inst/app-blorr/logic/logic_validation.R

conf_result <- eventReactive(input$submit_conf, {
  
  if (input$conf_use_prev) {
    k <- model()
  } else {
    data <- final_split$train
    k <- glm(input$conf_fmla, data = data, family = binomial(link = "logit"))
  }

  if (input$conf_use_test_data) {
    out <- blr_confusion_matrix(k, cutoff = input$conf_cutoff, data = final_split$test)
  } else {
    out <- blr_confusion_matrix(k, cutoff = input$conf_cutoff)
  }

  return(out)
  
})

confusion_title <- eventReactive(input$submit_conf, {
  column(12, align = 'center', h4('Confusion Matrix & Model Performance Measures'))
})

output$conf_title <- renderUI({
  confusion_title()
})


output$conf_out <- renderPrint({
  conf_result()
})


# hosmer lemeshow test
hoslem_result <- eventReactive(input$submit_hoslem, {

  if (input$hoslem_use_prev) {
    k <- model()
  } else {
    data <- final_split$train
    k <- glm(input$hoslem_fmla, data = data, family = binomial(link = "logit"))
  }

  if (input$hoslem_use_test_data) {
    out <- blr_test_hosmer_lemeshow(k, data = final_split$test)
  } else {
    out <- blr_test_hosmer_lemeshow(k)
  }

  return(out)

})

output$hoslem_out <- renderPrint({
  hoslem_result()
})

# gains chart and roc curve
lift_result <- eventReactive(input$submit_lift, {

  if (input$lift_use_prev) {
    k <- model()
  } else {
    data <- final_split$train
    k <- glm(input$lift_fmla, data = data, family = binomial(link = "logit"))
  }

  if (input$lift_use_test_data) {
    out <- blr_gains_table(k, data = final_split$test)
  } else {
    out <- blr_gains_table(k)
  }

  return(out)

})

gains_title <- eventReactive(input$submit_lift, {
  column(12, align = 'center', h4('Gains Table & Lift Chart'))
})

output$lift_title <- renderUI({
  gains_title()
})

output$gains_table_out <- renderPrint({
  lift_result()
})

output$lift_out <- renderPlot({
	plot(lift_result())
})

# ROC curve
roc_result <- eventReactive(input$submit_roc, {

  if (input$roc_use_prev) {
    k <- model()
  } else {
    data <- final_split$train
    k <- glm(input$roc_fmla, data = data, family = binomial(link = "logit"))
  }

  if (input$roc_use_test_data) {
    out <- blr_gains_table(k, data = final_split$test)
  } else {
    out <- blr_gains_table(k)
  }

  return(out)

})


output$roc_out <- renderPlot({
	blr_roc_curve(roc_result())
})

# KS chart
ks_result <- eventReactive(input$submit_ks, {

  if (input$ks_use_prev) {
    k <- model()
  } else {
    data <- final_split$train
    k <- glm(input$ks_fmla, data = data, family = binomial(link = "logit"))
  }

  if (input$ks_use_test_data) {
    out <- blr_gains_table(k, data = final_split$test)
  } else {
    out <- blr_gains_table(k)
  }

  return(out)

})


output$ks_out <- renderPlot({
	blr_ks_chart(ks_result())
})


# lorenz curve
lorenz_result <- eventReactive(input$submit_lorenz, {

  if (input$lorenz_use_prev) {
    k <- model()
  } else {
    data <- final_split$train
    k <- glm(input$lorenz_fmla, data = data, family = binomial(link = "logit"))
  }

  return(k)

})

output$lorenz_out <- renderPlot({
	blorr::blr_lorenz_curve(lorenz_result())
})

Try the xplorerr package in your browser

Any scripts or data that you put into this service are public.

xplorerr documentation built on Feb. 28, 2019, 5:05 p.m.