arid_countreg.RdFunction to create class object similar to sci-kit learn's object structure for inferential purposes. Given a data frame, the response, and certain specifications return a generalized regression model interface for count data (either using a poisson or a negative binomial distribution) with a fit, predict, and score functions as well as attributes obtained from the statistical analysis.
arid_countreg( X, y, alpha = 0.05, fit_intercept = TRUE, verbose = FALSE, model = "additive", family = "poisson" )
| X | (data_frame): the input data frame with the explanatory variables to fit the model. |
|---|---|
| y | (integer): an integer vector with the response to be fitted (only natural numbers). |
| alpha | (double): a double vector of length 1 indicating the significance level (default: 0.05) |
| fit_intercept | (logical): if the model should include the intercept (TRUE or FALSE). (default: FALSE) |
| verbose | (logical): if results should include a written explanation (TRUE or FALSE). (default: FALSE) |
| model | (character): type of model to be fitted, either "additive" or "interactive". (default: "additive") |
| family | (character): distributional family to be used in generalized linear model. (default: "poisson") |
a class object with three methods and statistical attributes
X <- as.data.frame(matrix(rnorm(40 * 3), 40, 3)) y <- sample(c(1:60), 40, replace = TRUE) arid_countreg(X,y,0.1)#> $Env #> <environment: 0x000000003bf984e8> #> #> $fit #> function (X, y) #> { #> model_df <- X #> model_df$response <- y #> if (model == "additive") { #> columns <- stringr::str_c(colnames(X), collapse = " + ") #> formula <- stringr::str_c(c("response", columns), collapse = " ~ ") #> } #> else { #> columns <- stringr::str_c(colnames(X), collapse = " * ") #> formula <- stringr::str_c(c("response", columns), collapse = " ~ ") #> } #> if (fit_intercept == FALSE) { #> formula <- formula <- stringr::str_c(c(formula, "1"), #> collapse = " - ") #> } #> count_model <- glm(formula, data = model_df, family = family) #> if (family == "poisson") { #> if (AER::dispersiontest(count_model)[[2]] < alpha) { #> assign("family_", "negative binomial", Env) #> count_model <- MASS::glm.nb(formula, data = model_df) #> if (verbose == TRUE) { #> print("The Poisson model has overdispersion and it is underestimating the\n variance of the model, hence the negative binomial model will be used") #> print(" ") #> } #> } #> } #> if (fit_intercept == TRUE) { #> initial_value = 2 #> } #> else { #> initial_value = 1 #> } #> if (verbose == TRUE) { #> for (i in seq(initial_value, nrow(broom::tidy(count_model)))) { #> if (broom::tidy(count_model)$p.value[i] < alpha) { #> print(" ") #> print(paste("The variable", broom::tidy(count_model)$term[i], #> "has a statistically\n significant association over the response")) #> } #> } #> } #> set_attributes(count_model) #> return(count_model) #> } #> <bytecode: 0x000000003bf88670> #> <environment: 0x000000003bf984e8> #> #> $predict_count #> function (model, new_X) #> { #> return(exp(predict(model, new_X))) #> } #> <bytecode: 0x000000003bf954f0> #> <environment: 0x000000003bf984e8> #> #> $score #> function (model) #> { #> return(tibble::tibble(In_Sample_Metric = c("AIC", "Deviance"), #> Value = c(model$aic, model$deviance))) #> } #> <bytecode: 0x000000003bf95170> #> <environment: 0x000000003bf984e8> #> #> $intercept_ #> [1] 29.87914 #> #> $coef_ #> [1] 0.9227020 0.9976305 0.9799423 #> #> $p_values_ #> [1] 0.4827375 0.9841780 0.8648655 #> #> $count_model_ #> #> Call: MASS::glm.nb(formula = formula, data = model_df, init.theta = 1.852692242, #> link = log) #> #> Coefficients: #> (Intercept) V1 V2 V3 #> 3.397161 -0.080449 -0.002372 -0.020262 #> #> Degrees of Freedom: 39 Total (i.e. Null); 36 Residual #> Null Deviance: 44.55 #> Residual Deviance: 44.03 AIC: 357.4 #> #> $alpha_ #> [1] 0.1 #> #> $fit_intercept_ #> [1] TRUE #> #> $type_ #> [1] "additive" #> #> $family_ #> [1] "negative binomial" #> #> attr(,"class") #> [1] "arid_countreg"