Review of Views and Copies#
The last reading on views and copies covered a lot, so here’s a brief summary of the key takeaways:
When an array is subset using simple indexing (i.e., by passing an index or range of indices denoted with a
:), the result is just a reference to the original array’s data. This is called a view.Because the view created through simple indexing is sharing data with the original array, changes to one will also impact the other.
While we often only refer to the newly created array as a “view,” the relationship between the original array and the view is symmetric, meaning changes to either may impact the other (if the change impacts an entry that is shared).
When you create a subset using fancy indexing or Boolean subsetting with a logical test, numpy will create a copy, not a view.
A view can be converted into a copy with
.copy().Creating a view is much faster than creating a copy; with that said, for most sizes of datasets you will encounter in life, both are exceedingly fast.