library(shiny) #nGramAll <- fread('predictionTableFull.csv')
shinyUI(fluidPage(
titlePanel("Next Word Prediction App"),
sidebarLayout( sidebarPanel( strong('Introduction'), p('This is the Shiny app for next word prediction for the Coursera Data Science Capstone Project'), p("The algorithm is based on Katz's back-off model using 2-gram to 7-grams"), strong('Instructions'), p('The inputs are as follows'), tags$ul( tags$li('Query word/phrase: word or phrase for prediction'), tags$li('Number of predicted next word: the number of predictions') ), strong('Output'), p('The predicted next word will show up in the order of most frequently used to less frequently used') ),
# Show a plot of the generated distribution
mainPanel(
h4('Query word/phrase:'),
tags$textarea(id = 'query', rows = 2, cols = 50),
HTML("<br>"), HTML("<br>"),
h4('Number of predictions:'),
sliderInput('wordN', '',
min = 1, max = 5, value = 1, step = 1),
HTML("<br>"), hr(), HTML("<br>"),
h4('Predicted next word'),
verbatimTextOutput('predicted')
)
) ))