LDAvis = { /* Original code taken from https://github.com/cpsievert/LDAvis */ /* Copyright 2013, AT&T Intellectual Property */ /* MIT Licence */ 'use strict'; var LDAvis = function(to_select, data_or_file_name, width) { // This section sets up the logic for event handling var current_clicked = { what: "nothing", element: undefined }, current_hover = { what: "nothing", element: undefined }, old_winning_state = { what: "nothing", element: undefined }, vis_state = { lambda: 1, topic: 0, term: "" }; // Set up a few 'global' variables to hold the data: var K, // number of topics R, // number of terms to display in bar chart mdsData, // (x,y) locations and topic proportions mdsData3, // topic proportions for all terms in the viz lamData, // all terms that are among the top-R most relevant for all topics, lambda values lambda = { old: 1, current: 1 }, color1 = "#1f77b4", // baseline color for default topic circles and overall term frequencies color2 = "#d62728"; // 'highlight' color for selected topics and term-topic frequencies // Set the duration of each half of the transition: var duration = 750; // Set global margins used for everything var margin = { top: 30, right: 30, bottom: 70, left: 30 }, mdswidth = (width-100)/2, //was: 530, mdsheight = 530, barwidth = (width-100)/2, //was: 530, barheight = 530, termwidth = 90, // width to add between two panels to display terms mdsarea = mdsheight * mdswidth; // controls how big the maximum circle can be // doesn't depend on data, only on mds width and height: var rMax = 60*width/530; // was: 60; // proportion of area of MDS plot to which the sum of default topic circle areas is set var circle_prop = 0.25; var word_prop = 0.25; // opacity of topic circles: var base_opacity = 0.2, highlight_opacity = 0.6; // topic/lambda selection names are specific to *this* vis var topic_select = to_select + "-topic"; var lambda_select = to_select + "-lambda"; // get rid of the # in the to_select (useful) for setting ID values var visID = to_select.replace("#", ""); var topicID = visID + "-topic"; var lambdaID = visID + "-lambda"; var termID = visID + "-term"; var topicDown = topicID + "-down"; var topicUp = topicID + "-up"; var topicClear = topicID + "-clear"; var leftPanelID = visID + "-leftpanel"; var barFreqsID = visID + "-bar-freqs"; var topID = visID + "-top"; var lambdaInputID = visID + "-lambdaInput"; var lambdaZeroID = visID + "-lambdaZero"; var sliderDivID = visID + "-sliderdiv"; var lambdaLabelID = visID + "-lamlabel"; ////////////////////////////////////////////////////////////////////////////// // sort array according to a specified object key name // Note that default is decreasing sort, set decreasing = -1 for increasing // adpated from http://stackoverflow.com/questions/16648076/sort-array-on-key-value function fancysort(key_name, decreasing) { decreasing = (typeof decreasing === "undefined") ? 1 : decreasing; return function(a, b) { if (a[key_name] < b[key_name]) return 1 * decreasing; if (a[key_name] > b[key_name]) return -1 * decreasing; return 0; }; } function visualize(data) { // set the number of topics to global variable K: K = data['mdsDat'].x.length; // R is the number of top relevant (or salient) words whose bars we display R = Math.min(data['R'], 30); // a (K x 5) matrix with columns x, y, topics, Freq, cluster (where x and y are locations for left panel) mdsData = []; for (var i = 0; i < K; i++) { var obj = {}; for (var key in data['mdsDat']) { obj[key] = data['mdsDat'][key][i]; } mdsData.push(obj); } // a huge matrix with 3 columns: Term, Topic, Freq, where Freq is all non-zero probabilities of topics given terms // for the terms that appear in the barcharts for this data mdsData3 = []; for (var i = 0; i < data['token.table'].Term.length; i++) { var obj = {}; for (var key in data['token.table']) { obj[key] = data['token.table'][key][i]; } mdsData3.push(obj); } // large data for the widths of bars in bar-charts. 6 columns: Term, logprob, loglift, Freq, Total, Category // Contains all possible terms for topics in (1, 2, ..., k) and lambda in the user-supplied grid of lambda values // which defaults to (0, 0.01, 0.02, ..., 0.99, 1). lamData = []; for (var i = 0; i < data['tinfo'].Term.length; i++) { var obj = {}; for (var key in data['tinfo']) { obj[key] = data['tinfo'][key][i]; } lamData.push(obj); } var dat3 = lamData.slice(0, R); // Create the topic input & lambda slider forms. Inspired from: // http://bl.ocks.org/d3noob/10632804 // http://bl.ocks.org/d3noob/10633704 init_forms(topicID, lambdaID, visID); // When the value of lambda changes, update the visualization console.log('lambda_select', lambda_select); d3.select(lambda_select) .on("input", function() { console.log('lambda_select mouseup'); // store the previous lambda value lambda.old = lambda.current; lambda.current = document.getElementById(lambdaID).value; vis_state.lambda = +this.value; // adjust the text on the range slider d3.select(lambda_select).property("value", vis_state.lambda); d3.select(lambda_select + "-value").text(vis_state.lambda); // transition the order of the bars var increased = lambda.old < vis_state.lambda; if (vis_state.topic > 0) reorder_bars(increased); // store the current lambda value state_save(true); document.getElementById(lambdaID).value = vis_state.lambda; }); d3.select("#" + topicUp) .on("click", function() { // remove term selection if it exists (from a saved URL) var termElem = document.getElementById(termID + vis_state.term); if (termElem !