NYC Taxi Crossfilter

Introduction

This vignette demonstrates a complex multi-view dashboard with crossfilter interactions. The example visualizes 1 million NYC taxi rides using: - Raster density maps for pickup and dropoff locations - Histogram of pickup times - Interactive brushing across all views

Example: NYC Taxi Rides with Crossfilter

library(rMosaic)

# Define the spec
taxi_spec <- list(
  meta = list(
    title       = "NYC Taxi Rides",
    description = paste(
      "Pickup/dropoff points for 1M NYC taxi rides on Jan 1–3, 2010.",
      "Drag to filter both maps.",
      "Requires DuckDB 'httpfs' + 'spatial' extensions.",
      sep = "\n"
    )
  ),
  config = list(extensions = c("httpfs", "spatial")),
  data = list(
    rides = list(
      file   = "https://idl.uw.edu/mosaic-datasets/data/nyc-rides-2010.parquet",
      select = c(
        "pickup_datetime::TIMESTAMP AS datetime",
        "ST_Transform(ST_Point(pickup_latitude, pickup_longitude),'EPSG:4326','ESRI:102718') AS pick",
        "ST_Transform(ST_Point(dropoff_latitude, dropoff_longitude),'EPSG:4326','ESRI:102718') AS drop"
      )
    ),
    trips = paste(
      "SELECT",
      "  (HOUR(datetime) + MINUTE(datetime)/60) AS time,",
      "  ST_X(pick) AS px, ST_Y(pick) AS py,",
      "  ST_X(drop) AS dx, ST_Y(drop) AS dy",
      "FROM rides",
      sep = "\n"
    )
  ),
  params = list(filter = list(select = "crossfilter")),
  vconcat = list(
    list(
      hconcat = list(
        # Left: pickup raster + brush
        list(
          plot = list(
            list(mark = "raster", data = list(from = "trips", filterBy = "$filter"),
                 x = "px", y = "py", bandwidth = 0),
            list(select = "intervalXY", as = "$filter"),
            list(mark = "text", data = list(list(label = "Taxi Pickups")),
                 dx = 10, dy = 10, text = "label",
                 fill = "black", fontSize = "1.2em",
                 frameAnchor = "top-left")
          ),
          width       = 335, height = 550,
          margin      = 0, xAxis = NULL, yAxis = NULL,
          xDomain     = c(975000, 1005000),
          yDomain     = c(190000, 240000),
          colorScale  = "symlog", colorScheme = "blues"
        ),
        # Spacer
        list(hspace = 10),
        # Right: dropoff raster + same brush
        list(
          plot = list(
            list(mark = "raster", data = list(from = "trips", filterBy = "$filter"),
                 x = "dx", y = "dy", bandwidth = 0),
            list(select = "intervalXY", as = "$filter"),
            list(mark = "text", data = list(list(label = "Taxi Dropoffs")),
                 dx = 10, dy = 10, text = "label",
                 fill = "black", fontSize = "1.2em",
                 frameAnchor = "top-left")
          ),
          width       = 335, height = 550,
          margin      = 0, xAxis = NULL, yAxis = NULL,
          xDomain     = c(975000, 1005000),
          yDomain     = c(190000, 240000),
          colorScale  = "symlog", colorScheme = "oranges"
        )
      )
    ),
    # Vertical space
    list(vspace = 10),
    # Histogram + brush
    list(
      plot = list(
        list(mark = "rectY", data = list(from = "trips", filterBy = "$filter"),
             x = list(bin = "time"),
             y = list(count = NULL),
             fill = "steelblue", inset = 0.5),
        list(select = "intervalX", as = "$filter")
      ),
      yTickFormat = "s",
      xLabel      = "Pickup Hour →",
      width       = 680,
      height      = 100
    )
  )
)

# Run the app
runMosaicApp(
  spec     = taxi_spec,
  specType = "yaml",
  data     = NULL,
  title    = "NYC Taxi Rides (Raster + Crossfilter)",
  backend  = "wasm"
)

Key Features

Crossfilter Selection

The filter parameter uses select = "crossfilter"
Brushing in any view filters all other views
Multiple selections are intersected

Remote Data Loading

Data loaded directly from Parquet files via URL
No need to download data locally
DuckDB handles efficient querying

Spatial Transformations

Uses DuckDB’s spatial extension
Projects coordinates to appropriate CRS
Efficient raster rendering with bandwidth = 0

Performance

1M rows rendered smoothly
WASM backend for browser-side computation
Raster aggregation for dense point clouds

Try It Yourself

Brush a region in the pickup map
Notice the dropoff map and histogram update
Brush a time range in the histogram
See both maps update to show only rides in that time window

This demonstrates Mosaic’s ability to handle large datasets with fluid, interactive crossfilter selections.