Merge pull request #91 from NGO-Algorithm-Audit/feature/feature-7-finetuning

devhelpr · web-flow · commit 885628c0ee7d · 2025-06-08T13:27:45.000+02:00
Feature/feature 7 finetuning
diff --git a/src/assets/bias-detection-python-code.tsx b/src/assets/bias-detection-python-code.tsx
@@ -27,6 +27,7 @@ from js import targetColumn
 from js import dataType
 from js import higherIsBetter
 from js import isDemo
+from js import dataTypeText
 
 def t_test_on_cluster(test_df, bias_score, cluster_label):
 
@@ -64,6 +65,7 @@ def t_test_on_cluster(test_df, bias_score, cluster_label):
 
     # if significant differences were found, print the variables with their differences
     for var, res in t_test_results.items():
+        
         if res['p_val'] < 0.05:
             direction = res['direction']
             if direction == "higher":
@@ -139,7 +141,19 @@ def chi2_test_on_cluster(decoded_X_test, bias_score, cluster_label):
         print(f"91mNo statistically significant differences in means found.")
 
     # if significant differences were found, print the variables with their differences
+    current_column = ""
+
     for var, res in chi2_results.items():
+        if (current_column != var[0]):
+            current_column = var[0]
+            if res['p_val'] < 0.05:
+                comparisons.append({
+                        'key': 'biasAnalysis.biasedCluster.differenceCategorical.feature',
+                        'params': {
+                            'feature': var[0],
+                        }
+                    })
+
         if res['p_val'] < 0.05:
             direction = res['direction']
             if direction == "higher":
@@ -318,7 +332,7 @@ def run():
             'iterations': localIterations,
             'minClusterSize': localClusterSize,
             'performanceMetric': bias_score,
-            'dataType': localDataType
+            'dataType': dataTypeText
         }
     }))
     setResult(json.dumps({
@@ -597,7 +611,8 @@ def run():
         setResult(json.dumps({
             'type': 'accordion',
             'titleKey': 'biasAnalysis.biasedCluster.accordionTitle',
-            'comparisons': comparisons
+            'comparisons': comparisons,
+            'className': 'biasAnalysis-biasedClusterAccordion'
         }))
 
 
diff --git a/src/components/BiasSettings.tsx b/src/components/BiasSettings.tsx
@@ -162,6 +162,7 @@ export default function BiasSettings({
                     form.getValues().whichPerformanceMetricValueIsBetter ===
                     'higher',
                 isDemo: true,
+                dataTypeText: t(`biasSettings.dataType.categorical`),
             }
         );
     };
@@ -224,6 +225,9 @@ export default function BiasSettings({
             higherIsBetter:
                 formData.whichPerformanceMetricValueIsBetter === 'higher',
             isDemo: false,
+            dataTypeText: t(
+                `biasSettings.dataType.${formData.selectedDataType}`
+            ),
         });
     };
 
diff --git a/src/components/bias-detection-interfaces/BiasDetectionParameters.ts b/src/components/bias-detection-interfaces/BiasDetectionParameters.ts
@@ -5,4 +5,5 @@ export interface BiasDetectionParameters {
     dataType: string;
     higherIsBetter: boolean;
     isDemo: boolean;
+    dataTypeText: string;
 }
diff --git a/src/components/componentMapper.tsx b/src/components/componentMapper.tsx
@@ -108,6 +108,7 @@ export default function ComponentMapper({
                                     key={index}
                                     title={t(resultItem.titleKey || '')}
                                     content={content}
+                                    className={resultItem?.className ?? ''}
                                 />
                             );
                         }
diff --git a/src/components/ui/accordion.tsx b/src/components/ui/accordion.tsx
@@ -59,11 +59,12 @@ AccordionContent.displayName = AccordionPrimitive.Content.displayName;
 export interface AccordionProps {
     title: string;
     content: React.ReactNode;
+    className?: string;
 }
 