== undefined) term_off(termElem); vis_state.term = ""; var value_old = document.getElementById(topicID).value; var value_new = Math.min(K, +value_old + 1).toFixed(0); // increment the value in the input box document.getElementById(topicID).value = value_new; topic_off(document.getElementById(topicID + value_old)); topic_on(document.getElementById(topicID + value_new)); vis_state.topic = value_new; state_save(true); }); d3.select("#" + topicDown) .on("click", function() { // remove term selection if it exists (from a saved URL) var termElem = document.getElementById(termID + vis_state.term); if (termElem !== undefined) term_off(termElem); vis_state.term = ""; var value_old = document.getElementById(topicID).value; var value_new = Math.max(0, +value_old - 1).toFixed(0); // increment the value in the input box document.getElementById(topicID).value = value_new; topic_off(document.getElementById(topicID + value_old)); topic_on(document.getElementById(topicID + value_new)); vis_state.topic = value_new; state_save(true); }); d3.select("#" + topicID) .on("keyup", function() { // remove term selection if it exists (from a saved URL) var termElem = document.getElementById(termID + vis_state.term); if (termElem !== undefined) term_off(termElem); vis_state.term = ""; topic_off(document.getElementById(topicID + vis_state.topic)); var value_new = document.getElementById(topicID).value; if (!isNaN(value_new) && value_new > 0) { value_new = Math.min(K, Math.max(1, value_new)); topic_on(document.getElementById(topicID + value_new)); vis_state.topic = value_new; state_save(true); document.getElementById(topicID).value = vis_state.topic; } }); d3.select("#" + topicClear) .on("click", function() { state_reset(); state_save(true); }); // create linear scaling to pixels (and add some padding on outer region of scatterplot) var xrange = d3.extent(mdsData, function(d) { return d.x; }); //d3.extent returns min and max of an array var xdiff = xrange[1] - xrange[0], xpad = 0.05; var yrange = d3.extent(mdsData, function(d) { return d.y; }); var ydiff = yrange[1] - yrange[0], ypad = 0.05; if (xdiff > ydiff) { var xScale = d3.scale.linear() .range([0, mdswidth]) .domain([xrange[0] - xpad * xdiff, xrange[1] + xpad * xdiff]); var yScale = d3.scale.linear() .range([mdsheight, 0]) .domain([yrange[0] - 0.5*(xdiff - ydiff) - ypad*xdiff, yrange[1] + 0.5*(xdiff - ydiff) + ypad*xdiff]); } else { var xScale = d3.scale.linear() .range([0, mdswidth]) .domain([xrange[0] - 0.5*(ydiff - xdiff) - xpad*ydiff, xrange[1] + 0.5*(ydiff - xdiff) + xpad*ydiff]); var yScale = d3.scale.linear() .range([mdsheight, 0]) .domain([yrange[0] - ypad * ydiff, yrange[1] + ypad * ydiff]); } // Create new svg element (that will contain everything): var svg = d3.select(to_select).append("svg") .attr("width", mdswidth + barwidth + margin.left + termwidth + margin.right) .attr("height", mdsheight + 2 * margin.top + margin.bottom + 2 * rMax); // Create a group for the mds plot var mdsplot = svg.append("g") .attr("id", leftPanelID) .attr("class", "points") .attr("transform", "translate(" + margin.left + "," + 2 * margin.top + ")"); // Clicking on the mdsplot should clear the selection mdsplot .append("rect") .attr("x", 0) .attr("y", 0) .attr("height", mdsheight) .attr("width", mdswidth) .style("fill", color1) .attr("opacity", 0) .on("click", function() { state_reset(); state_save(true); }); mdsplot.append("line") // draw x-axis .attr("x1", 0) .attr("x2", mdswidth) .attr("y1", mdsheight / 2) .attr("y2", mdsheight / 2) .attr("stroke", "gray") .attr("opacity", 0.3); mdsplot.append("text") // label x-axis .attr("x", 0) .attr("y", mdsheight/2 - 5) .text(data['plot.opts'].xlab) .attr("fill", "gray"); mdsplot.append("line") // draw y-axis .attr("x1", mdswidth / 2) .attr("x2", mdswidth / 2) .attr("y1", 0) .attr("y2", mdsheight) .attr("stroke", "gray") .attr("opacity", 0.3); mdsplot.append("text") // label y-axis .attr("x", mdswidth/2 + 5) .attr("y", 7) .text(data['plot.opts'].ylab) .attr("fill", "gray"); // new definitions based on fixing the sum of the areas of the default topic circles: var newSmall = Math.sqrt(0.02*mdsarea*circle_prop/Math.PI); var newMedium = Math.sqrt(0.05*mdsarea*circle_prop/Math.PI); var newLarge = Math.sqrt(0.10*mdsarea*circle_prop/Math.PI); var cx = 10 + newLarge, cx2 = cx + 1.5 * newLarge; // circle guide inspired from // http://www.nytimes.com/interactive/2012/02/13/us/politics/2013-budget-proposal-graphic.html?