2b:null 2c:["$","title",null,{"children":"ClarityLens: See your Data and Model | Akhila Sulgante | Observable"}] 31:T5b8,async function split_attr(attr){ let plots = [] let tableData = [] for (let value of Object.keys(indexed_data[attr])){ let data = raw_adult.filter((d)=> { return d[attr] == value }) let select_data = (d) => { let arr = [] for(let feature of trainFeatures){ arr.push(indexed_data[feature][d[feature]]) } return arr } let Xtrain = data.slice(0, split(data.length, trainSplit)).map((d)=> { return select_data(d) }) let Xtest = data.slice(Xtrain.length+1, -1).map((d)=> { return select_data(d) }) let Ytrain = data.slice(0, split(data.length, trainSplit)).map((d)=>{ return d["income"] === ">50K" ? 1 : 0 }) let Ytest = data.slice(Xtrain.length+1, -1).map((d)=>{ return d["income"] === ">50K" ? 1 : 0 }) let nb = new ML.NaiveBayes.MultinomialNB() console.log({Xtrain,Ytrain}) nb.train(Xtrain, Ytrain); let results = nb.predict(Xtest); let {chart,plotData} = await confusion_matrix(results, Ytest) tableData.push([value,...plotData]) // ${model_split} : ${value} plots.push(htl.html`

${await chart}

`) } return htl.html`${plots} ${Inputs.table(tableData,{header:[attr,"TP","FP","FN","TN", "Accuracy", "Misclassification", "Precision", "True Positive Rate(Sensitivity)", "False Positive Rate(Specificity)"]})}` }32:T4c3,confusion_matrix = async (nb_results, Ytest) =>{ // >50k: 1, <=50: 0 let tp =0,tn=0,fp=0,fn=0; for(let i=0;i\nThere are libraries and tools in place to tackle and identify biases in the dataset, and this tool also aims to contribute to a more equitable space in machine learning.","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":11,"value":"The primary objective behind this tool is to assist users in pinpointing biases within their dataset both before and after model training. As said by John Tukey, _\"The greatest value of a picture is when it forces us to notice what we never expected to see\"_.
\nThis tool **empowers** users to customize visualizations, allowing for tailored views of distribution or composition. Moreover, it allows the assessment of attribute correlations, enhancing comprehension of their interconnections.\n\nSometimes, merely observing data isn't sufficient to identify bias. It is essential to train our models and assess their performance to ensure fairness and avoid disadvantaging any particular group. This tool offers users the choice to train a binary classifier and partition the data based on suspected biased attribute values to get a closer look at Accuracy, Misclassification, Precision, Sentivity and Specifity.","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":39,"value":"## Tool Overview/ Capabilities\nThis tool has the following options for the users\n\n** In terms of Visualizing the data **\n1. Users have the option to upload their preferred dataset (For this demonstration, the tool can be explored using the UCI adult dataset)\n2. Users can select a feature they wish to visualize, displaying it in either distribution or composition format\n3. Users can select two features to explore and understand any correlation between the chosen features\n\n** In terms of evaluating the model**\n1. Users have the flexibility to set the train and test split ratio\n2. Users can select the feature used to train the model\n3. Users can specify the groups to scrutinize for potential bias (in this demonstration, the tool offers options to assess bias related to gender and race)\n4. Users can view the confusion matrix, accuracy, precision, misclassification rate, sensitivity, and specificity of the groups within the selected attribute.\n","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":50,"value":"## Let's break down our exploration of the tool into steps.","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":55,"value":"### Step1: Upload your data\n_We're in the process of making the tool applicable to various datasets. Presently, we're utilizing the UCI adult dataset for our exploration_","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":66,"value":"viewof data = dataInput({\n value: adult3,\n delimiter: ',', // Optional\n format: \"auto\" // Optional, one of MongoExport, JSON, CSV, CSVNoAutoType\n})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":81,"value":"### Step2: Explore your data","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":86,"value":"Select the features you want to visualize!
\n**Data distribution** refers to the way data values are spread or distributed across a dataset.
\n**Data composition** refers to the makeup or the components of a dataset. ","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":94,"value":"viewof selected_columns = searchCheckbox(\n Object.keys(data[0]), // What elements to chose from\n {// Any other options you pass here will be passed to the checkbox and search inputs\n value: [\"sex\"], // Start with everything selected\n label: \"Features\",\n height: 200\n // You can also pass specific options for the search and the checkboxes\n // optionsSearch: {\n // filter: fullSearchFilter\n // }\n }\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":103,"value":"viewof chart_type = Inputs.select([\"Distribution\", \"Composition\"], {label:\"Select chart type\"})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":106,"value":"htl.html`${await Promise.all(selected_columns.map((d)=> {\n if(chart_type === \"Distribution\"){\n return display_histogram(data, d)\n } else {\n return display_pieChart(data, d)\n }\n \n}))}`","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":115,"value":"### Step3: Attribute correlation\nSelect features from the dataset to find correlations between them","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":118,"value":"viewof y_axis = Inputs.select(Object.keys(data[0]),{label:\"Y-axis\",value:\"sex\"})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":121,"value":"viewof x_axis = Inputs.select(Object.keys(data[0]),{label: \"X-axis\",value:\"age\"})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":126,"value":"barChart(data,x_axis,y_axis)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":135,"value":"### Step4: Train your model\nSelect from the features and train-test split to train your model","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":138,"value":"viewof trainFeatures = searchCheckbox(\n Object.keys(adult[0]).filter(d=>d!='income'), // What elements to chose from\n {// Any other options you pass here will be passed to the checkbox and search inputs\n value: Object.keys(adult[0]), // Start with everything selected\n label: \"Train Features\",\n height: 200\n // You can also pass specific options for the search and the checkboxes\n // optionsSearch: {\n // filter: fullSearchFilter\n // }\n }\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":147,"value":"viewof trainSplit = Inputs.range([0.1,0.9],{step:0.1,label:\"Train Split\"})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":143,"value":"### Step5: Select the characteristics you believe might exhibit bias\n_We have selected race and gender for demo purposes_","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":150,"value":"viewof model_split = Inputs.select([\"sex\", \"race\"], {label:\"Split model by\"})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":153,"value":"htl.html`

