remove obsolete poisson log likelihood

src/evermore/loss.py

@@ -1,8 +1,5 @@
-from collections.abc import Callable
 from typing import cast
 
-import equinox as eqx
-import jax
 import jax.numpy as jnp
 from jaxtyping import Array, PyTree
 
@@ -13,7 +10,6 @@
 __all__ = [
     "get_log_probs",
     "get_boundary_constraints",
-    "PoissonLogLikelihood",
 ]
 
 
@@ -38,37 +34,3 @@ def _constraint(param: Parameter) -> Array:
 
 def get_boundary_constraints(module: PyTree) -> PyTree:
     return params_map(lambda p: p.boundary_constraint, module)
-
-
-class PoissonLogLikelihood(eqx.Module):
-    """
-    Poisson log-likelihood.
-
-    Usage:
-
-    .. code-block:: python
-
-        import evermore as evm
-
-        nll = evm.loss.PoissonLogLikelihood()
-
-        def loss(model, x, y):
-            expectation = model(x)
-            loss = nll(expectation, y)
-            constraints = evm.loss.get_log_probs(model)
-            loss += evm.util.sum_over_leaves(constraints)
-            return -jnp.sum(loss)
-    """
-
-    @property
-    def log_prob(self) -> Callable:
-        return jax.scipy.stats.poisson.logpmf
-
-    @jax.named_scope("evm.loss.PoissonLogLikelihood")
-    def __call__(self, expectation: Array, observation: Array) -> Array:
-        # poisson log-likelihood
-        return jnp.sum(
-            self.log_prob(observation, expectation)
-            - self.log_prob(observation, observation),
-            axis=-1,
-        )

tests/test_loss.py

@@ -6,11 +6,6 @@
 import evermore as evm
 
 
-def test_PoissonLogLikelihood():
-    f = evm.loss.PoissonLogLikelihood()
-    assert f(jnp.array([1.0]), jnp.array([1.0])) == 0.0
-
-
 def test_get_log_probs():
     params = {
         "a": evm.NormalParameter(value=0.5),