_r=0 var circleGuide = function(rSize, size) { d3.select("#" + leftPanelID).append("circle") .attr('class', "circleGuide" + size) .attr('r', rSize) .attr('cx', cx) .attr('cy', mdsheight + rSize) .style('fill', 'none') .style('stroke-dasharray', '2 2') .style('stroke', '#999'); d3.select("#" + leftPanelID).append("line") .attr('class', "lineGuide" + size) .attr("x1", cx) .attr("x2", cx2) .attr("y1", mdsheight + 2 * rSize) .attr("y2", mdsheight + 2 * rSize) .style("stroke", "gray") .style("opacity", 0.3); }; circleGuide(newSmall, "Small"); circleGuide(newMedium, "Medium"); circleGuide(newLarge, "Large"); var defaultLabelSmall = "2%"; var defaultLabelMedium = "5%"; var defaultLabelLarge = "10%"; d3.select("#" + leftPanelID).append("text") .attr("x", 10) .attr("y", mdsheight - 10) .attr('class', "circleGuideTitle") .style("text-anchor", "left") .style("fontWeight", "bold") .text("Marginal topic distribtion"); d3.select("#" + leftPanelID).append("text") .attr("x", cx2 + 10) .attr("y", mdsheight + 2 * newSmall) .attr('class', "circleGuideLabelSmall") .style("text-anchor", "start") .text(defaultLabelSmall); d3.select("#" + leftPanelID).append("text") .attr("x", cx2 + 10) .attr("y", mdsheight + 2 * newMedium) .attr('class', "circleGuideLabelMedium") .style("text-anchor", "start") .text(defaultLabelMedium); d3.select("#" + leftPanelID).append("text") .attr("x", cx2 + 10) .attr("y", mdsheight + 2 * newLarge) .attr('class', "circleGuideLabelLarge") .style("text-anchor", "start") .text(defaultLabelLarge); // bind mdsData to the points in the left panel: var points = mdsplot.selectAll("points") .data(mdsData) .enter(); // text to indicate topic points.append("text") .attr("class", "txt") .attr("x", function(d) { return (xScale(+d.x)); }) .attr("y", function(d) { return (yScale(+d.y) + 4); }) .attr("stroke", "black") .attr("opacity", 1) .style("text-anchor", "middle") .style("fontWeight", 100) .text(function(d) { return d.topics; }); // draw circles points.append("circle") .attr("class", "dot") .style("opacity", 0.2) .style("fill", color1) .attr("r", function(d) { //return (rScaleMargin(+d.Freq)); return (Math.sqrt((d.Freq/100)*mdswidth*mdsheight*circle_prop/Math.PI)); }) .attr("cx", function(d) { return (xScale(+d.x)); }) .attr("cy", function(d) { return (yScale(+d.y)); }) .attr("stroke", "black") .attr("id", function(d) { return (topicID + d.topics); }) .on("mouseover", function(d) { var old_topic = topicID + vis_state.topic; if (vis_state.topic > 0 && old_topic != this.id) { topic_off(document.getElementById(old_topic)); } topic_on(this); }) .on("click", function(d) { // prevent click event defined on the div container from firing // http://bl.ocks.org/jasondavies/3186840 d3.event.stopPropagation(); var old_topic = topicID + vis_state.topic; if (vis_state.topic > 0 && old_topic != this.id) { topic_off(document.getElementById(old_topic)); } // make sure topic input box value and fragment reflects clicked selection document.getElementById(topicID).value = vis_state.topic = d.topics; state_save(true); topic_on(this); }) .on("mouseout", function(d) { if (vis_state.topic != d.topics) topic_off(this); if (vis_state.topic > 0) topic_on(document.getElementById(topicID + vis_state.topic)); }); svg.append("text") .text("Intertopic Distance Map (via multidimensional scaling)") .attr("x", mdswidth/2 + margin.left) .attr("y", 30) .style("font-size", "larger") .style("text-anchor", "middle"); // establish layout and vars for bar chart var barDefault2 = dat3.filter(function(d) { return d.Category == "Default"; }); var y = d3.scale.ordinal() .domain(barDefault2.map(function(d) { return d.Term; })) .rangeRoundBands([0, barheight], 0.15); var x = d3.scale.linear() .domain([1, d3.max(barDefault2, function(d) { return d.Total; })]) .range([0, barwidth]) .nice(); var yAxis = d3.svg.axis() .scale(y); // Add a group for the bar chart var chart = svg.append("g") .attr("transform", "translate(" + +(mdswidth + margin.left + termwidth) + "," + 2 * margin.top + ")") .attr("id", barFreqsID); // bar chart legend/guide: var barguide = {"width": 100, "height": 15}; d3.select("#" + barFreqsID).append("rect") .attr("x", 0) .attr("y", mdsheight + 10) .attr("height", barguide.height) .attr("width", barguide.width) .style("fill", color1) .attr("opacity", 0.4); d3.select("#" + barFreqsID).append("text") .attr("x", barguide.width + 5) .attr("y", mdsheight + 10 + barguide.height/2) .style("dominant-baseline", "middle") .text("Overall term frequency"); d3.select("#" + barFreqsID).append("rect") .attr("x", 0) .attr("y", mdsheight + 10 + barguide.height + 