From f81eca75e67c88be3cdc5c30844f95be9a58d6e9 Mon Sep 17 00:00:00 2001
From: "jannik.luethje" <=>
Date: Thu, 11 Apr 2024 15:48:57 +0200
Subject: [PATCH] Release version 0.0.11
---
dep/mcpp | 2 +-
gaussian process/training/python/utils.py | 25 +++++++++++++++++++----
2 files changed, 22 insertions(+), 5 deletions(-)
diff --git a/dep/mcpp b/dep/mcpp
index 69a763e..fb61bbb 160000
--- a/dep/mcpp
+++ b/dep/mcpp
@@ -1 +1 @@
-Subproject commit 69a763e55faf6557af0839a06fa82eb919b062e4
+Subproject commit fb61bbb8733b23a0b48cb86f016429594ee062c5
diff --git a/gaussian process/training/python/utils.py b/gaussian process/training/python/utils.py
index 7831b30..2caa3ae 100644
--- a/gaussian process/training/python/utils.py
+++ b/gaussian process/training/python/utils.py
@@ -4,6 +4,7 @@ from pathlib import Path
import json
import torch
from gpytorch.means import ConstantMean, ZeroMean
+from gpytorch.likelihoods import GaussianLikelihood, FixedNoiseGaussianLikelihood
# Import MeLOn
import maingopy.melonpy as melonpy
# If you are not getting MeLOn through MAiNGO, use:
@@ -51,8 +52,16 @@ def generate_melon_gp_object(GP_model, GP_likelihood, X, y, matern, scaler):
noise = GP_likelihood.noise.detach().numpy()
cov_mat = GP_model.covar_module(X)
- gp_data.K = cov_mat.numpy() + noise * np.eye(N=gp_data.nX)
- gp_data.invK = np.linalg.inv(gp_data.K)
+
+ if isinstance(GP_likelihood, GaussianLikelihood):
+ K_numpy = cov_mat.numpy() + noise * np.eye(N=gp_data.nX)
+ elif isinstance(GP_likelihood, FixedNoiseGaussianLikelihood):
+ K_numpy = cov_mat.numpy() + np.diag(noise)
+ else:
+ raise Exception(f'Likelihood {type(GP_likelihood)} currently not supported.')
+
+ gp_data.K = K_numpy
+ gp_data.invK = np.linalg.inv(K_numpy)
gp_data.matern = matern
kernel_data = melonpy.KernelData()
@@ -106,8 +115,16 @@ def save_model_to_json(filepath, filename, GP_model, GP_likelihood, X, y, matern
noise = GP_likelihood.noise.detach().numpy()
cov_mat = GP_model.covar_module(X).numpy()
- prediction_parameters["K"] = (cov_mat + noise * np.eye(N=prediction_parameters["nX"])).tolist()
- prediction_parameters["invK"] = np.linalg.inv(cov_mat + noise * np.eye(N=prediction_parameters["nX"])).tolist()
+
+ if isinstance(GP_likelihood, GaussianLikelihood):
+ K_numpy = cov_mat.numpy() + noise * np.eye(N=prediction_parameters["nX"])
+ elif isinstance(GP_likelihood, FixedNoiseGaussianLikelihood):
+ K_numpy = cov_mat.numpy() + np.diag(noise)
+ else:
+ raise Exception(f'Likelihood {type(GP_likelihood)} currently not supported.')
+
+ prediction_parameters["K"] = K_numpy.tolist()
+ prediction_parameters["invK"] = np.linalg.inv(K_numpy).tolist()
if not 'input' in scalers or not isinstance(scalers['input'], MinMaxScaler):
raise Exception("There has to be an inputscaler which is a scikit-learn MinMaxScaler instance")
--
GitLab