library(shiny)
library(shinydashboard)
library(shinydashboardPlus)
library(shinyWidgets)
library(shinycssloaders)
library(DT)
library(plotly)
library(scico)
library(ggthemes)
library(scales)
library(data.table)
library(dtplyr)
library(parallel)
library(Rnumerai)
# ==============================================================================
# Helper Functions
# ==============================================================================
# Download raw data
download_raw_data <- function(model_name) {
# Download data from Numerai
d_raw <- round_model_performances(model_name)
# Remove rows without CORR
d_raw <- d_raw[!is.na(d_raw$corr), ]
# Add the model name
d_raw$model <- model_name
# Return
return(as.data.table(d_raw))
}
# Reformat
reformat_data <- function(d_raw) {
# Keep some columns only
col_keep <- c("model", "roundNumber",
"roundOpenTime", "roundResolveTime",
"roundResolved", "selectedStakeValue",
"corr", "corrPercentile",
"fncV3", "fncV3Percentile",
"tc", "tcPercentile",
"corrWMetamodel",
"roundPayoutFactor", "payout")
d_munged <- d_raw[, col_keep, with = FALSE]
# Date
d_munged[, roundOpenTime := as.Date(roundOpenTime)]
d_munged[, roundResolveTime := as.Date(roundResolveTime)]
# Reformat percentile
d_munged[, corrPercentile := round(corrPercentile * 100, 6)]
d_munged[, fncV3Percentile := round(fncV3Percentile * 100, 6)]
d_munged[, tcPercentile := round(tcPercentile * 100, 6)]
# Rename columns
colnames(d_munged) <- c("model", "round",
"date_open", "date_resolved",
"resolved", "stake",
"corr", "corr_pct",
"fncv3", "fncv3_pct",
"tc", "tc_pct",
"corr_meta",
"pay_ftr", "payout")
# Return
return(d_munged)
}
# ==============================================================================
# UI
# ==============================================================================
ui <- shinydashboardPlus::dashboardPage(
title = "Shiny Numerati",
skin = "black-light",
options = list(sidebarExpandOnHover = TRUE),
header = shinydashboardPlus::dashboardHeader(
title = "✨ Shiny Numerati",
userOutput("user")
),
# ============================================================================
# Sidebar
# ============================================================================
sidebar = shinydashboardPlus::dashboardSidebar(
id = "sidebar",
sidebarMenu(
menuItem(text = "Start Here", tabName = "start", icon = icon("play")),
menuItem(text = "Payout Summary", tabName = "payout", icon = icon("credit-card")),
menuItem(text = "Model Performance", tabName = "performance", icon = icon("line-chart")),
menuItem(text = "About", tabName = "about", icon = icon("question-circle"))
),
minified = TRUE,
collapsed = FALSE
),
# ============================================================================
# Main Body
# ============================================================================
body = dashboardBody(
tabItems(
# ========================================================================
# Start Here
# ========================================================================
tabItem(tabName = "start",
fluidPage(
# ==============================================================
# Special script to keep the session alive for a bit longer
# ==============================================================
tags$head(
HTML(
"
"
)
),
# ==============================================================
# First Page
# ==============================================================
markdown("# **Shiny Numerati**"),
markdown("### Community Dashboard for the Numerai Classic Tournament"),
br(),
fluidRow(
column(6,
markdown("## **Step 1 - Select Your Models**"),
markdown("### First, click this ⬇"),
pickerInput(inputId = "model",
label = " ",
choices = sort(Rnumerai::get_leaderboard()$username),
multiple = TRUE,
width = "100%",
options = list(
`title` = "---------->>> HERE <<<----------",