5) .attr("height", barguide.height) .attr("width", barguide.width/2) .style("fill", color2) .attr("opacity", 0.8); d3.select("#" + barFreqsID).append("text") .attr("x", barguide.width/2 + 5) .attr("y", mdsheight + 10 + (3/2)*barguide.height + 5) .style("dominant-baseline", "middle") .text("Estimated term frequency within the selected topic"); // footnotes: d3.select("#" + barFreqsID) .append("a") .attr("xlink:href", "http://vis.stanford.edu/files/2012-Termite-AVI.pdf") .attr("target", "_blank") .append("text") .attr("x", 0) .attr("y", mdsheight + 10 + (6/2)*barguide.height + 5) .style("dominant-baseline", "middle") .text("1. saliency(term w) = frequency(w) * [sum_t p(t | w) * log(p(t | w)/p(t))] for topics t; see Chuang et. al (2012)"); d3.select("#" + barFreqsID) .append("a") .attr("xlink:href", "http://nlp.stanford.edu/events/illvi2014/papers/sievert-illvi2014.pdf") .attr("target", "_blank") .append("text") .attr("x", 0) .attr("y", mdsheight + 10 + (8/2)*barguide.height + 5) .style("dominant-baseline", "middle") .text("2. relevance(term w | topic t) = \u03BB * p(w | t) + (1 - \u03BB) * p(w | t)/p(w); see Sievert & Shirley (2014)"); // Bind 'default' data to 'default' bar chart var basebars = chart.selectAll(to_select + " .bar-totals") .data(barDefault2) .enter(); // Draw the gray background bars defining the overall frequency of each word basebars .append("rect") .attr("class", "bar-totals") .attr("x", 0) .attr("y", function(d) { return y(d.Term); }) .attr("height", y.rangeBand()) .attr("width", function(d) { return x(d.Total); }) .style("fill", color1) .attr("opacity", 0.4); // Add word labels to the side of each bar basebars .append("text") .attr("x", -5) .attr("class", "terms") .attr("y", function(d) { return y(d.Term) + 12; }) .attr("cursor", "pointer") .attr("id", function(d) { return (termID + d.Term); }) .style("text-anchor", "end") // right align text - use 'middle' for center alignment .text(function(d) { return d.Term; }) .on("mouseover", function() { term_hover(this); }) // .on("click", function(d) { // var old_term = termID + vis_state.term; // if (vis_state.term != "" && old_term != this.id) { // term_off(document.getElementById(old_term)); // } // vis_state.term = d.Term; // state_save(true); // term_on(this); // debugger; // }) .on("mouseout", function() { vis_state.term = ""; term_off(this); state_save(true); }); var title = chart.append("text") .attr("x", barwidth/2) .attr("y", -30) .attr("class", "bubble-tool") // set class so we can remove it when highlight_off is called .style("text-anchor", "middle") .style("font-size", "larger") .style("text-anchor", "middle") .text("Top-" + R + " Most Salient Terms"); title.append("tspan") .attr("baseline-shift", "super") .attr("font-size", "smaller") .text("(1)"); // barchart axis adapted from http://bl.ocks.org/mbostock/1166403 var xAxis = d3.svg.axis().scale(x) .orient("top") .tickSize(-barheight) .tickSubdivide(true) .ticks(6); chart.attr("class", "xaxis") .call(xAxis); // dynamically create the topic and lambda input forms at the top of the page: function init_forms(topicID, lambdaID, visID) { // create container div for topic and lambda input: var inputDiv = document.createElement("div"); inputDiv.setAttribute("id", topID); inputDiv.setAttribute("style", "width: 100%"); document.getElementById(visID).appendChild(inputDiv); // topic input container: var topicDiv = document.createElement("div"); topicDiv.setAttribute("style", "padding: 5px; background-color: #e8e8e8; display: inline-block; width: " + mdswidth + "px; height: 50px; float: left"); inputDiv.appendChild(topicDiv); var topicLabel = document.createElement("label"); topicLabel.setAttribute("for", topicID); topicLabel.setAttribute("style", "font-family: sans-serif;"); topicLabel.innerHTML = "Selected Topic: "; topicDiv.appendChild(topicLabel); var topicInput = document.createElement("input"); topicInput.setAttribute("style", "width: 20px"); topicInput.type = "text"; topicInput.min = "0"; topicInput.max = K; // assumes the data has already been read in topicInput.step = "1"; topicInput.value = "0"; // a value of 0 indicates no topic is selected topicInput.id = topicID; topicDiv.appendChild(topicInput); var previous = document.createElement("button"); previous.setAttribute("id", topicDown); previous.setAttribute("style", "margin-left: 5px"); previous.innerHTML = "Previous Topic"; topicDiv.appendChild(previous); var next = document.createElement("button"); next.setAttribute("id", topicUp); next.setAttribute("style", "margin-left: 5px"); next.innerHTML = "Next Topic"; topicDiv.appendChild(next); var clear = document.createElement("button"); clear.setAttribute("id", topicClear); clear.setAttribute("style", "margin-left: 5px"); clear.innerHTML = "Clear Topic"; topicDiv.appendChild(clear); // lambda inputs //var lambdaDivLeft = 8 + mdswidth + margin.left + termwidth; var