FEA-Bench / testbed /fairlearn__fairlearn /visualization /FairnessDashboard /src /Controls /ModelComparisonChart.tsx
| import React from "react"; | |
| import { AccessibleChart, ChartBuilder, IPlotlyProperty, PlotlyMode, SelectionContext } from "mlchartlib"; | |
| import { IFairnessContext } from "../IFairnessContext"; | |
| import _ from "lodash"; | |
| import { MetricsCache } from "../MetricsCache"; | |
| import { IAccuracyPickerProps, IParityPickerProps, IFeatureBinPickerProps } from "../FairnessWizard"; | |
| import { ParityModes } from "../ParityMetrics"; | |
| import { Stack, StackItem } from "office-ui-fabric-react/lib/Stack"; | |
| import { ChoiceGroup, IChoiceGroupOption } from 'office-ui-fabric-react/lib/ChoiceGroup'; | |
| import { localization } from "../Localization/localization"; | |
| import { Spinner, SpinnerSize } from "office-ui-fabric-react/lib/Spinner"; | |
| import { mergeStyleSets } from "@uifabric/styling"; | |
| import { ActionButton } from "office-ui-fabric-react/lib/Button"; | |
| import { AccuracyOptions } from "../AccuracyMetrics"; | |
| import { FormatMetrics } from "../FormatMetrics"; | |
| import { PredictionTypes } from "../IFairnessProps"; | |
| export interface IModelComparisonProps { | |
| dashboardContext: IFairnessContext; | |
| selections: SelectionContext; | |
| metricsCache: MetricsCache; | |
| modelCount: number; | |
| accuracyPickerProps: IAccuracyPickerProps; | |
| parityPickerProps: IParityPickerProps; | |
| featureBinPickerProps: IFeatureBinPickerProps; | |
| onEditConfigs: () => void; | |
| } | |
| export interface IState { | |
| accuracyArray?: number[]; | |
| disparityArray?: number[]; | |
| disparityInOutcomes: boolean; | |
| } | |
| export class ModelComparisonChart extends React.PureComponent<IModelComparisonProps, IState> { | |
| private readonly plotlyProps: IPlotlyProperty = { | |
| config: { displaylogo: false, responsive: true, modeBarButtonsToRemove: ['toggleSpikelines', 'hoverClosestCartesian', 'hoverCompareCartesian', 'zoom2d', 'pan2d', 'select2d', 'lasso2d', 'zoomIn2d', 'zoomOut2d', 'autoScale2d', 'resetScale2d'] }, | |
| data: [ | |
| { | |
| datapointLevelAccessors: { | |
| customdata: { | |
| path: ['index'], | |
| plotlyPath: 'customdata' | |
| } | |
| }, | |
| mode: PlotlyMode.markers, | |
| marker: { | |
| size: 14 | |
| }, | |
| type: 'scatter', | |
| xAccessor: 'Accuracy', | |
| yAccessor: 'Parity', | |
| hoverinfo: 'text' | |
| } | |
| ], | |
| layout: { | |
| autosize: true, | |
| font: { | |
| size: 10 | |
| }, | |
| margin: { | |
| t: 4, | |
| r:0 | |
| }, | |
| hovermode: 'closest', | |
| xaxis: { | |
| automargin: true, | |
| fixedrange: true, | |
| mirror: true, | |
| linecolor: '#CCCCCC', | |
| linewidth: 1, | |
| title:{ | |
| text: 'Error' | |
| } | |
| }, | |
| yaxis: { | |
| automargin: true, | |
| fixedrange: true, | |
| title:{ | |
| text: 'Disparity' | |
| } | |
| }, | |
| } as any | |
| }; | |
| private static readonly classNames = mergeStyleSets({ | |
| frame: { | |
| flex: 1, | |
| display: "flex", | |
| flexDirection: "column" | |
| }, | |
| spinner: { | |
| margin: "auto", | |
| padding: "40px" | |
| }, | |
| header: { | |
| backgroundColor: "#EBEBEB", | |
| padding: "0 90px", | |
| height: "90px", | |
| display: "inline-flex", | |
| flexDirection: "row", | |
| justifyContent: "space-between", | |
| alignItems: "center" | |
| }, | |
| headerTitle: { | |
| color: "#333333", | |
| fontSize: "32px", | |
| lineHeight: "39px", | |
