dygraphs
A demo is worth a thousand words:
(Mouse over to highlight individual values. Click and drag to zoom. Double-click to zoom back out. Change the number and hit enter to adjust the averaging period.)
Some things to notice:
- There's less seasonal temperature variation in SF than in NY.
- The difference is about 15° F for SF vs. 50° F for NY.
- The daily data (set rolling period to 1) is quite noisy and hides this conclusion.
- Using a 14-day moving average makes it clearer. A 100-day rolling period averages out nearly all the specifics from the data.
- There's a gap in the data for SF, when the weather station was down (zoom into October 2007 to see it).
- The bands around each point indicate average highs and lows.
- There is a lot of data in this chart: low, average and high for each city on each day of a three year period ≈ 6000 data points in all.
dygraphs allows the user to explore the data and discover these facts.
For more demos, browse the dygraph tests directory.
Features
Some of the features of dygraphs:
- Plots time series without using an external server or Flash
- Works in Internet Explorer (using excanvas)
- Lightweight (69kb) and responsive
- Displays values on mouseover, making interaction easily discoverable
- Supports error bands around data series
- Interactive zoom
- Displays Annotations on the chart
- Adjustable averaging period
- Can intelligently chart fractions
- Customizable click-through actions
- Compatible with the Google Visualization API
- Intelligent defaults make it easy to use
Usage
To use dygraphs, include the dygraph-combined.js
JavaScript file and instantiate a Dygraph
object.
Here's a basic example to get things started:
HTML
<html>
<head>
<script type="text/javascript"
src="dygraph-combined.js"></script>
</head>
<body>
<div id="graphdiv"></div>
<script type="text/javascript">
g = new Dygraph(
// containing div
document.getElementById("graphdiv"),
// CSV or path to a CSV file.
"Date,Temperature\n" +
"2008-05-07,75\n" +
"2008-05-08,70\n" +
"2008-05-09,80\n"
);
</script>
</body>
</html>
OUTPUT
In order to keep this example self-contained, the second parameter is raw CSV data. The dygraphs library parses this data (including column headers), resizes its container to a reasonable default, calculates appropriate axis ranges and tick marks and draws the graph.
In most applications, it makes more sense to include a CSV file instead. If the second parameter to the constructor doesn't contain a newline, it will be interpreted as the path to a CSV file. The Dygraph will perform an XMLHttpRequest to retrieve this file and display the data when it becomes available. Make sure your CSV file is readable and serving from a place that understands XMLHttpRequest's! In particular, you cannot specify a CSV file using "file:///"
. Here's an example: (data from Weather Underground)
HTML
<html>
<head>
<script type="text/javascript"
src="dygraph-combined.js"></script>
</head>
<body>
<div id="graphdiv2"
style="width:500px; height:300px;"></div>
<script type="text/javascript">
g2 = new Dygraph(
document.getElementById("graphdiv2"),
"temperatures.csv", // path to CSV file
{} // options
);
</script>
</body>
</html>
OUTPUT
The file used is temperatures.csv
.
There are a few things to note here:
- The Dygraph sent off an XHR to get the temperatures.csv file.
- The labels were taken from the first line of
temperatures.csv
, which isDate,High,Low
. - The Dygraph automatically chose two different, easily-distinguishable colors for the two data series.
- The labels on the x-axis have switched from days to months. If you zoom in, they'll switch to weeks and then days.
- Some heuristics are used to determine a good vertical range for the data. The idea is to make all the data visible and have human-friendly values on the axis (i.e. 200 instead of 193.4). Generally this works well.
- The data is very spiky. A moving average would be easier to interpret.
This problem can be fixed by specifying the appropriate options in the "additional options" parameter to the Dygraph constructor. To set the number of days for a moving average, use the rollPeriod
option. Here's how it's done:
HTML
<html>
<head>
<script type="text/javascript"
src="dygraph-combined.js"></script>
</head>
<body>
<div id="graphdiv3"
style="width:500px; height:300px;"></div>
<script type="text/javascript">
g3 = new Dygraph(
document.getElementById("graphdiv3"),
"temperatures.csv",
{
rollPeriod: 7,
showRoller: true
}
);
</script>
</body>
</html>
OUTPUT
A rolling average can be set using the text box in the lower left-hand corner of the graph (the showRoller attribute is what makes this appear). Also note that we've explicitly set the size of the chart div.