lambdaDivWidth = barwidth; var lambdaDiv = document.createElement("div"); lambdaDiv.setAttribute("id", lambdaInputID); lambdaDiv.setAttribute("style", "padding: 5px; background-color: #e8e8e8; display: inline-block; height: 50px; width: " + lambdaDivWidth + "px; float: right;"); inputDiv.appendChild(lambdaDiv); var lambdaZero = document.createElement("div"); lambdaZero.setAttribute("style", "padding: 5px; height: 20px; width: 100%; font-family: sans-serif; float: left"); lambdaZero.setAttribute("id", lambdaZeroID); lambdaDiv.appendChild(lambdaZero); var xx = d3.select("#" + lambdaZeroID) .append("text") .attr("x", 0) .attr("y", 0) .text("Slide to adjust relevance metric:"); var yy = d3.select("#" + lambdaZeroID) .append("text") .attr("x", 125) .attr("y", -5) .style("font-size", "smaller") .style("position", "absolute") .text("(2)"); var sliderDiv = document.createElement("div"); sliderDiv.setAttribute("id", sliderDivID); sliderDiv.setAttribute("style", "width: 70%; float: right; margin-top: -5px;"); lambdaDiv.appendChild(sliderDiv); var lambdaInput = document.createElement("input"); lambdaInput.setAttribute("style", "width: 100%; margin-left: 0px; margin-right: 0px"); lambdaInput.type = "range"; lambdaInput.min = 0; lambdaInput.max = 1; lambdaInput.step = data['lambda.step']; lambdaInput.value = vis_state.lambda; lambdaInput.id = lambdaID; lambdaInput.setAttribute("list", "ticks"); // to enable automatic ticks (with no labels, see below) sliderDiv.appendChild(lambdaInput); var lambdaLabel = document.createElement("label"); lambdaLabel.setAttribute("id", lambdaLabelID); lambdaLabel.setAttribute("for", lambdaID); lambdaLabel.setAttribute("style", "width: 20%; font-family: sans-serif; padding:5px"); lambdaLabel.innerHTML = "λ = " + vis_state.lambda + ""; lambdaDiv.appendChild(lambdaLabel); /* was: slider scale thru svg; does not scale // Create the svg to contain the slider scale: var scaleContainer = d3.select("#" + sliderDivID).append("svg") .attr("width", 250) .attr("height", 25); var sliderScale = d3.scale.linear() .domain([0, 1]) .range([7.5, 242.5]) // trimmed by 7.5px on each side to match the input type=range slider: .nice(); // adapted from http://bl.ocks.org/mbostock/1166403 var sliderAxis = d3.svg.axis() .scale(sliderScale) .orient("bottom") .tickSize(10) .tickSubdivide(true) .ticks(6); // group to contain the elements of the slider axis: var sliderAxisGroup = scaleContainer.append("g") .attr("class", "slideraxis") .attr("margin-top", "-10px") .call(sliderAxis); // Another strategy for tick marks on the slider; simpler, but not labels // var sliderTicks = document.createElement("datalist"); // sliderTicks.setAttribute("id", "ticks"); // for (var tick = 0; tick <= 10; tick++) { // var tickOption = document.createElement("option"); // //tickOption.value = tick/10; // tickOption.innerHTML = tick/10; // sliderTicks.appendChild(tickOption); // } // append the forms to the containers //lambdaDiv.appendChild(sliderTicks); */ } // function to re-order the bars (gray and red), and terms: function reorder_bars(increase) { // grab the bar-chart data for this topic only: var dat2 = lamData.filter(function(d) { //return d.Category == "Topic" + Math.min(K, Math.max(0, vis_state.topic)) // fails for negative topic numbers... return d.Category == "Topic" + vis_state.topic; }); // define relevance: for (var i = 0; i < dat2.length; i++) { dat2[i].relevance = vis_state.lambda * dat2[i].logprob + (1 - vis_state.lambda) * dat2[i].loglift; } // sort by relevance: dat2.sort(fancysort("relevance")); // truncate to the top R tokens: var dat3 = dat2.slice(0, R); var y = d3.scale.ordinal() .domain(dat3.map(function(d) { return d.Term; })) .rangeRoundBands([0, barheight], 0.15); var x = d3.scale.linear() .domain([1, d3.max(dat3, function(d) { return d.Total; })]) .range([0, barwidth]) .nice(); // Change Total Frequency bars var graybars = d3.select("#" + barFreqsID) .selectAll(to_select + " .bar-totals") .data(dat3, function(d) { return d.Term; }); // Change word labels var labels = d3.select("#" + barFreqsID) .selectAll(to_select + " .terms") .data(dat3, function(d) { return d.Term; }); // Create red bars (drawn over the gray ones) to signify the frequency under the selected topic var redbars = d3.select("#" + barFreqsID) .selectAll(to_select + " .overlay") .data(dat3, function(d) { return d.Term; }); // adapted from http://bl.ocks.org/mbostock/1166403 var xAxis = d3.svg.axis().scale(x) .orient("top") .tickSize(-barheight) .tickSubdivide(true) .ticks(6); // New axis definition: var newaxis = d3.selectAll(to_select + " .xaxis"); // define the new elements to enter: var graybarsEnter = graybars.enter().append("rect") .attr("class", "bar-totals") .attr("x", 0) .attr("y", function(d) { return y(d.Term) + barheight + margin.bottom + 2 * rMax; }) .attr("height", y.rangeBand()) .style("fill", color1) .attr("opacity", 