@@ -59,10 +59,10 @@ Security & maintainability limitations
5959pickle (and joblib by extension), has some issues regarding maintainability
6060and security. Because of this,
6161
62- * Never unpickle untrusted data as it could lead to malicious code being
62+ * Never unpickle untrusted data as it could lead to malicious code being
6363 executed upon loading.
64- * While models saved using one version of scikit-learn might load in
65- other versions, this is entirely unsupported and inadvisable. It should
64+ * While models saved using one version of scikit-learn might load in
65+ other versions, this is entirely unsupported and inadvisable. It should
6666 also be kept in mind that operations performed on such data could give
6767 different and unexpected results.
6868
@@ -77,12 +77,11 @@ additional metadata should be saved along the pickled model:
7777This should make it possible to check that the cross-validation score is in the
7878same range as before.
7979
80- Since a model internal representation may be different on two different
81- architectures, dumping a model on one architecture and loading it on
82- another architecture is not a supported behaviour, even if it might work
83- on some cases.
84- To overcome the issue of portability, pickle models are often deployed in
85- production using containers, like docker.
80+ Aside for a few exceptions, pickled models should be portable across
81+ architectures assuming the same versions of dependencies and Python are used.
82+ If you encounter an estimator that is not portable please open an issue on
83+ GitHub. Pickled models are often deployed in production using containers, like
84+ Docker, in order to freeze the environment and dependencies.
8685
8786If you want to know more about these issues and explore other possible
8887serialization methods, please refer to this
@@ -108,7 +107,7 @@ models between different machine learning frameworks, and to improve their
108107portability on different computing architectures. More details are available
109108from the `ONNX tutorial <https://onnx.ai/get-started.html >`_.
110109To convert scikit-learn model to ONNX a specific tool `sklearn-onnx
111- <http://onnx.ai/sklearn-onnx/> `_ has been developed.
110+ <http://onnx.ai/sklearn-onnx/> `_ has been developed.
112111
113112PMML is an implementation of the `XML
114113<https://en.wikipedia.org/wiki/XML> `_ document standard
0 commit comments