Error Bars
Another significant feature of the dygraphs library is the ability to display error bars around data series. One standard deviation must be specified for each data point. A ±n sigma band will be drawn around the data series at that point. If a moving average is being displayed, dygraphs will compute the standard deviation of the average at each point. I.E. σ = sqrt( (σ12 + σ22 + ... + σn2) / n )
Here's a demonstration. There are two data series. One is N(100,10)
with a standard deviation of 10 specified at each point. The other is N(80,20)
with a standard deviation of 20 specified at each point. The CSV file was generated using Octave and can be viewed at twonormals.csv.
HTML
<html>
<head>
<script type="text/javascript"
src="combined.js"></script>
</head>
<body>
<div id="graphdiv4"
style="width:480px; height:320px;"></div>
<script type="text/javascript">
g4 = new Dygraph(
document.getElementById("graphdiv4"),
"twonormals.csv",
{
rollPeriod: 7,
showRoller: true,
errorBars: true,
valueRange: [50,125]
}
);
</script>
</body>
</html>
OUTPUT
Things to note here:
- The errorBars option affects both the interpretation of the CSV file and the display of the graph. When errorBars is set to true, each line is interpreted as YYYYMMDD, A, sigma_A, B, sigma_B, …
- The first line of the CSV file doesn't mention the error columns. In this case, it's just "Date,Series1,Series2".
- The averaging visibly affects the error bars. This is most clear if you crank up the rolling period to something like 100 days. For the earliest dates, there won't be 100 data points to average so the signal will be noisier. The error bars get smaller like sqrt(N) going forward in time until there's a full 100 points to average.
- The error bars are partially transparent. This can be seen when they overlap one another.
Internet Explorer Compatibility
The dygraphs library relies heavily on the HTML5 <canvas>
tag, which Microsoft Internet Explorer did not traditionally support. To use Microsoft's native canvas implementation in IE9, you need to set an HTML5 doctype on your page:
<!DOCTYPE html>
When IE9 is in HTML5 mode, dygraphs works just like in other modern browsers.
If you want to support previous versions of Internet Explorer (IE6–IE8), you'll need to include the excanvas library, which emulates the <canvas>
tag using VML. You can add excanvas by including the following snippet:
<!DOCTYPE html> <html> <head> <meta http-equiv="X-UA-Compatible" content="IE=EmulateIE7; IE=EmulateIE9"> <!--[if IE]><script src="path/to/excanvas.js"></script><![endif]--> </head>
(This is surprisingly tricky because the HTML5 doctype breaks excanvas in IE8. See this discussion for details.)
While VML emulation sounds like it would be slow, it works well in practice for most charts.
One common gotcha to look out for: make sure you don't have any trailing commas in parameter lists, e.g.
new Dygraph(el, data, { showRoller: true, // note trailing comma })
Most browsers will ignore the trailing comma, but it will break under IE.
GViz Data
The Google Visualization API provides a standard interface for describing data. Once you've specified your data using this API, you can plug in any GViz-compatible visualization. dygraphs is such a visualization. In particular, it can be used as a drop-in replacement for the AnnotatedTimeline visualization used on Google Finance and other sites. To see how this works, check out the gviz annotation demo.
For a simple demonstration of how to use dygraphs a GViz visualization, see http://danvk.org/dygraphs/tests/gviz.html. dygraphs can also be used as a GViz gadget. This allows it to be embedded inside of a Google Spreadsheet. For a demonstration of this, see this spreadsheet. The URL for the gadget is http://danvk.org/dygraphs/gadget.xml
.