0.4); var labelsEnter = labels.enter() .append("text") .attr("x", -5) .attr("class", "terms") .attr("y", function(d) { return y(d.Term) + 12 + barheight + margin.bottom + 2 * rMax; }) .attr("cursor", "pointer") .style("text-anchor", "end") .attr("id", function(d) { return (termID + d.Term); }) .text(function(d) { return d.Term; }) .on("mouseover", function() { term_hover(this); }) // .on("click", function(d) { // var old_term = termID + vis_state.term; // if (vis_state.term != "" && old_term != this.id) { // term_off(document.getElementById(old_term)); // } // vis_state.term = d.Term; // state_save(true); // term_on(this); // }) .on("mouseout", function() { vis_state.term = ""; term_off(this); state_save(true); }); var redbarsEnter = redbars.enter().append("rect") .attr("class", "overlay") .attr("x", 0) .attr("y", function(d) { return y(d.Term) + barheight + margin.bottom + 2 * rMax; }) .attr("height", y.rangeBand()) .style("fill", color2) .attr("opacity", 0.8); if (increase) { graybarsEnter .attr("width", function(d) { return x(d.Total); }) .transition().duration(duration) .delay(duration) .attr("y", function(d) { return y(d.Term); }); labelsEnter .transition().duration(duration) .delay(duration) .attr("y", function(d) { return y(d.Term) + 12; }); redbarsEnter .attr("width", function(d) { return x(d.Freq); }) .transition().duration(duration) .delay(duration) .attr("y", function(d) { return y(d.Term); }); graybars.transition().duration(duration) .attr("width", function(d) { return x(d.Total); }) .transition().duration(duration) .attr("y", function(d) { return y(d.Term); }); labels.transition().duration(duration) .delay(duration) .attr("y", function(d) { return y(d.Term) + 12; }); redbars.transition().duration(duration) .attr("width", function(d) { return x(d.Freq); }) .transition().duration(duration) .attr("y", function(d) { return y(d.Term); }); // Transition exiting rectangles to the bottom of the barchart: graybars.exit() .transition().duration(duration) .attr("width", function(d) { return x(d.Total); }) .transition().duration(duration) .attr("y", function(d, i) { return barheight + margin.bottom + 6 + i * 18; }) .remove(); labels.exit() .transition().duration(duration) .delay(duration) .attr("y", function(d, i) { return barheight + margin.bottom + 18 + i * 18; }) .remove(); redbars.exit() .transition().duration(duration) .attr("width", function(d) { return x(d.Freq); }) .transition().duration(duration) .attr("y", function(d, i) { return barheight + margin.bottom + 6 + i * 18; }) .remove(); // https://github.com/mbostock/d3/wiki/Transitions#wiki-d3_ease newaxis.transition().duration(duration) .call(xAxis) .transition().duration(duration); } else { graybarsEnter .attr("width", 100) // FIXME by looking up old width of these bars .transition().duration(duration) .attr("y", function(d) { return y(d.Term); }) .transition().duration(duration) .attr("width", function(d) { return x(d.Total); }); labelsEnter .transition().duration(duration) .attr("y", function(d) { return y(d.Term) + 12; }); redbarsEnter .attr("width", 50) // FIXME by looking up old width of these bars .transition().duration(duration) .attr("y", function(d) { return y(d.Term); }) .transition().duration(duration) .attr("width", function(d) { return x(d.Freq); }); graybars.transition().duration(duration) .attr("y", function(d) { return y(d.Term); }) .transition().duration(duration) .attr("width", function(d) { return x(d.Total); }); labels.transition().duration(duration) .attr("y", function(d) { return y(d.Term) + 12; }); redbars.transition().duration(duration) .attr("y", function(d) { return y(d.Term); }) .transition().duration(duration) .attr("width", function(d) { return x(d.Freq); }); // Transition exiting rectangles to the bottom of the barchart: graybars.exit() .transition().duration(duration) .attr("y", function(d, i) { return barheight + margin.bottom + 6 + i * 18 + 2 * rMax; }) .remove(); labels.exit() .transition().duration(duration) .attr("y", function(d, i) { return barheight + margin.bottom + 18 + i * 18 + 2 * rMax; }) .remove(); redbars.exit() .transition().duration(duration) .attr("y", function(d, i) { return barheight + margin.bottom + 6 + i * 18 + 2 * rMax; }) .remove(); // https://github.com/mbostock/d3/wiki/Transitions#wiki-d3_ease newaxis.transition().duration(duration) .transition().duration(duration) .call(xAxis); } } ////////////////////////////////////////////////////////////////////////////// // function to update bar chart when a topic is selected // the circle argument should be the appropriate circle element function topic_on(circle) { if (circle == null) return null; // grab data bound to this element var d = circle.