Labels¶

Warning

These functions look at future prices by design – they build supervised-learning targets (y), never features (X). Route their output through a purged / embargoed split (fynance.data.split) before training; never feed a label back into a model as an input.

The AFML (Lopez de Prado, Advances in Financial Machine Learning) labeling stack: path-dependent triple-barrier labels (triple_barrier), meta-labels for a secondary bet-sizing model (meta_labels), and overlap-aware sample weights (label_concurrency, uniqueness_weights).

Triple-barrier labeling¶

`triple_barrier`(prices[, events, pt, sl, ...])	Path-dependent triple-barrier labels (AFML ch.
`meta_labels`(side, labels)	Meta-label a primary side prediction against realized outcomes.

Overlap-aware sample weights¶

`label_concurrency`(t_in, t_out, T)	Number of labels alive at each bar (inclusive `[t_in, t_out]` spans).
`uniqueness_weights`(t_in, t_out, T)	Average-uniqueness sample weights for overlapping labels (AFML ch.