`header` = "Notes: 1) Use the search box below to find and select your models. 2) Use 'Select All' for quick selection.",
size = 20,
`actions-box` = TRUE,
`live-search` = TRUE,
`live-search-placeholder` = "For example, try lgbm_v4 or integration_test",
`virtual-scroll` = TRUE,
`multiple-separator` = ", ",
`selected-text-format`= "count > 3",
`count-selected-text` = "{0} models selected (out of {1})",
`deselect-all-text` = "Deselect All",
`select-all-text` = "Select All"
)
)
),
column(6,
markdown("## **Step 2 - Download Data**"),
markdown("### Next, click this ⬇ (it may take a while)"),
br(),
actionBttn(inputId = "button_download",
label = "Download Data from Numerai",
color = "primary",
icon = icon("cloud-download"),
style = "gradient",
block = TRUE
)
)
),
br(),
h3(strong(textOutput(outputId = "text_download"))),
verbatimTextOutput(outputId = "print_download"),
br(),
h3(strong(textOutput(outputId = "text_preview"))),
shinycssloaders::withSpinner(DTOutput("dt_model")),
br(),
h3(strong(textOutput(outputId = "text_next"))),
h3(strong(textOutput(outputId = "text_soon")))
)
),
# ========================================================================
# Payout Summary
# ========================================================================
tabItem(tabName = "payout",
fluidPage(
markdown("# **Payout Summary**"),
markdown("### Remember to refresh the charts after making changes to model selection or settings below"),
br(),
fluidRow(
column(6,
markdown("## **Step 1 - Define the Range**"),
sliderInput(inputId = "range_round",
label = "Numerai Classic Tournament Rounds",
width = "100%",
min = 168, # first tournament round
max = Rnumerai::get_current_round(),
# note: daily rounds from round 339
value = c(339, Rnumerai::get_current_round())
)
),
column(6,
markdown("## **Step 2 - Visualise**"),
br(),
actionBttn(inputId = "button_filter",
label = "Create / Refresh Charts",
color = "primary",
icon = icon("refresh"),
style = "gradient",
block = TRUE)
)
), # end of fluidRow
br(),
tabsetPanel(type = "tabs",
tabPanel("All Models",
br(),
h3(strong(textOutput(outputId = "text_payout"))),
fluidRow(
class = "text-center",
valueBoxOutput("payout_confirmed", width = 2),
valueBoxOutput("payout_pending", width = 2),
valueBoxOutput("payout_total", width = 2),
valueBoxOutput("payout_n_round", width = 2),
valueBoxOutput("payout_average", width = 2),
valueBoxOutput("payout_avg_ror", width = 2)
),
br(),
shinycssloaders::withSpinner(plotlyOutput("plot_payout_stacked")),
br(),
br(),
DTOutput("dt_payout_summary")
),
tabPanel("Individual Models",
br(),
shinycssloaders::withSpinner(plotlyOutput("plot_payout_individual"))
)
) # end of tabsetPanel
) # end of fluidPage
),
# ========================================================================
# Model Performance
# ========================================================================
tabItem(tabName = "performance",
fluidPage(
markdown("![image](https://media.giphy.com/media/cftSzNoCTfSyAWctcl/giphy.gif)")
)
),
# ========================================================================
# About
# ========================================================================
tabItem(tabName = "about",
markdown("## **About this App**"),
markdown('#### Yet another Numerai community dashboard by Jo-fai Chow.'),
br(),
markdown("## **Acknowledgements**"),
markdown("- #### This hobby project was inspired by Rajiv's shiny-kmeans on 🤗 Spaces."),
markdown('- #### The Rnumerai package from Omni Analytics Group.'),
br(),
markdown("## **Changelog**"),
markdown(
"
- #### **0.1.0** — First prototype with an interactive table output
- #### **0.1.1** — Added a functional `Payout Summary` page
- #### **0.1.2** — `Payout Summary` layout updates
"),
br(),
markdown("## **Session Info**"),
verbatimTextOutput(outputId = "session_info"),
