commonMain.org.jetbrains.letsPlot.bistro.residual.ResidualUtil.kt Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of lets-plot-kotlin-kernel Show documentation
Lets-Plot Kotlin API without dependencies.
There is a newer version: 4.9.2
Show newest version
/*
 * Copyright (c) 2022. JetBrains s.r.o.
 * Use of this source code is governed by the MIT license that can be found in the LICENSE file.
 */

package org.jetbrains.letsPlot.bistro.residual

import org.jetbrains.letsPlot.commons.interval.DoubleSpan
import org.jetbrains.letsPlot.core.commons.data.SeriesUtil
import org.jetbrains.letsPlot.core.plot.base.DataFrame
import org.jetbrains.letsPlot.core.plot.base.data.DataFrameUtil
import org.jetbrains.letsPlot.core.plot.base.util.SamplingUtil
import kotlin.random.Random

internal object ResidualUtil {
    fun appendResiduals(
        data: Map>,
        x: String,
        y: String,
        colorBy: String?,
        model: Model,
        loessCriticalSize: Int,
        samplingSeed: Long
    ): Map<*, *> {

        fun emptyDataWithResiduals() = (data.keys + ResidualVar.RESIDUAL).associateWith { emptyList() }

        if (data.isEmpty() || data.values.first().isEmpty()) {
            return emptyDataWithResiduals()
        }

        val groupResults = splitByGroup(data, colorBy).map { groupData ->
            if (groupData.isEmpty() || !groupData.containsKey(x) || !groupData.containsKey(y)) {
                return@map emptyDataWithResiduals()
            }
            val xs = groupData[x]!!
            val ys = groupData[y]!!
            val indices = xs.zip(ys)
                .withIndex()
                .filter { (_, pair) -> SeriesUtil.allFinite(pair.first as? Double, pair.second as? Double) }
                .map { it.index }

            if (indices.size < 2) {
                return@map emptyDataWithResiduals()
            }

            val values = groupData.mapValues { SeriesUtil.pickAtIndices(it.value, indices) }
            appendResidualsToGroup(values, x, y, model, loessCriticalSize, samplingSeed)
        }

        // merge maps
        return groupResults.fold(mutableMapOf>()) { mergedMap, map ->
            for ((key, value) in map) {
                mergedMap.getOrPut(key, ::mutableListOf).addAll(value)
            }
            mergedMap
        }
    }

    private fun appendResidualsToGroup(
        groupData: Map>,
        x: String,
        y: String,
        model: Model,
        loessCriticalSize: Int,
        samplingSeed: Long
    ): Map> {
        val df = DataFrameUtil.fromMap(groupData)
        val afterSampling = applySampling(df, model, loessCriticalSize, samplingSeed)
        val xs = getNumeric(afterSampling, x)
        val xRange = SeriesUtil.range(xs)
        val predictor: (Double) -> Double = model.getPredictor(
            xs = xs,
            ys = getNumeric(afterSampling, y)
        )

        val result = groupData.toMutableMap()
        result[ResidualVar.RESIDUAL] = calculateResiduals(
            xs = getNumeric(df, x),
            ys = getNumeric(df, y),
            predictor,
            range = xRange
        )
        return result
    }

    private fun calculateResiduals(
        xs: List,
        ys: List,
        predictor: (Double) -> Double,
        range: DoubleSpan?
    ): List {
        return (xs zip ys).map { (x, y) ->
            if (SeriesUtil.allFinite(x, y) && range?.contains(x!!) == true) {
                y!! - predictor(x!!)
            } else {
                null
            }
        }
    }

    private fun applySampling(
        df: DataFrame,
        model: Model,
        loessCriticalSize: Int,
        samplingSeed: Long
    ): DataFrame {
        return if (model.method == Model.Method.LOESS && df.rowCount() > loessCriticalSize) {
            SamplingUtil.sampleWithoutReplacement(loessCriticalSize, Random(samplingSeed), df)
        } else {
            df
        }
    }

    private fun getNumeric(df: DataFrame, varName: String): List {
        return df.getNumeric(DataFrameUtil.findVariableOrFail(df, varName))
    }

    private fun splitByGroup(data: Map>, groupName: String?): List>> {
        if (groupName == null || !data.containsKey(groupName)) {
            return listOf(data)
        }

        val groupIndices = mutableMapOf>()
        data[groupName]!!.forEachIndexed { index, v ->
            groupIndices.getOrPut(v, ::mutableListOf).add(index)
        }

        val result = groupIndices.map { (_, indices) ->
            data.mapValues { (_, values) -> SeriesUtil.pickAtIndices(values, indices) }
        }
        return result
    }
}