| fontWeight: "100" | |
| }, | |
| editButton: { | |
| color: "#333333", | |
| fontSize: "12px", | |
| lineHeight: "20px", | |
| fontWeight: "400" | |
| }, | |
| main: { | |
| height: "100%", | |
| flex: 1, | |
| display: "inline-flex", | |
| flexDirection: "row" | |
| }, | |
| mainRight: { | |
| padding: "30px 0 0 35px", | |
| width: "300px" | |
| }, | |
| rightTitle: { | |
| color: "#333333", | |
| fontSize: "15px", | |
| lineHeight: "16px", | |
| fontWeight: "500", | |
| paddingBottom: "18px", | |
| borderBottom: "1px solid #CCCCCC" | |
| }, | |
| rightText: { | |
| padding: "16px 15px 30px 0", | |
| color: "#333333", | |
| fontSize: "15px", | |
| lineHeight: "18px", | |
| fontWeight: "400", | |
| borderBottom: "0.5px dashed #CCCCCC" | |
| }, | |
| insights: { | |
| textTransform: "uppercase", | |
| color: "#333333", | |
| fontSize: "15px", | |
| lineHeight: "16px", | |
| fontWeight: "500", | |
| padding: "18px 0", | |
| }, | |
| insightsText: { | |
| color: "#333333", | |
| fontSize: "15px", | |
| lineHeight: "16px", | |
| fontWeight: "400", | |
| paddingBottom: "18px", | |
| paddingRight: "15px", | |
| borderBottom: "1px solid #CCCCCC" | |
| }, | |
| chart: { | |
| padding: "60px 0 0 0", | |
| flex: 1 | |
| }, | |
| textSection: { | |
| paddingBottom: "5px" | |
| }, | |
| radio: { | |
| paddingBottom: "30px", | |
| paddingLeft: "75px" | |
| } | |
| }); | |
| constructor(props: IModelComparisonProps) { | |
| super(props); | |
| this.state = { | |
| disparityInOutcomes: true | |
| }; | |
| } | |
| public render(): React.ReactNode { | |
| if (!this.state || this.state.accuracyArray === undefined || this.state.disparityArray === undefined) { | |
| this.loadData(); | |
| return ( | |
| <Spinner className={ModelComparisonChart.classNames.spinner} size={SpinnerSize.large} label={localization.calculating}/> | |
| ); | |
| } | |
| const data = this.state.accuracyArray.map((accuracy, index) => { | |
| return { | |
| Parity: this.state.disparityArray[index], | |
| Accuracy: accuracy, | |
| index: index | |
| }; | |
| }); | |
| let minAccuracy: number = Number.MAX_SAFE_INTEGER; | |
| let maxAccuracy: number = Number.MIN_SAFE_INTEGER; | |
| let maxDisparity: number = Number.MIN_SAFE_INTEGER; | |
| let minDisparity: number = Number.MAX_SAFE_INTEGER; | |
| let minAccuracyIndex: number; | |
| let maxAccuracyIndex: number; | |
| let minDisparityIndex: number; | |
| let maxDisparityIndex: number; | |
| this.state.accuracyArray.forEach((value, index) => { | |
| if (value >= maxAccuracy) { | |
| maxAccuracyIndex = index; | |
| maxAccuracy = value; | |
| } | |
| if (value <= minAccuracy) { | |
| minAccuracyIndex = index; | |
| minAccuracy = value; | |
| } | |
| }); | |
| this.state.disparityArray.forEach((value, index) => { | |
| if (value >= maxDisparity) { | |
| maxDisparityIndex = index; | |
| maxDisparity = value; | |
| } | |
| if (value <= minDisparity) { | |
| minDisparityIndex = index; | |
| minDisparity = value; | |
| } | |
| }); | |
| const formattedMinAccuracy = FormatMetrics.formatNumbers(minAccuracy, this.props.accuracyPickerProps.selectedAccuracyKey); | |
| const formattedMaxAccuracy = FormatMetrics.formatNumbers(maxAccuracy, this.props.accuracyPickerProps.selectedAccuracyKey); | |
| const formattedMinDisparity = FormatMetrics.formatNumbers(minDisparity, this.props.accuracyPickerProps.selectedAccuracyKey); | |
| const formattedMaxDisparity = FormatMetrics.formatNumbers(maxDisparity, this.props.accuracyPickerProps.selectedAccuracyKey); | |