br(),
textOutput("keepAlive") # trick to keep session alive
)
# ========================================================================
) # end of tabItems
),
footer = shinydashboardPlus::dashboardFooter(
left = "Powered by ❤️, ☕, Shiny, and 🤗 Spaces",
right = paste0("Version 0.1.2"))
)
# ==============================================================================
# Server
# ==============================================================================
server <- function(input, output) {
# About Joe
output$user <- renderUser({
dashboardUser(
name = "JC",
image = "https://numerai-public-images.s3.amazonaws.com/profile_images/aijoe_v5_compressed-iJWEo1WeHkpH.jpg",
subtitle = "@matlabulous",
footer = p('"THE NMR LIFE CHOSE ME."', class = 'text-center')
)
})
# ============================================================================
# Reactive: Data
# ============================================================================
react_ls_model <- eventReactive(input$button_download, {sort(input$model)})
output$print_download <- renderPrint({react_ls_model()})
output$text_download <- renderText({
if (length(react_ls_model()) >= 1) "Your Selection:" else " "
})
output$text_preview <- renderText({
if (length(react_ls_model()) >= 1) "Data Preview:" else " "
})
output$text_next <- renderText({
if (length(react_ls_model()) >= 1) "⬅ [NEW] Payout Summary 📊💸" else " "
})
output$text_soon <- renderText({
if (length(react_ls_model()) >= 1) "⬅ [COMING SOON] Model Performance 📈🔥" else " "
})
react_d_raw <- eventReactive(
input$button_download,
{
# Parallelised download
d_raw <- rbindlist(mclapply(X = input$model,
FUN = download_raw_data,
mc.cores = detectCores()))
# Return
d_raw
}
)
# ============================================================================
# Reactive: DataTable
# ============================================================================
output$dt_model <- DT::renderDT({
# Raw Data
d_raw <- react_d_raw()
# Reformat
d_munged <- reformat_data(d_raw)
# Main DT
DT::datatable(
# Data
d_munged,
# Other Options
rownames = FALSE,
extensions = "Buttons",
options =
list(
dom = 'Bflrtip', # https://datatables.net/reference/option/dom
buttons = list('csv', 'excel', 'copy', 'print'), # https://rstudio.github.io/DT/003-tabletools-buttons.html
order = list(list(0, 'asc'), list(1, 'asc')),
pageLength = 5,
lengthMenu = c(5, 10, 20, 100, 500, 1000, 50000),
columnDefs = list(list(className = 'dt-center', targets = "_all")))
) |>
# Reformat individual columns
formatRound(columns = c("corr", "tc", "fncv3", "corr_meta", "pay_ftr"), digits = 4) |>
formatRound(columns = c("corr_pct", "tc_pct", "fncv3_pct"), digits = 1) |>
formatRound(columns = c("stake", "payout"), digits = 2) |>
formatStyle(columns = c("model"),
fontWeight = "bold") |>
formatStyle(columns = c("stake"),
fontWeight = "bold",
color = styleInterval(cuts = -1e-15, values = c("#D24141", "#2196F3"))) |>
formatStyle(columns = c("corr", "fncv3"),
color = styleInterval(cuts = -1e-15, values = c("#D24141", "black"))) |>
formatStyle(columns = c("tc"),
color = styleInterval(cuts = -1e-15, values = c("#D24141", "#A278DC"))) |>
formatStyle(columns = c("corr_pct", "tc_pct", "fncv3_pct"),
color = styleInterval(cuts = c(1, 5, 15, 85, 95, 99),
values = c("#692020", "#9A2F2F", "#D24141",
"#D1D1D1", # light grey
"#00A800", "#007000", "#003700"))) |>
formatStyle(columns = c("payout"),
fontWeight = "bold",
color = styleInterval(cuts = c(-1e-15, 1e-15),
values = c("#D24141", "#D1D1D1", "#00A800")))
})
# ============================================================================
# Reactive: filtering data for all charts
# ============================================================================
react_d_filter <- eventReactive(
input$button_filter,
{
# Reformat and Filter
d_filter <- reformat_data(react_d_raw())
