|
WeightedLowess
A C++ library for LOWESS with various weighting schemes
|
Options for compute().
More...
#include <Options.hpp>
Public Attributes | |
| Data_ | span = 0.3 |
| bool | span_as_proportion = true |
| Data_ | minimum_width = 0 |
| size_t | anchors = 200 |
| int | iterations = 3 |
| Data_ | delta = -1 |
| Data_ * | weights = NULL |
| bool | frequency_weights = true |
| int | num_threads = 1 |
| Data_ | Floating-point type for the data. |
| size_t WeightedLowess::Options< Data_ >::anchors = 200 |
The number of points that can be used as "anchors". LOWESS smoothing is performed exactly for each anchor, while the fitted values for all intervening points are computed by linear interpolation. A higher number of anchor points improves accuracy at the cost of computational work.
Note that this number is only used as a guideline by our LOWESS implementation. The actual number of selected anchors depends on the distribution of x-values; in addition, the first and last points are always used as the anchors,. If the specified number of anchors is greater than the number of points, LOWESS smoothing is performed directly for each point.
This setting is ignored if Options::delta is non-negative.
| Data_ WeightedLowess::Options< Data_ >::delta = -1 |
Delta value used to identify anchors. Seeds are identified greedily, by walking through the ordered x-coordinate values and marking a point y as a anchor if there are no anchors in [y - delta, y]. If set to zero, all unique points are used as anchors. If set to a negative value, an appropriate delta is determined from the number of points specified in set_points(). Otherwise, the chosen delta should have similar magnitude to the range of the x-values.
| bool WeightedLowess::Options< Data_ >::frequency_weights = true |
Whether the weights (if provided in Options::weights) should be interpreted as frequency weights. This means that they will be involved in both the span calculations for the smoothing window around each point, as well as in the LOWESS calculations themselves. If false, the weights will only be used for the latter.
| int WeightedLowess::Options< Data_ >::iterations = 3 |
The number of robustness iterations. At each iteration, each point is weighted according to its difference from the smoothed value, and the smoothing is repeated with these weights. More iterations increase robustness to outliers at the cost of computational work.
| Data_ WeightedLowess::Options< Data_ >::minimum_width = 0 |
Minimum width of the window centered around each point. This is useful for forcing the creation of a larger window in highly dense regions of the covariate range. Otherwise, overly small windows can lead to overfitting and a "bumpy" trend.
| int WeightedLowess::Options< Data_ >::num_threads = 1 |
Number of threads to use for various steps. This should be a positive integer. The parallelization scheme is determined by the #WEIGHTEDLOWESS_CUSTOM_PARALLEL macro.
| Data_ WeightedLowess::Options< Data_ >::span = 0.3 |
Span of the smoothing window around each point. By default, this is interpreted as a proportion of the total number of points used in compute() and should be between 0 and 1. Each window is defined as the smallest interval centered on the current point that covers the specified proportion.
If weights are provided to compute() and Options::frequency_weights = true, the span is instead defined from the proportion of the total weight across all points. This interprets the weights on each observation as relative frequencies.
See also Options::span_as_proportion, which changes the interpretation of this option.
| bool WeightedLowess::Options< Data_ >::span_as_proportion = true |
Whether the span should be interpreted as a proportion of the total number of points. If false, the value in Options::span is directly interpreted as the number of points that must fall inside the window. If false and weights are provided to compute() and Options::frequency_weights = true, the value in Options::span is interpreted as the sum of weights inside the window.
| Data_* WeightedLowess::Options< Data_ >::weights = NULL |
Pointer to an array of length equal to the number of points used in compute(). Each element should be a positive weight for the corresponding point in x and y. Alternatively, this may be NULL in which case all points are equally weighted.
1.9.8