I'm trying to use Kandy for a scatter plot, but I ...

# datascience

[martmists]

I'm trying to use Kandy for a scatter plot, but I get this error:

Unequal column sizes. Expected rows count: 35. Actual column sizes:
x: 35
y: 35
x*: 31

My code:

val clusters = clusterKMeans(3, vectors)

        plot {
            points {
                for ((cluster, clusterColor) in clusters.zip(colors)) {
                    x(cluster.map { it[0] })
                    y(cluster.map { it[1] })
                    color = clusterColor
                }
            }
        }.save("kmeans-3.png")

[Andrei Kislitsyn]

Hi! What you are trying to do is called "series approach" - you have a series of data (clusters in your case) and you want to add points of corresponding color for each series. Unfortunately, this approach is not supported in Kandy at the moment, so you need to transform your data a bit and change the code - you need 3 lists - all Xs and Ys points and the cluster name/number for the corresponding point and also add color scale from cluster name/index.
here's more on that:
https://kotlin.github.io/kandy/series-hack-guide.html

[Andrei Kislitsyn]

You need something like that:

plot {
    points {
        x(clusters.flatMap { cluster -> cluster.map { it[0] } })
        y(clusters.flatMap { cluster -> cluster.map { it[1] } })
        color(clusters.flatMapIndexed { index, cluster -> List(cluster.size) { index } }) {
            scale = categorical(colors, colors.indices.toList())
        }
    }
}

[martmists]

I see. And how can I specify colors in a bar chart? I tried doing

"color" to colors.keys.map {
                val r = (it[0] * 255).roundToInt()
                val g = (it[1] * 255).roundToInt()
                val b = (it[2] * 255).roundToInt()
                Color.rgb(r, g, b)
            }

and then specifying fillColor("color") but that didn't seem to work, instead putting the colors as strings in a legend.

[Andrei Kislitsyn]

Unfortunately, Kandy hasn't identity color scale, so you should provide list of cluster labels into color

[Andrei Kislitsyn]

And then use scale to set how labels are mapped to color.
Example:

"color" to listOf("a", "b", "c", "a", etc)

fillColor("color") {
   scale = categorical("a" to Color.RED, "b" to Color.BLUE, "c" to... )
   scale = categorical("a" to Color.RED, "b" to Color.BLUE, "c" to... )

}

[Andrei Kislitsyn]

In previous example I provided indices into color mapping and add scale from them