d_filter <- d_filter[pay_ftr > 0, ] # ignoring the new daily rounds for now
d_filter <- d_filter[round >= input$range_round[1], ]
d_filter <- d_filter[round <= input$range_round[2], ]
# Return
d_filter
})
react_d_payout_summary <- eventReactive(
input$button_filter,
{
# Summarise payout
d_smry <-
react_d_filter() |>
lazy_dt() |>
group_by(round, date_open, date_resolved, resolved) |>
summarise(staked_models = n(),
total_stake = sum(stake, na.rm = T),
net_payout = sum(payout, na.rm = T)) |>
as.data.table()
d_smry$rate_of_return <- (d_smry$net_payout / d_smry$total_stake) * 100
# Return
d_smry
})
# ============================================================================
# Reactive: Payout Value Boxes
# ============================================================================
output$text_payout <- renderText({
if (nrow(react_d_filter()) >= 1) "Payouts in NMR" else " "
})
output$payout_confirmed <- renderValueBox({
valueBox(value = round(sum(react_d_filter()[resolved == TRUE, ]$payout, na.rm = T), 2),
subtitle = "Realised",
color = "olive")
})
output$payout_pending <- renderValueBox({
valueBox(value = round(sum(react_d_filter()[resolved == FALSE, ]$payout, na.rm = T), 2),
subtitle = "Pending",
color = "yellow")
})
output$payout_total <- renderValueBox({
valueBox(value = round(sum(react_d_filter()$payout, na.rm = T), 2),
subtitle = "Realised + Pending",
color = "light-blue")
})
output$payout_n_round <- renderValueBox({
# Use rounds with stake > 0 only
valueBox(value = nrow(react_d_payout_summary()[total_stake > 0, ]),
subtitle = "Staked Rounds",
color = "light-blue")
})
output$payout_average <- renderValueBox({
# Use rounds with stake > 0 only
valueBox(value = round(mean(react_d_payout_summary()[total_stake > 0, ]$net_payout, na.rm = T), 2),
subtitle = "Avg. Round Payout",
color = "light-blue")
})
output$payout_avg_ror <- renderValueBox({
# Use rounds with stake > 0 only
valueBox(value = paste(round(mean(react_d_payout_summary()[total_stake > 0, ]$rate_of_return), 2), "%"),
subtitle = "Avg. Round ROR",
color = "light-blue")
})
# ============================================================================
# Reactive: Payout Charts
# ============================================================================
# Stacked Bar Chart
output$plot_payout_stacked <- renderPlotly({
# Data
d_filter <- react_d_filter()
# ggplot
p <- ggplot(d_filter,
aes(x = date_resolved, y = payout, fill = payout,
text = paste("Model:", model,
"\nRound:", round,
"\nRound Open Date:", date_open,
"\nRound Resolved Date:", date_resolved,
"\nRound Status:", resolved,
"\nPayout:", round(payout,2), "NMR"))) +
geom_bar(position = "stack", stat = "identity") +
theme(
panel.border = element_rect(fill = 'transparent',
color = "grey", linewidth = 0.25),
panel.background = element_rect(fill = 'transparent'),
plot.background = element_rect(fill = 'transparent', color = NA),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.background = element_rect(fill = 'transparent'),
strip.text = element_text(),
strip.clip = "on",
legend.background = element_rect(fill = 'transparent'),
legend.box.background = element_rect(fill = 'transparent')
) +
geom_hline(aes(yintercept = 0), linewidth = 0.25, color = "grey") +
scale_fill_scico(palette = "vikO", direction = -1, midpoint = 0) +
scale_x_date(breaks = breaks_pretty(10),
labels = label_date_short(format = c("%Y", "%b", "%d"), sep = "\n")
) +
xlab(" \nDate (Round Resolved / Resolving)") +
ylab("Payout (NMR)")
# Generate plotly
ggplotly(p, tooltip = "text")
})
# Individual
output$plot_payout_individual <- renderPlotly({
# Data
d_filter <- react_d_filter()
# Get the number of unique models
n_model <- length(unique(d_filter$model))
# Base plot
p <- ggplot(d_filter,
aes(x = round, y = payout, fill = payout,
text = paste("Model:", model,
