From 1499b0bb59d951102de18076abd8b2672adb82ae Mon Sep 17 00:00:00 2001
From: Dominik <dotto@fredhutch.org>
Date: Wed, 16 Oct 2024 17:08:17 -0700
Subject: [PATCH] revert 84355740992c28a3d7344a1e91323c738fa7e14e

---
 mellon/conditional.py |  8 +++-----
 mellon/util.py        | 27 ---------------------------
 2 files changed, 3 insertions(+), 32 deletions(-)

diff --git a/mellon/conditional.py b/mellon/conditional.py
index 74ff2bd..1cef8ac 100644
--- a/mellon/conditional.py
+++ b/mellon/conditional.py
@@ -5,7 +5,7 @@
 from jax.numpy import diag as diagonal
 from jax.numpy.linalg import cholesky
 from jax.scipy.linalg import solve_triangular
-from .util import ensure_2d, stabilize, DEFAULT_JITTER, add_variance, add_projected_variance
+from .util import ensure_2d, stabilize, DEFAULT_JITTER, add_variance
 from .base_predictor import Predictor, ExpPredictor, PredictorTime
 from .decomposition import DEFAULT_SIGMA
 
@@ -278,9 +278,9 @@ def __init__(
             LLB = stabilize(LLB, jitter)
         else:
             logger.debug("Assuming y is not the mean of the GP.")
-            y_cov_factor = _sigma_to_y_cov_factor(sigma, y_cov_factor, x.shape[0])
+            y_cov_factor = _sigma_to_y_cov_factor(sigma, y_cov_factor, xu.shape[0])
             sigma = None
-            LLB = add_projected_variance(LLB, A, y_cov_factor, jitter=jitter)
+            LLB = add_variance(LLB, y_cov_factor, jitter=jitter)
 
         L_B = cholesky(LLB)
         r = y - mu
@@ -304,8 +304,6 @@ def __init__(
         self.L = L
         self._state_variables.add("L")
 
-        y_cov_factor = _sigma_to_y_cov_factor(sigma, y_cov_factor, xu.shape[0])
-
         C = solve_triangular(L_B, dot(A, y_cov_factor), lower=True)
         Z = solve_triangular(L_B.T, C)
         W = solve_triangular(L.T, Z)
diff --git a/mellon/util.py b/mellon/util.py
index f6ec0a4..b9b621d 100644
--- a/mellon/util.py
+++ b/mellon/util.py
@@ -315,33 +315,6 @@ def add_variance(K, M=None, jitter=DEFAULT_JITTER):
     return K
 
 
-def add_projected_variance(K, A, y_cov_factor, jitter=DEFAULT_JITTER):
-    """
-    Adds the projected observation noise covariance to K and stabilizes it.
-
-    Parameters
-    ----------
-    K : array_like, shape (n_landmarks, n_landmarks)
-        The initial covariance matrix.
-    A : array_like, shape (n_landmarks, n_obs)
-        The projection matrix from observations to inducing points.
-    y_cov_factor : array_like, shape (n_obs, n_obs)
-        The observation noise covariance matrix.
-    jitter : float, optional
-        A small number to stabilize the covariance matrix. Defaults to 1e-6.
-
-    Returns
-    -------
-    stabilized_K : array_like, shape (n_landmarks, n_landmarks)
-        The stabilized covariance matrix with added projected variance.
-    """
-    noise = A @ y_cov_factor @ A.T
-    noise_diag = diagonal(noise)
-    diff = where(noise_diag < jitter, jitter - noise_diag, 0)
-    K = K + noise + diagonal(diff)
-    return K
-
-
 def mle(nn_distances, d):
     R"""
     Nearest Neighbor distribution maximum likelihood estimate for log density