Performance and large datasets

Rendering cost depends on the number and complexity of drawn primitives, the backend, canvas size, and raster supersampling. Measure the chart and workload you actually plan to ship.

from dataviz import Plot, Theme, save

def main() raises:
    var x = List[Float64]()
    var y = List[Float64]()
    for i in range(10_000):
        x.append(Float64(i))
        y.append(Float64((i * 17) % 101))
    var plot = (
        Plot().mark_point().encode(x=x, y=y)
        .theme(Theme(raster_supersample=1))
        .labels(title="Ten Thousand Points")
    )
    save(plot, "large.png")

Start with the default backend and settings, then profile. Avoid labels for every observation in dense charts. Reduce output dimensions or raster supersampling when antialiasing is not worth the cost. Aggregation or sampling can improve both performance and readability, but should preserve the pattern the chart is meant to show.

Run pixi run bench for the repository benchmark suite. See the benchmark methodology and Data shapes.