pelt documentation

pelt.predict(signal, penalty, segment_cost_function='l1', jump=10, minimum_segment_length=2)

Calculate the changepoints.

Parameters:

• signal (numpy.typing.NDArray[numpy.float64]) – 1D or 2D input signal array. Can only contain numbers. None values are not accepted.

• penalty (float) – Penalty value for each changepoint added. Larger values result in fewer changepoints detected.

• segment_cost_function (str, optional) –

  Determines how the cost of each potential segment is calculated. Must be one of:

  • "l1" - L1 cost function (least absolute deviation)

  • "l2" - L2 cost function (least squared deviation)

  Defaults to "l1".

• jump (int, optional) – Step size between candidate changepoint positions. Must be > 0. If jump = 1, a check is done every possible prior change point, guaranteeing an exact solution, finding the true minimum of the objective function. If jump > 1, previous change points are considered at intervals of jump. This speeds up the computation, but the solution becomes approximate. Defaults to 10.

• minimum_segment_length (int, optional) – Minimum number of allowable number of data points within a segment. Must be positive. Defaults to 2.

Returns:

1D array of zero-based indices where changes in the signal were detected.

Return type:

numpy.typing.NDArray[numpy.uint64]

Raises:

ValueError – If the array has invalid dimensions or if any of the parameters are outside their valid ranges.

Examples

>>> from pelt import predict
>>> changepoints = predict(signal, penalty=20.0, segment_cost_function="l1", jump=10, minimum_segment_length=2)
>>> print(changepoints)