_factors[k] = norm_factor
marginal_prior_exponents[k] = exponents[0]

Minimal Reproducer

import numpy as np
from gmr import GMM

np.random.seed(42)
data = np.random.randn(100, 2)

gmm = GMM(n_components=2, random_state=42)
gmm.from_samples(data)

# This fails with ValueError before the fix
conditional = gmm.condition([1], [[0.75]])

Testing

• Added test_conditional_distribution_input_formats() to verify various input formats work
• All 61 existing tests pass

[AlexanderFabisch]

Thanks for the contribution!