performance: Evaluate Interpolation Performance Across Multiple Time...

In castels/interpTools: Tools for the Systematic Testing of Interpolation Algorithms

Evaluate Interpolation Performance Across Multiple Time Series

Description

Function to calculate the performance metrics between lists of original and interpolated series.

See additional documentation provided in this package ("~/metric_definitions.pdf/") for a full descriptions of the the performance criteria.
Resulting object is of class 'pf'.

Usage

performance(OriginalData, IntData, GappyData, custom = NULL)

Arguments

OriginalData	list; A list object of dimension D x N of original (complete) time series
IntData	list; A list object of dimension D x M x P x G x K x N of interpolated time series (output of parInterpolate())
GappyData	list; A list object of dimension D x P x G x K x N of the gappy original time series (output of simulateGaps())
custom	character; A vector of names of user-defined functions used to calculate custom performance metrics (see details)

Details

The following is a description of the list of base performance metrics included in the returned object:

ID	Criterion	Optimal
......	...........	.........
1	pearson_r	max
2	r_squared	max
3	AD	min
4	MBE	min
5	ME	min
6	MAE	min
7	MRE	min
8	MARE	min
9	MAPE	min
10	SSE	min
11	MSE	min
12	RMS	min
13	NMSE	min
14	RE	max
15	RMSE	min
16	NRMSD	min
17	RMSS	min
18	MdAPE	min

Users can define and pass-in their own custom performance metric functions, but must adhere to the following rules:

• Inputs are limited to *ONLY* x (lowercase; the original time series) and X (uppercase; the interpolated time series)
  

• Output must be a single numeric value
  

Examples

 # User-defined functions to calculate a custom performance metric (see Details for rules)
 
 my_metric1 <- function(x,X){
 
  # Sum of original + interpolated values
  
  val <- x + X
  
  return(val) # return value must be a single numeric element
  
  }
  
 my_metric2 <- function(x,X){
 
  # Sum of index positions of interpolated values
 
  val <- sum(which(x != X))
 
  return(val) # return value must be a single numeric element
 
 } 
 
 # Implementing in eval_performance()
 
 performance(OriginalData = OriginalData, IntData = IntData, GappyData = GappyData, custom = c("my_metric1", "my_metric2"))

castels/interpTools documentation built on June 7, 2024, 4:20 p.m.