__data__; var Freq = Math.round(d.Freq * 10) / 10, topics = d.topics; // change opacity and fill of the selected circle circle.style.opacity = highlight_opacity; circle.style.fill = color2; // Remove 'old' bar chart title var text = d3.select(to_select + " .bubble-tool"); text.remove(); // append text with info relevant to topic of interest d3.select("#" + barFreqsID) .append("text") .attr("x", barwidth/2) .attr("y", -30) .attr("class", "bubble-tool") // set class so we can remove it when highlight_off is called .style("text-anchor", "middle") .style("font-size", "larger") .text("Top-" + R + " Most Relevant Terms for Topic " + topics + " (" + Freq + "% of tokens)"); // grab the bar-chart data for this topic only: var dat2 = lamData.filter(function(d) { return d.Category == "Topic" + topics; }); // define relevance: for (var i = 0; i < dat2.length; i++) { dat2[i].relevance = lambda.current * dat2[i].logprob + (1 - lambda.current) * dat2[i].loglift; } // sort by relevance: dat2.sort(fancysort("relevance")); // truncate to the top R tokens: var dat3 = dat2.slice(0, R); // scale the bars to the top R terms: var y = d3.scale.ordinal() .domain(dat3.map(function(d) { return d.Term; })) .rangeRoundBands([0, barheight], 0.15); var x = d3.scale.linear() .domain([1, d3.max(dat3, function(d) { return d.Total; })]) .range([0, barwidth]) .nice(); // remove the red bars if there are any: d3.selectAll(to_select + " .overlay").remove(); // Change Total Frequency bars d3.selectAll(to_select + " .bar-totals") .data(dat3) .attr("x", 0) .attr("y", function(d) { return y(d.Term); }) .attr("height", y.rangeBand()) .attr("width", function(d) { return x(d.Total); }) .style("fill", color1) .attr("opacity", 0.4); // Change word labels d3.selectAll(to_select + " .terms") .data(dat3) .attr("x", -5) .attr("y", function(d) { return y(d.Term) + 12; }) .attr("id", function(d) { return (termID + d.Term); }) .style("text-anchor", "end") // right align text - use 'middle' for center alignment .text(function(d) { return d.Term; }); // Create red bars (drawn over the gray ones) to signify the frequency under the selected topic d3.select("#" + barFreqsID).selectAll(to_select + " .overlay") .data(dat3) .enter() .append("rect") .attr("class", "overlay") .attr("x", 0) .attr("y", function(d) { return y(d.Term); }) .attr("height", y.rangeBand()) .attr("width", function(d) { return x(d.Freq); }) .style("fill", color2) .attr("opacity", 0.8); // adapted from http://bl.ocks.org/mbostock/1166403 var xAxis = d3.svg.axis().scale(x) .orient("top") .tickSize(-barheight) .tickSubdivide(true) .ticks(6); // redraw x-axis d3.selectAll(to_select + " .xaxis") //.attr("class", "xaxis") .call(xAxis); } function topic_off(circle) { if (circle == null) return circle; // go back to original opacity/fill circle.style.opacity = base_opacity; circle.style.fill = color1; var title = d3.selectAll(to_select + " .bubble-tool") .text("Top-" + R + " Most Salient Terms"); title.append("tspan") .attr("baseline-shift", "super") .attr("font-size", 12) .text(1); // remove the red bars d3.selectAll(to_select + " .overlay").remove(); // go back to 'default' bar chart var dat2 = lamData.filter(function(d) { return d.Category == "Default"; }); var y = d3.scale.ordinal() .domain(dat2.map(function(d) { return d.Term; })) .rangeRoundBands([0, barheight], 0.15); var x = d3.scale.linear() .domain([1, d3.max(dat2, function(d) { return d.Total; })]) .range([0, barwidth]) .nice(); // Change Total Frequency bars d3.selectAll(to_select + " .bar-totals") .data(dat2) .attr("x", 0) .attr("y", function(d) { return y(d.Term); }) .attr("height", y.rangeBand()) .attr("width", function(d) { return x(d.Total); }) .style("fill", color1) .attr("opacity", 0.4); //Change word labels d3.selectAll(to_select + " .terms") .data(dat2) .attr("x", -5) .attr("y", function(d) { return y(d.Term) + 12; }) .style("text-anchor", "end") // right align text - use 'middle' for center alignment .text(function(d) { return d.Term; }); // adapted from http://bl.ocks.org/mbostock/1166403 var xAxis = d3.svg.axis().scale(x) .orient("top") .tickSize(-barheight) .tickSubdivide(true) .ticks(6); // redraw x-axis d3.selectAll(to_select + " .xaxis") .attr("class", "xaxis") .call(xAxis); } // event definition for mousing over a term function term_hover(term) { var old_term = termID + vis_state.term; if (vis_state.term != "" && old_term != term.id) { term_off(document.getElementById(old_term)); } vis_state.term = term.innerHTML; term_on(term); state_save(true); } // updates vis when a term is selected via click or hover function term_on(term) { if (term == null) return null; term.style["fontWeight"] = "bold"; var d = term.