Spatial Transformation Function

In this repository, there are two versions of the spatialtransform function. One in R and the other in Python. You can use either of them in their repsective folders. Below is documentation of for each function.

R

transform_to_ind(formula, trainData, trainLocs, testData, testLocs, MaternParam=NULL, smoothness = 0.5, M = 30, ncores)

Description

transform_to_ind is a function designed to decorrelate spatially dependent data, specifically in the continuous univariate case.

Arguments

• formula: An object of class "formula" describing the model to be decorrelated.
• trainData: An object of class data.frame containing the training data with the response variable provided.
• trainLocs: A matrix object containing the coordinates of the training data. The dimensions should be nx2.
• testData: An object of class data.frame containing the data to be predicted.
• testLocs: A matrix object containing the coordinates of the test data. The dimensions should be nx2.
• MaternParams: A vector of two parameters: range and nugget. Range represents how fast the correlation decays with distance and nugget represents the variability in one location. The default is NULL (rng, nug) where the range and nugget parameter are estimated automatically.
• smoothness: The smoothness parameter, which controls the smoothness of the function. The default is 1/2 which results in an exponential kernel.
• M: The number of neighbors to consider when creating a correlation matrix for each individual observation. The default is 30.
• ncores: The number of cores to parallelize the decorrelation process.

back_transform_to_spatial(preds, transformObj)

Description

back_transform_to_spatial is a function designed to back transform the predictions to their spatial state in the continous univariate case.

Arguments

• preds: A vector of predictions from the machine learning model
• transformObj: The object outputted IndData after running transform_to_ind

Look to the folder named R Function for guidance on how to implemenet these two functions.

Python

transform_to_ind(formula, trainData, trainLocs, testData, testLocs, range_param, nugget, smoothness = 0.5, M = 30, ncores)

Description

transform_to_ind is a function designed to decorrelate spatially dependent data, specifically in the continuous univariate case.

Arguments

• formula: An object of class "formula" describing the model to be decorrelated.
• trainData: An object of class data.frame containing the training data with the response variable provided.
• trainLocs: A matrix object containing the coordinates of the training data. The dimensions should be nx2.
• testData: An object of class data.frame containing the data to be predicted.
• testLocs: A matrix object containing the coordinates of the test data. The dimensions should be nx2.
• range_param: A value that represents how fast the correlation decays with distance
• nugget: A value that represents the variability in one location.
• smoothness: The smoothness parameter, which controls the smoothness of the function. The default is 1/2 which results in an exponential kernel.
• M: The number of neighbors to consider when creating a correlation matrix for each individual observation. The default is 30.
• ncores: The number of cores to parallelize the decorrelation process. At this time keep the ncores at 1 (We will be updating this shortly).

back_transform_to_spatial(preds, transformObj)

Description

back_transform_to_spatial is a function designed to back transform the predictions to their spatial state in the continous univariate case.

Arguments

• preds: A vector of predictions from the machine learning model
• transformObj: The object outputted IndData after running transform_to_ind

Look to folder named Python Function for guidance on how to implemenet these two functions.