| const selectedMetric = AccuracyOptions[this.props.accuracyPickerProps.selectedAccuracyKey]; | |
| const insights2 = localization.formatString( | |
| localization.ModelComparison.insightsText2, | |
| selectedMetric.title, | |
| formattedMinAccuracy, | |
| formattedMaxAccuracy, | |
| formattedMinDisparity, | |
| formattedMaxDisparity | |
| ); | |
| const insights3 = localization.formatString( | |
| localization.ModelComparison.insightsText3, | |
| selectedMetric.title.toLowerCase(), | |
| selectedMetric.isMinimization ? formattedMinAccuracy : formattedMaxAccuracy, | |
| FormatMetrics.formatNumbers(this.state.disparityArray[selectedMetric.isMinimization ? minAccuracyIndex : maxAccuracyIndex], this.props.accuracyPickerProps.selectedAccuracyKey) | |
| ); | |
| const insights4 = localization.formatString( | |
| localization.ModelComparison.insightsText4, | |
| selectedMetric.title.toLowerCase(), | |
| FormatMetrics.formatNumbers(this.state.accuracyArray[minDisparityIndex], this.props.accuracyPickerProps.selectedAccuracyKey), | |
| formattedMinDisparity | |
| ); | |
| const howToReadText = localization.formatString( | |
| localization.ModelComparison.howToReadText, | |
| this.props.modelCount.toString(), | |
| selectedMetric.title.toLowerCase(), | |
| selectedMetric.isMinimization ? localization.ModelComparison.lower : localization.ModelComparison.higher | |
| ); | |
| const props = _.cloneDeep(this.plotlyProps); | |
| props.data = ChartBuilder.buildPlotlySeries(props.data[0], data).map(series => { | |
| series.name = this.props.dashboardContext.modelNames[series.name]; | |
| series.text = this.props.dashboardContext.modelNames; | |
| return series; | |
| }); | |
| const accuracyMetricTitle = AccuracyOptions[this.props.accuracyPickerProps.selectedAccuracyKey].title | |
| props.layout.xaxis.title = accuracyMetricTitle; | |
| props.layout.yaxis.title = this.state.disparityInOutcomes ? localization.ModelComparison.disparityInOutcomes : | |
| localization.formatString(localization.ModelComparison.disparityInAccuracy, accuracyMetricTitle.toLowerCase()) as string | |
| return ( | |
| <Stack className={ModelComparisonChart.classNames.frame}> | |
| <div className={ModelComparisonChart.classNames.header}> | |
| <h2 className={ModelComparisonChart.classNames.headerTitle}>{localization.ModelComparison.title}</h2> | |
| <ActionButton iconProps={{iconName: "Edit"}} onClick={this.props.onEditConfigs} className={ModelComparisonChart.classNames.editButton}>{localization.Report.editConfiguration}</ActionButton> | |
| </div> | |
| <div className={ModelComparisonChart.classNames.main}> | |
| <div className={ModelComparisonChart.classNames.chart}> | |
| <AccessibleChart | |
| plotlyProps={props} | |
| sharedSelectionContext={this.props.selections} | |
| theme={undefined} | |
| /> | |
| </div> | |
| <div className={ModelComparisonChart.classNames.mainRight}> | |
| <div className={ModelComparisonChart.classNames.rightTitle}>{localization.ModelComparison.howToRead}</div> | |
| <div className={ModelComparisonChart.classNames.rightText}>{howToReadText}</div> | |
| <div className={ModelComparisonChart.classNames.insights}>{localization.ModelComparison.insights}</div> | |
| <div className={ModelComparisonChart.classNames.insightsText}> | |
| <div className={ModelComparisonChart.classNames.textSection}>{insights2}</div> | |
| <div className={ModelComparisonChart.classNames.textSection}>{insights3}</div> | |
| <div>{insights4}</div> | |
| </div> | |