 export const Accordion = (props: AccordionProps) => (
     <>
-        <div className="hideonprint">
+        <div className={`hideonprint ${props.className ?? ''}`}>
             <AccordionPrimitive.Root type="single" collapsible>
                 <AccordionItem value="item-1">
                     <AccordionTrigger>{props.title}</AccordionTrigger>
@@ -79,7 +80,7 @@ export const Accordion = (props: AccordionProps) => (
                 </AccordionItem>
             </AccordionPrimitive.Root>
         </div>
-        <div className="hidden showonprint">
+        <div className={`hidden showonprint ${props.className ?? ''}`}>
             {typeof props.content === 'string' ? (
                 <MarkdownWithTooltips className="text-gray-800 markdown">
                     {props.content}
diff --git a/src/index.css b/src/index.css
@@ -43,6 +43,22 @@
     }
 }
 
+.biasAnalysis-biasedClusterAccordion {
+    .content-list,
+    .markdown {
+        ul li {
+            line-height: 1;
+        }
+
+        ul {
+            @apply mb-0;
+        }
+        span:has(> ul) {
+            @apply block;
+        }
+    }
+}
+
 .tooltip-content {
     @apply p-4;
 }
diff --git a/src/locales/en.ts b/src/locales/en.ts
@@ -21,6 +21,10 @@ export const en = {
         'Loading core packages. On average this takes 10-15 seconds.',
     installingPackages: 'Installing additional packages...',
     biasSettings: {
+        dataType: {
+            numeric: 'Numeric',
+            categorical: 'Categorical',
+        },
         exportToPDF: 'Download bias analysis report as pdf',
         exportToJSON: 'Export clusters as json',
         form: {
@@ -263,8 +267,9 @@ we recommend to impute the missing data. For {tooltip:syntheticData.missingDataM
 
 A two-sided t-test is performed to accept or reject <i class="font-serif">H</i><sub>0</sub>:.
 
-p_value : {{p_val}}
+{tooltip:biasAnalysis.p_valueTooltip}p_value{/tooltip} : {{p_val}} 
         `,
+        p_valueTooltip: `p_value tooltip`,
         dataSetPreview: {
             heading: '1. Preview of data',
         },
@@ -329,10 +334,9 @@ In this example, we analyze which group is most adversely affected by the risk p
 `,
             },
             differenceCategorical: {
-                deviatingMoreOften: `**{{feature}}: {{value}}** in the most deviating cluster occurs **more** often than in the rest of the dataset.
-`,
-                deviatingLessOften: `**{{feature}}: {{value}}** in the most deviating cluster occurs **less** often than in the rest of the dataset.
-`,
+                feature: '- {{feature}}',
+                deviatingMoreOften: `  - **{{value}}** in the most deviating cluster occurs **more** often than in the rest of the dataset.`,
+                deviatingLessOften: `  - **{{value}}** in the most deviating cluster occurs **less** often than in the rest of the dataset.`,
             },
         },
         nodifference: {
@@ -341,8 +345,7 @@ In this example, we analyze which group is most adversely affected by the risk p
         },
         distribution: {
             mainHeading: '6. Cluster differences wrt. features',
-            heading:
-                '"{{variable}}" distribution per cluster:',
+            heading: '"{{variable}}" distribution per cluster:',
         },
         splittingDataset: {
             heading: '3. Splitting dataset',
diff --git a/src/locales/nl.ts b/src/locales/nl.ts
@@ -18,6 +18,10 @@ export const nl = {
     loadingPackages: 'Laden van packages. Dit duurt gemiddeld 10-15 seconden.',
     installingPackages: 'Aanvullende packages laden',
     biasSettings: {
+        dataType: {
+            numeric: 'Numeriek',
+            categorical: 'Categorisch',
+        },
         exportToPDF: 'Download bias analyse rapport als pdf',
         exportToJSON: 'Export clusters als json',
 
