Fix several typos (#2729)

fix typos in various files

Co-authored-by: Kaiwen Wu <37524685+kayween@users.noreply.github.com>

Author: benediktjohannes

docs/source/conf.py

@@ -51,7 +51,7 @@ def find_version(*file_paths):
 os.mkdir(examples_dest)
 
 # Include examples in documentation
-# This adds a lot of time to the doc buiod; to bypass use the environment variable SKIP_EXAMPLES=true
+# This adds a lot of time to the doc build; to bypass use the environment variable SKIP_EXAMPLES=true
 for root, dirs, files in os.walk(examples_source):
     for dr in dirs:
         os.mkdir(os.path.join(root.replace(examples_source, examples_dest), dr))

docs/source/keops_kernels.rst

@@ -12,7 +12,7 @@ These kernels are compatible with the GPyTorch KeOps integration.
 For more information, see the `KeOps tutorial`_.
 
 .. note::
-   Only some standard kernels have KeOps impementations.
+   Only some standard kernels have KeOps implementations.
    If there is a kernel you want that's missing, consider submitting a pull request!
 
 

docs/source/likelihoods.rst

@@ -22,7 +22,7 @@ One-Dimensional Likelihoods
 Likelihoods for GPs that are distributions of scalar functions.
 (I.e. for a specific :math:`\mathbf x` we expect that :math:`f(\mathbf x) \in \mathbb{R}`.)
 
-One-dimensional likelihoods should extend :obj:`gpytoch.likelihoods._OneDimensionalLikelihood` to
+One-dimensional likelihoods should extend :obj:`gpytorch.likelihoods._OneDimensionalLikelihood` to
 reduce the variance when computing approximate GP objective functions.
 (Variance reduction is accomplished by using 1D Gauss-Hermite quadrature rather than MC-integration).
 

docs/source/variational.rst

@@ -13,7 +13,7 @@ most GP approximations:
 
 - :obj:`VariationalDistribution`, which define the form of the approximate inducing value
   posterior :math:`q(\mathbf u)`.
-- :obj:`VarationalStrategies`, which define how to compute :math:`q(\mathbf f(\mathbf X))` from
+- :obj:`VariationalStrategies`, which define how to compute :math:`q(\mathbf f(\mathbf X))` from
   :math:`q(\mathbf u)`.
 - :obj:`~gpytorch.mlls._ApproximateMarginalLogLikelihood`, which defines the objective function
   to learn the approximate posterior (e.g. variational ELBO).

examples/04_Variational_and_Approximate_GPs/index.rst

@@ -12,7 +12,7 @@ All approximate models consist of the following 3 composible objects:
 
 - :obj:`VariationalDistribution`, which define the form of the approximate inducing value
   posterior :math:`q(\mathbf u)`.
-- :obj:`VarationalStrategies`, which define how to compute :math:`q(\mathbf f(\mathbf X))` from
+- :obj:`VariationalStrategies`, which define how to compute :math:`q(\mathbf f(\mathbf X))` from
   :math:`q(\mathbf u)`.
 - :obj:`~gpytorch.mlls._ApproximateMarginalLogLikelihood`, which defines the objective function
   to learn the approximate posterior (e.g. variational ELBO).

gpytorch/distributions/multitask_multivariate_normal.py

@@ -86,7 +86,7 @@ def event_shape(self):
     @classmethod
     def from_batch_mvn(cls, batch_mvn, task_dim=-1):
         """
-        Reinterprate a batch of multivariate normal distributions as an (independent) multitask multivariate normal
+        Reinterpret a batch of multivariate normal distributions as an (independent) multitask multivariate normal
         distribution.
 
         :param ~gpytorch.distributions.MultivariateNormal batch_mvn: The base MVN distribution.

gpytorch/distributions/multivariate_normal.py

@@ -83,7 +83,7 @@ def _repr_sizes(mean: Tensor, covariance_matrix: Tensor | LinearOperator) -> str
     @property
     def _unbroadcasted_scale_tril(self) -> Tensor:
         if self.islazy and self.__unbroadcasted_scale_tril is None:
-            # cache root decoposition
+            # cache root decomposition
             ust = to_dense(self.lazy_covariance_matrix.cholesky())
             self.__unbroadcasted_scale_tril = ust
         return self.__unbroadcasted_scale_tril