${ await\n(split_attr([model_split]))\n}

`","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":179,"value":"### Step6: Make decisions\nFrom the above matrix, make your best judgments to decide if the model is biased towards any particular group","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":189,"value":"### Iteration 2\n1. Add Adverse Impact and chi-squared calculations for the dataset\n2. Add Disparity index, Statistical Parity Difference, and Equal Opportunity Difference table\n3. Add % change in the evaluation matrix","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":165,"value":"$31","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":169,"value":"split = (total_length, percentage)=>{\n return percentage * total_length\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":159,"value":"function generateIndices(data, attributes){\n let indices = {}\n attributes.forEach((d)=>{\n indices[d] = {};\n Array.from(new Set(data.map((v)=>v[d]))).forEach((v,i)=> {\n indices[d][v] = i;\n })\n })\n return indices\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":172,"value":"$32","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":162,"value":"indexed_data = generateIndices(adult, Object.keys(adult[0]))","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":128,"value":"function barChart(data,x_axis,y_axis){\n let x = vl.x().fieldN(x_axis)\n let y = vl.y().fieldN(y_axis)\n\n if(x_axis === \"fnlwgt\"){\n x = vl.x().fieldQ(x_axis)\n }\n if(y_axis === \"fnlwgt\"){\n y = vl.y().fieldQ(y_axis)\n }\n return vl.markRect({tooltip:true})\n .encode(\n x,y,\n vl.color().count()\n )\n .data(data)\n .height(400)\n .width(width-200)\n .render()\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":108,"value":"function display_histogram(data, attr) {\n \n let x = vl.x().field(attr)\n \n\n if(attr === \"fnlwgt\"){\n x = vl.x().fieldQ(attr)\n }\n \n return vl.markBar({tooltip: true})\n.encode(\n x, vl.y().count()\n \n)\n .height(400)\n .width(width)\n.data(data)\n.render()\n \n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":111,"value":"function display_pieChart(data, attr) {\n\n return vl.markArc({tooltip: true})\n.encode(\n vl.color().field(attr),\n vl.theta().count()\n)\n .height(400)\n .width(width-200)\n.data(data)\n.render()\n \n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":68,"value":"import {dataInput} from \"@john-guerra/data-input\"","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":96,"value":"import {searchCheckbox} from \"@john-guerra/search-checkbox\"","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":155,"value":"ML = require('ml@5.0.0/dist/ml.js')","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":71,"value":"adult3 = FileAttachment(\"adult-3.data\").csv({typed:true})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":140,"value":"adult = adult3.filter(d=>d[\"sex\"] || d[\"race\"])","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":175,"value":"raw_adult = _.shuffle( await FileAttachment(\"adult-3.data\").csv({typed:true}))","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W33","stdlib":"1","titling":"auto"}} 30:"$2f:body"