@@ -191,6 +195,9 @@ export const nl = {
             '&nbsp;&nbsp;\n  \n  \n  \nWil je meer weten over synthetische data?\n  \n  \n  \n- [python-synthpop op Github](https://github.com/NGO-Algorithm-Audit/python-synthpop)\n- [local-first web app op Github](https://github.com/NGO-Algorithm-Audit/local-first-web-tool/tree/main)\n- [Synthetische Data: wat, waarom en hoe?](https://royalsociety.org/-/media/policy/projects/privacy-enhancing-technologies/Synthetic_Data_Survey-24.pdf)\n- [Kennis Netwerk Synthetische Data](https://online.rijksinnovatiecommunity.nl/groups/399-kennisnetwerk-synthetischedata/welcome) (for Dutch public organizations)\n- [Synthetische data portaal van DUO](https://duo.nl/open_onderwijsdata/footer/synthetische-data.jsp)\n- [CART: synthpop resources](https://synthpop.org.uk/resources.html)\n- [Gaussian Copula - Synthetic Data Vault](https://docs.sdv.dev/sdv)',
     },
     biasAnalysis: {
+        dataSetPreview: {
+            heading: '1. Preview van de data',
+        },
         demo: {
             heading: 'Informatie over de demodataset',
             description: `Als voorbeeld wordt de [COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) dataset](https://github.com/propublica/compas-analysis/tree/master) gebruikt. De dataset bevat kenmerken van criminele verdachten en hun risico op recidive, zoals voorspeld door het COMPAS-algoritme. De dataset bevat demografische gegevens zoals leeftijd, geslacht en ras, evenals strafblad, details over de aanklacht en het voorspelde risicolabel. Deze dataset wordt gebruikt als benchmark voor onderzoek naar eerlijke algoritmes. Een beschrijving van alle variabelen is te vinden in onderstaande tabel.
@@ -219,8 +226,9 @@ In dit voorbeeld analyseren welke groepen het COMPAS risicotaxatie-algoritme afw
 
 Er wordt een tweezijdige t-toets uitgevoerd om <i class="font-serif">H</i><sub>0</sub> te aanvaarden of te verwerpen.
 
-p-waarde : {{p_val}}
+{tooltip:biasAnalysis.p_valueTooltip}p-waarde{/tooltip} : {{p_val}}
         `,
+        p_valueTooltip: `p-waarde tooltip`,
         parameters: {
             heading: '2. Geselecteerde parameters',
             iterations: 'Aantal iteraties: {{value}}',
@@ -236,8 +244,7 @@ p-waarde : {{p_val}}
 `,
         },
         distribution: {
-            mainHeading:
-                '6. Verschil clusters mbt. eigenschappen',
+            mainHeading: '6. Verschil clusters mbt. eigenschappen',
             heading: '"{{variable}}" verdeling per cluster:',
         },
         splittingDataset: {
diff --git a/src/routes/BiasDetection.tsx b/src/routes/BiasDetection.tsx
@@ -81,6 +81,7 @@ export default function BiasDetection() {
         dataType: 'numeric',
         higherIsBetter: false,
         isDemo: false,
+        dataTypeText: t(`biasSettings.dataType.numeric`),
     });
     const onFileLoad: (
         data: Record<string, string>[],
@@ -127,6 +128,7 @@ export default function BiasDetection() {
                     dataType: params.dataType,
                     higherIsBetter: params.higherIsBetter,
                     isDemo: params.isDemo,
+                    dataTypeText: params.dataTypeText,
                 },
             },
         });

Original file line number	Diff line number	Diff line change
`@@ -5,4 +5,5 @@ export interface BiasDetectionParameters {`
`5`	`5`	`dataType: string;`
`6`	`6`	`higherIsBetter: boolean;`
`7`	`7`	`isDemo: boolean;`
	`8`	`+ dataTypeText: string;`
`8`	`9`	`}`
Original file line number	Diff line number	Diff line change
`@@ -108,6 +108,7 @@ export default function ComponentMapper({`
`108`	`108`	`key={index}`
`109`	`109`	`title={t(resultItem.titleKey \|\| '')}`
`110`	`110`	`content={content}`
	`111`	`+ className={resultItem?.className ?? ''}`
`111`	`112`	`/>`
`112`	`113`	`);`
`113`	`114`	`}`