__data__; var Term = d.Term; var dat2 = mdsData3.filter(function(d2) { return d2.Term == Term; }); var k = dat2.length; // number of topics for this token with non-zero frequency var radius = []; for (var i = 0; i < K; ++i) { radius[i] = 0; } for (i = 0; i < k; i++) { radius[dat2[i].Topic - 1] = dat2[i].Freq; } var size = []; for (var i = 0; i < K; ++i) { size[i] = 0; } for (i = 0; i < k; i++) { // If we want to also re-size the topic number labels, do it here // 11 is the default, so leaving this as 11 won't change anything. size[dat2[i].Topic - 1] = 11; } var rScaleCond = d3.scale.sqrt() .domain([0, 1]).range([0, rMax]); // Change size of bubbles according to the word's distribution over topics d3.selectAll(to_select + " .dot") .data(radius) .transition() .attr("r", function(d) { //return (rScaleCond(d)); return (Math.sqrt(d*mdswidth*mdsheight*word_prop/Math.PI)); }); // re-bind mdsData so we can handle multiple selection d3.selectAll(to_select + " .dot") .data(mdsData); // Change sizes of topic numbers: d3.selectAll(to_select + " .txt") .data(size) .transition() .style("font-size", function(d) { return +d; }); // Alter the guide d3.select(to_select + " .circleGuideTitle") .text("Conditional topic distribution given term = '" + term.innerHTML + "'"); } function term_off(term) { if (term == null) return null; term.style["fontWeight"] = "normal"; d3.selectAll(to_select + " .dot") .data(mdsData) .transition() .attr("r", function(d) { //return (rScaleMargin(+d.Freq)); return (Math.sqrt((d.Freq/100)*mdswidth*mdsheight*circle_prop/Math.PI)); }); // Change sizes of topic numbers: d3.selectAll(to_select + " .txt") .transition() .style("font-size", "11px") .transition() .style("font-size", null); // Go back to the default guide d3.select(to_select + " .circleGuideTitle") .text("Marginal topic distribution"); d3.select(to_select + " .circleGuideLabelLarge") .text(defaultLabelLarge); d3.select(to_select + " .circleGuideLabelSmall") .attr("y", mdsheight + 2 * newSmall) .text(defaultLabelSmall); d3.select(to_select + " .circleGuideSmall") .attr("r", newSmall) .attr("cy", mdsheight + newSmall); d3.select(to_select + " .lineGuideSmall") .attr("y1", mdsheight + 2 * newSmall) .attr("y2", mdsheight + 2 * newSmall); } // serialize the visualization state using fragment identifiers -- http://en.wikipedia.org/wiki/Fragment_identifier // location.hash holds the address information var params = location.hash.split("&"); if (params.length > 1) { vis_state.topic = params[0].split("=")[1]; vis_state.lambda = params[1].split("=")[1]; vis_state.term = params[2].split("=")[1]; // Idea: write a function to parse the URL string // only accept values in [0,1] for lambda, {0, 1, ..., K} for topics (any string is OK for term) // Allow for subsets of the three to be entered: // (1) topic only (lambda = 1 term = "") // (2) lambda only (topic = 0 term = "") visually the same but upon hovering a topic, the effect of lambda will be seen // (3) term only (topic = 0 lambda = 1) only fires when the term is among the R most salient // (4) topic + lambda (term = "") // (5) topic + term (lambda = 1) // (6) lambda + term (topic = 0) visually lambda doesn't make a difference unless a topic is hovered // (7) topic + lambda + term // Short-term: assume format of "#topic=k&lambda=l&term=s" where k, l, and s are strings (b/c they're from a URL) // Force k (topic identifier) to be an integer between 0 and K: vis_state.topic = Math.round(Math.min(K, Math.max(0, vis_state.topic))); // Force l (lambda identifier) to be in [0, 1]: vis_state.lambda = Math.min(1, Math.max(0, vis_state.lambda)); // impose the value of lambda: document.getElementById(lambdaID).value = vis_state.lambda; document.getElementById(lambdaID + "-value").innerHTML = vis_state.lambda; // select the topic and transition the order of the bars (if approporiate) if (!isNaN(vis_state.topic)) { document.getElementById(topicID).value = vis_state.topic; if (vis_state.topic > 0) { topic_on(document.getElementById(topicID + vis_state.topic)); } if (vis_state.lambda < 1 && vis_state.topic > 0) { reorder_bars(false); } } lambda.current = vis_state.lambda; var termElem = document.getElementById(termID + vis_state.term); if (termElem !== undefined) term_on(termElem); } function state_url() { return location.origin + location.pathname + "#topic=" + vis_state.topic + "&lambda=" + vis_state.lambda + "&term=" + vis_state.term; } function state_save(replace) { if (replace) history.replaceState(vis_state, "Query", state_url()); else history.pushState(vis_state, "Query", state_url()); } function state_reset() { if (vis_state.topic > 0) { topic_off(document.getElementById(topicID + vis_state.topic)); } if (vis_state.term != "") { term_off(document.getElementById(termID + vis_state.term)); } vis_state.term = ""; document.getElementById(topicID).value = vis_state.topic = 0; state_save(true); } } if (typeof data_or_file_name === 'string') d3.json(data_or_file_name, function(error, data) {visualize(data);}); else visualize(data_or_file_name); // var current_clicked = { // what: "nothing", // element: undefined // }, //debugger; }; return LDAvis; }