Position-based vs label-based indexing

[eivindtostesen]

The data set Y could be, for example, a time series of stock prices. The P = PeakTree(Y) object is a tree structure that describes features of Y.

Position-based vs label-based indexing is an issue in three cases:

1. Y has an integer index, but it also has labels, for example time stamps.
2. P has internal data as dicts, and each dict has both keys and integer positions (available as P._index).
3. Peaks are located by starts and stops that can be either integers (as in slices of Y) or labels (pairs of time stamps).

In previous versions

• The idea was that all indexing should be label-based, since integer positions would be tied to the Y array and not be robust to resampling of the data.
• Starts and stops should not be necessary to compute, but possible.
• P's dict keys was argmin labels, which approximately located the peaks.

In future versions, maybe

• Starts and stops are computed by default. Easier to understand than argmin positions.
• Integer slice notation will be used to locate peaks, by default. Much more lightweight than pairs of time stamps. But requires Y for reference.
• P dict keys can be changed freely
• Label-based indexing of Y should not be expected, but possible

[eivindtostesen]

There is a somewhat similar issue in pandas development.

Quote: "We propose that it should only work with positional indexing, and the translation of keys to positions should be entirely done at a higher level."