dfalbel
commented
4 years ago You need to find a plot library that allows you to dynamically modify the data via js. For example you can display the metrics using something like this:
library(shiny)
library(keras)
ui <- fluidPage(
shinyjs::useShinyjs(), # Include shinyjs
actionButton("button", "Click me"),
textOutput('metrics')
)
server <- function(input, output) {
observeEvent(input$button, {
model <- keras_model_sequential() %>%
layer_dense(1)
model %>%
compile(loss = "mse" , optimizer = "adam", metrics = "mse")
cb <- callback_lambda(on_epoch_end = function(epoch, logs) {
shinyjs::html("metrics", paste("mse: ", logs$mse))
})
model %>%
fit(
x = matrix(runif(100), ncol = 1),
y = matrix(runif(100), ncol = 1),
callbacks = list(cb),
verbose = 0
)
})
}
# Run the application
shinyApp(ui = ui, server = server)
fabrizio1208
Hi! I was wondering how to insert in a shiny app a “live” history plot while the model is training/validating. I mean, the same plot that is shown in RStudio viewer pane.
Due to the research I have done, I believe that one way to do it is by using a custom callback function, but I’m not sure how to send the info (make it reactive) while the keras
fit()function is working.Any ideas?
Thanks in advance.