"\nRound:", round,
"\nRound Open Date:", date_open,
"\nRound Resolved Date:", date_resolved,
"\nRound Status:", resolved,
"\nPayout:", round(payout,2), "NMR"))) +
geom_bar(stat = "identity") +
theme(
panel.border = element_rect(fill = 'transparent',
color = "grey", linewidth = 0.25),
panel.background = element_rect(fill = 'transparent'),
plot.background = element_rect(fill = 'transparent', color = NA),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.background = element_rect(fill = 'transparent'),
strip.text = element_text(),
strip.clip = "on",
legend.background = element_rect(fill = 'transparent'),
legend.box.background = element_rect(fill = 'transparent'),
axis.text.x = element_text(angle = 45, hjust = 1)
) +
geom_hline(aes(yintercept = 0), linewidth = 0.25, color = "grey") +
scale_fill_scico(palette = "vikO", direction = -1, midpoint = 0) +
scale_x_continuous(breaks = breaks_pretty(5)) +
xlab(" \nTournament Round") +
ylab("Confirmed / Pending Payout (NMR)")
# Facet setting
if (n_model %% 5 == 0) {
p <- p + facet_wrap(. ~ model, ncol = 5)
} else {
p <- p + facet_wrap(. ~ model)
}
# Dynamic height adjustment
height <- 600 # default
if (n_model > 10) height = 800
if (n_model > 15) height = 1000
if (n_model > 20) height = 1200
if (n_model > 25) height = 1400
if (n_model > 30) height = 1600
if (n_model > 35) height = 1800
if (n_model > 40) height = 2000
if (n_model > 45) height = 2200
if (n_model > 50) height = 2400
# Generate plotly
ggplotly(p, height = height, tooltip = "text")
})
# ============================================================================
# Reactive: Payout Summary Table
# ============================================================================
output$dt_payout_summary <- DT::renderDT({
# Generate a new DT
DT::datatable(
# Data
react_d_payout_summary(),
# Other Options
rownames = FALSE,
extensions = "Buttons",
options =
list(
dom = 'Bflrtip', # https://datatables.net/reference/option/dom
buttons = list('csv', 'excel', 'copy', 'print'), # https://rstudio.github.io/DT/003-tabletools-buttons.html
order = list(list(0, 'asc'), list(1, 'asc')),
pageLength = 100,
lengthMenu = c(5, 10, 20, 100, 500, 1000),
columnDefs = list(list(className = 'dt-center', targets = "_all")))
) |>
# Reformat individual columns
formatRound(columns = c("total_stake", "net_payout", "rate_of_return"), digits = 2) |>
formatStyle(columns = c("round"), fontWeight = "bold") |>
formatStyle(columns = c("resolved"),
target = "row",
backgroundColor = styleEqual(c(1,0), c("transparent", "#FFF8E1"))) |>
formatStyle(columns = c("total_stake"),
fontWeight = "bold",
color = styleInterval(cuts = -1e-15, values = c("#D24141", "#2196F3"))) |>
formatStyle(columns = c("net_payout"),
fontWeight = "bold",
color = styleInterval(cuts = c(-1e-15, 1e-15),
values = c("#D24141", "#D1D1D1", "#00A800"))) |>
formatStyle(columns = c("rate_of_return"),
fontWeight = "bold",
color = styleInterval(cuts = c(-1e-15, 1e-15),
values = c("#D24141", "#D1D1D1", "#00A800")))
})
# ============================================================================
# Session Info
# ============================================================================
output$session_info <- renderPrint({
sessionInfo()
})
# ============================================================================
# Trick to keep session alive
# https://tickets.dominodatalab.com/hc/en-us/articles/360015932932-Increasing-the-timeout-for-Shiny-Server
# ============================================================================
output$keepAlive <- renderText({
req(input$count)
# paste("keep alive ", input$count)
" "
})
}
# ==============================================================================
# App
# ==============================================================================
shinyApp(ui, server)