| Title: | Permutations Tests and Performance Indicator for Zero-Inflated Proportions Response |
|---|---|
| Description: | Permutations tests to identify factor correlated to zero-inflated proportions response. Provide a performance indicator based on Spearman correlation to quantify the part of correlation explained by the selected set of factors. See details for the method at the following preprint e.g.: <https://hal.archives-ouvertes.fr/hal-02936779v3>. |
| Authors: | Melina Ribaud |
| Maintainer: | Melina Ribaud <[email protected]> |
| License: | GPL-3 |
| Version: | 0.1.1 |
| Built: | 2025-02-12 06:01:59 UTC |
| Source: | https://github.com/cran/ZIprop |
Calculate the scalar delta.
This parameter comes from the optimal Spearman’s correlation
when the rank of two vectors X and proba are equal except on a given set of indices.
In our context, this set correspond to the zero-values of the vector proba.
delta(X, proba)delta(X, proba)
X |
a vector. |
proba |
a zero-inflated proportions response. |
Delta the scalar Delta calculated for the vector x
and the vector proba.
X = rnorm(100) proba = runif(100) proba[sample(1:100,80)]=0 Delta = delta(X,proba) print(Delta)X = rnorm(100) proba = runif(100) proba[sample(1:100,80)]=0 Delta = delta(X,proba) print(Delta)
Data for the comparison of COVID-19 mortality in European and North American geographic entities
data(diffFactors)data(diffFactors)
A data frame with 483 rows and 32 variables
geographic_entity_receptor are the entity receptor
geographic_entity_source are the entity source
proba is the probability that the receptor follows the mortality dynamics of the source
other columns are the difference between factors
Melina Ribaud, Davide Martinetti and Samuel Soubeyrand
Equine Influenza dataset
data(equineDiffFactors)data(equineDiffFactors)
A data frame with 2256 rows and 8 variables
ID.source are the ID of source hosts
ID.recep are the ID of receiver hosts
y are the vector of transmission probabilities source -> receiver
other columns are the factors
Melina Ribaud and Joseph Hughes
A dataset example to test the package functions. The factor X1 to X5 and F1 to F5 are correlated to the responses y.
data(example_data)data(example_data)
A data frame with 440 rows and 23 variables
ID.source are the ID of source hosts
ID.recep are the ID of receiver hosts
y are the vector of transmission probabilities source -> receiver
X1 to X10 are continuous factor
F1 to F10 are discrete factor
Turns a factor with several levels into a matrix with several columns composed of zeros and ones.
fact2mat(x)fact2mat(x)
x |
a vector. |
Columns with zeros and ones.
x = sample(1:3,100,replace = TRUE) fact2mat(x)x = sample(1:3,100,replace = TRUE) fact2mat(x)
Calculate the indicator for a vector X
and a zero-inflated proportions response proba.
indicator(X, proba)indicator(X, proba)
X |
a vector. |
proba |
a zero-inflated proportions response. |
a scalar represents the performance indicator
and the vector proba.
X = rnorm(100) proba = runif(100) proba[sample(1:100,80)]=0 print(indicator(X,proba))X = rnorm(100) proba = runif(100) proba[sample(1:100,80)]=0 print(indicator(X,proba))
Search for the set of parameters that maximize the indicator (equivalent to Spearman correlation). For a given set of factors scaled between 0 and 1 and a zero-inflated proportions response.
indicator_max( DT, ColNameFactor, ColNameWeight = "weight", bounds = c(-10, 10), max_generations = 200, hard_limit = TRUE, wait_generations = 50, other_class = NULL )indicator_max( DT, ColNameFactor, ColNameWeight = "weight", bounds = c(-10, 10), max_generations = 200, hard_limit = TRUE, wait_generations = 50, other_class = NULL )
DT |
a data table contains the factors and the response. |
ColNameFactor |
a char vector with the name of the selected factor. |
ColNameWeight |
a char with the name of the ZI response. |
bounds |
default is $[-10;10]$. Upper and Lower bounds. |
max_generations |
default is 200 see genoud for more information. |
hard_limit |
default is TRUE see genoud for more information. |
wait_generations |
default is 50 see genoud for more information. |
other_class |
a char vector with the name of other classes than numeric (factor or char). |
Return a list of two elements with the value of the indicator and the associate set of parameters (beta).
library(data.table) data(example_data) # For real cases increase max_generations and wait_generations I_max = indicator_max(example_data, names(example_data)[c(4:8, 14:18)], ColNameWeight = "proba", max_generations = 20, wait_generations = 5) print(I_max)library(data.table) data(example_data) # For real cases increase max_generations and wait_generations I_max = indicator_max(example_data, names(example_data)[c(4:8, 14:18)], ColNameWeight = "proba", max_generations = 20, wait_generations = 5) print(I_max)
Creates a design matrix by expanding factors to a set of dummy variables.