| </div> | |
| </div> | |
| <div> | |
| <ChoiceGroup | |
| className={ModelComparisonChart.classNames.radio} | |
| selectedKey={this.state.disparityInOutcomes ? "outcomes" : "accuracy"} | |
| options={[ | |
| { | |
| key: 'accuracy', | |
| text: localization.formatString(localization.ModelComparison.disparityInAccuracy, | |
| AccuracyOptions[this.props.accuracyPickerProps.selectedAccuracyKey].title.toLowerCase()) as string | |
| }, | |
| { | |
| key: 'outcomes', | |
| text: localization.ModelComparison.disparityInOutcomes | |
| } | |
| ]} | |
| onChange={this.disparityChanged} | |
| label={localization.ModelComparison.howToMeasureDisparity} | |
| required={false} | |
| ></ChoiceGroup> | |
| </div> | |
| </Stack>); | |
| } | |
| private async loadData(): Promise<void> { | |
| try { | |
| const accuracyPromises = new Array(this.props.modelCount).fill(0).map((unused, modelIndex) => { | |
| return this.props.metricsCache.getMetric( | |
| this.props.dashboardContext.binVector, | |
| this.props.featureBinPickerProps.selectedBinIndex, | |
| modelIndex, | |
| this.props.accuracyPickerProps.selectedAccuracyKey); | |
| }); | |
| const disparityMetric = this.state.disparityInOutcomes ? | |
| (this.props.dashboardContext.modelMetadata.predictionType === PredictionTypes.binaryClassification ? | |
| "selection_rate" : "average") : | |
| this.props.accuracyPickerProps.selectedAccuracyKey; | |
| const disparityPromises = new Array(this.props.modelCount).fill(0).map((unused, modelIndex) => { | |
| return this.props.metricsCache.getDisparityMetric( | |
| this.props.dashboardContext.binVector, | |
| this.props.featureBinPickerProps.selectedBinIndex, | |
| modelIndex, | |
| disparityMetric, | |
| ParityModes.difference); | |
| }); | |
| const accuracyArray = (await Promise.all(accuracyPromises)).map(metric => metric.global); | |
| const disparityArray = await Promise.all(disparityPromises); | |
| this.setState({accuracyArray, disparityArray}); | |
| } catch { | |
| // todo; | |
| } | |
| } | |
| private readonly disparityChanged = (ev: React.FormEvent<HTMLInputElement>, option: IChoiceGroupOption): void => { | |
| const disparityInOutcomes = option.key !== "accuracy"; | |
| if (this.state.disparityInOutcomes !== disparityInOutcomes) { | |
| this.setState({disparityInOutcomes, disparityArray: undefined}); | |
| } | |
| } | |
| // TODO: Reuse if multiselect re-enters design | |
| // private readonly applySelections = (chartId: string, selectionIds: string[], plotlyProps: IPlotlyProperty) => { | |
| // if (!plotlyProps.data || plotlyProps.data.length === 0) { | |
| // return; | |
| // } | |
| // const customData: string[] = (plotlyProps.data[0] as any).customdata; | |
| // if (!customData) { | |
| // return; | |
| // } | |
| // const colors = customData.map(modelIndex => { | |
| // const selectedIndex = this.props.selections.selectedIds.indexOf(modelIndex); | |
| // if (selectedIndex !== -1) { | |
| // return FabricStyles.plotlyColorPalette[selectedIndex % FabricStyles.plotlyColorPalette.length]; | |
| // } | |
| // return "#111111"; | |
| // }); | |
| // const shapes = customData.map(modelIndex => { | |
| // const selectedIndex = this.props.selections.selectedIds.indexOf(modelIndex); | |
| // if (selectedIndex !== -1) { | |
| // return 1 | |
| // } | |
| // return 0; | |
| // }); | |
| // Plotly.restyle(chartId, 'marker.color' as any, [colors] as any); | |
| // Plotly.restyle(chartId, 'marker.symbol' as any, [shapes] as any); | |
| // } | |
| } |