Here's an example of a published gviz gadget using dygraphs:
Charting Fractions
Situations often arise where you want to plot fractions, e.g. the fraction of respondents in a poll who said they'd vote for candidate X or the number of hits divided by at bats (baseball's batting average). Fractions require special treatment for two main reasons:
- The average of
a1/b1
anda2/b2
is(a1+a2)/(b1+b2)
, not(a1/b1 + a2/b2)/2
. - The normal approximation is not always applicable and more sophisticated confidence intervals (e.g. the Wilson confidence interval) must be employed to avoid ratios that exceed 100% or go below 0%.
Fortunately, dygraphs handles both of these for you! Here's a chart and the command that generated it:
Command:
new Dygraph( document.getElementById("baseballdiv"), "suzuki-mariners.txt", { fractions: true, errorBars: true, showRoller: true, rollPeriod: 15 } );
The fractions
option indicates that the values in each column should be parsed as fractions (e.g. "1/2" instead of "0.5"). The errorBars
option indicates that we'd like to see a confidence interval around each data point. By default, when fractions
is set, you get a Wilson confidence interval. If you look carefully at the chart, you can see that the error bars are asymmetric.
A couple things to notice about this chart:
- The error bars for Ichiro's batting average are larger than for the Mariners', since he has far fewer at bats than his team.
- dygraphs makes it easy to see "batting average over the last 30 games". This is ordinarily quite difficult to compute. It makes it clear where the "hot" and "cold" part of Suzuki's season were.
- If you set the averaging period to something large, like 200, you'll see the team's and player's batting average through that game. The final number is the overall batting average for the season.
- Where the error bars do not overlap, we can say with 95% confidence that the series differ. There is a better than 95% chance that Ichiro was a better hitter than his team as a whole in 2004, the year he won the batting title.
One last demo
This chart shows monthly closes of the Dow Jones Industrial Average, both in nominal and real (i.e. adjusted for inflation) dollars. The shaded areas show its monthly high and low. CPI values with a base from 1982-84 are used to adjust for inflation.
Display:
Other Options
In addition to the options mentioned above (showRoller
, rollPeriod
, errorBars
, valueRange
), there are many others.
For a full list, see the Dygraphs Options Reference page.
Common Gotchas
Here are a few problems that I've frequently run into while using the dygraphs library.
- If your chart doesn't display, be sure to check your browser's JavaScript error console. dygraphs makes every attempt to log errors and warnings, and these can often guide you in the right direction.
- Make sure your CSV files are readable! If your graph isn't showing up, the XMLHttpRequest for the CSV file may be failing. You can determine whether this is the case using tools like Firebug.
- Make sure your CSV files are in the correct format. They must be of the form
YYYYMMDD, series1, series2,
… . And if you set theerrorBars
property, make sure you alternate data series and standard deviations. - dygraphs are not happy when placed inside a
<center>
tag. This applies to the CSStext-align
property as well. If you want to center a Dygraph, put it inside a table withalign = center
set. - Don't set the
dateWindow
property to a date. It expects milliseconds since epoch, which can be obtained from a JavaScript Date object's valueOf method. - Make sure you don't have any trailing commas in your call to the Dygraph constructor or in the options parameter. Firefox, Chrome and Safari ignore these but they can cause a graph to not display in Internet Explorer.
GWT Compatibility
There is currently no GWT wrapper around Dygraphs, however there is a class that can be used to easily load Dygraphs into the browser. To use it, include the generated dygraph-gwt.jar file in your classpath and add the following line to your GWT module:
<inherits name="org.danvk.dygraphs"/>
Call org.danvk.Dygraphs.install() when your application starts to install the JavaScript code into the browser. You can use JSNI to call Dygraphs from your GWT code, as in the example below. The example uses the Visualization API for GWT and the Dygraphs GViz API.
public static native JavaScriptObject drawDygraph( Element element, DataTable dataTable, double minY, double maxY) /*-{ var chart = new $wnd.Dygraph.GVizChart(element); chart.draw(dataTable, { valueRange: [minY, maxY] }); return chart; }-*/;
Data Policy
dygraphs is purely client-side JavaScript. It does not send your data to any servers – the data is processed entirely in the client's browser.
Created May 9, 2008 by Dan Vanderkam