model_matrix(DT, ColNameFactor, other_class)model_matrix(DT, ColNameFactor, other_class)
DT |
a data table contains the factors and the response. |
ColNameFactor |
a char vector with the name of the selected factor. |
other_class |
a char vector with the name of other classes than numeric (factor or char). |
return the value.
library(data.table) data(example_data) m = model_matrix (example_data, colnames(example_data)[-c(1:3)], other_class = colnames(example_data)[14:23]) print(m)library(data.table) data(example_data) m = model_matrix (example_data, colnames(example_data)[-c(1:3)], other_class = colnames(example_data)[14:23]) print(m)
Permutations tests to identify factor correlated to a zero-inflated proportions response. The statistic are the Spearman's correlation for numeric factor and mean by level for other factor.
permDT( DT, ColNameFactor, B = 1000, nclust = 1, ColNameWeight = "weight", ColNameRecep = "ID.recep", ColNameSource = "ID.source", seed = NULL, no_const = FALSE, num_class = ColNameFactor, other_class = NULL, multiple_test = FALSE, adjust_method = "none", alpha = 0.05 )permDT( DT, ColNameFactor, B = 1000, nclust = 1, ColNameWeight = "weight", ColNameRecep = "ID.recep", ColNameSource = "ID.source", seed = NULL, no_const = FALSE, num_class = ColNameFactor, other_class = NULL, multiple_test = FALSE, adjust_method = "none", alpha = 0.05 )
DT |
a data table contains the factors and the response. |
ColNameFactor |
a char vector with the name of the selected factor. |
B |
number of permutations (use at least B=1000 permutations to get a correct accuracy of the p-value.) |
nclust |
number of proc for parallel computation. |
ColNameWeight |
a char with the name of the ZI response. |
ColNameRecep |
colname of the column with the target names |
ColNameSource |
colname of the column with the contributor names |
seed |
vector with the seed for the permutations: size( |
no_const |
FALSE for receiver block constraint for permutations: TRUE no constraint. |
num_class |
a char vector with the name of numeric factor. |
other_class |
a char vector with the name of other classes than numeric (factor or char). |
multiple_test |
useful option only for discrete factors: Set TRUE to calculate multiple tests. |
adjust_method |
p-values adjusted methods (default "none" ). c("holm", "hochberg", "hommel", "bonferroni", "BH", "BY","fdr", "none"). |
alpha |
significant level (default 0.05). |
A data frame with two columns. One for the statistics and the other one for the p-value.
library(data.table) data(example_data) res = permDT (example_data, colnames(example_data)[c(4,10,14,20)], B = 10, nclust = 1, ColNameWeight = "y", ColNameRecep = "ID.recep", ColNameSource = "ID.source", seed = NULL, num_class = colnames(example_data)[c(4,10)], other_class = colnames(example_data)[c(14,20)]) print(res)library(data.table) data(example_data) res = permDT (example_data, colnames(example_data)[c(4,10,14,20)], B = 10, nclust = 1, ColNameWeight = "y", ColNameRecep = "ID.recep", ColNameSource = "ID.source", seed = NULL, num_class = colnames(example_data)[c(4,10)], other_class = colnames(example_data)[c(14,20)]) print(res)
Scale a vector between 0 and 1.
scale_01(x)scale_01(x)
x |
a vector. |
the scaled vector of x.
x = runif(100,-10,10) x_scale = scale_01(x) range(x_scale)x = runif(100,-10,10) x_scale = scale_01(x) range(x_scale)
Statistic for non-numeric factor tests (same statistic as H-test).
T_stat_discr(permu, al)T_stat_discr(permu, al)
permu |
the response vector. |
al |
the factor. |
the statistic.
permu = runif(100,-10,10) al = as.factor(sample(1:3,100,replace=TRUE)) T_stat_discr(permu, al)permu = runif(100,-10,10) al = as.factor(sample(1:3,100,replace=TRUE)) T_stat_discr(permu, al)
Statistic for non-numeric factor multiple tests (difference in mean ranks).
T_stat_multi(permu, al)T_stat_multi(permu, al)
permu |
the response vector. |
al |
the factor. |
the means difference of two levels for a discrete factor.
permu = runif(100,-10,10) al = as.factor(sample(1:3,100,replace=TRUE)) T_stat_multi(permu, al)permu = runif(100,-10,10) al = as.factor(sample(1:3,100,replace=TRUE)) T_stat_multi(permu, al)
We propose a by block-permutation-based methodology (i) to identify factors (discrete or continuous) that are potentially significant, (ii) to define a performance indicator to quantify the percentage of correlation explained by the significant factors subset for Zero-Inflated Proportions data (ZIprop).
Melina Ribaud, Edith Gabriel, Joseph Hughes, Samuel Soubeyrand. Identifying potential significant factors impacting zero-inflated proportions data. 2020. hal-02936779