[dygraphs.git] / docs / data.html

<!--#include virtual="header.html" -->

<style type="text/css">
  code { white-space: pre; border: 1px dashed black; display: block; }
  pre  { white-space: pre; border: 1px dashed black; }
</style>

<h2>dygraphs Data Format</h2>

<p>When you create a Dygraph object, your code looks something like
this:</p>

<code>
  g = new Dygraph(document.getElementById("div"),
                  <i>data</i>,
                  { <i>options</i> });

</code>

<p>This document is about what you can put in the <i>data</i>
parameter.</p>

<p>There are five types of input that dygraphs will accept:</p>
<ol>
  <li><a href="#csv">CSV data</a>
  <li><a href="#url">URL</a>
  <li><a href="#array">array (native format)</a>
  <li><a href="#function">function</a>
  <li><a href="#datatable">DataTable</a>
</ol>

<p>These are all discussed below. If you're trying to debug why your input
won't parse, <b>check the JS error console</b>. dygraphs tries to log
informative errors explaining what's wrong with your data, and these can
often point you in the right direction.</p>

<p>There are several options which affect how your input data is
interpreted. These are:</p>
<ul>
  <li> <i>xValueParser</i> affects CSV only.
  <li> <i>errorBars</i> affects all input types.
  <li> <i>customBars</i> affects all input types.
  <li> <i>fractions</i> affects all input types.
  <li> <i>labels</i> affects all input types.
</ul>

<a name="csv"></a>
  <h3>CSV</h3>
<p>Here's an example of what CSV data should look like:</p>
<pre>
Date,Series1,Series2
2009/07/12,100,200  # comments are OK on data lines
2009/07/19,150,201
</pre>

<p>"CSV" is actually a bit of a misnomer: the data can be tab-delimited,
too. The delimiter is set by the <i>delimiter</i> option. It default to ",".
If no delimiter is found in the first row, it switches over to tab.</p>

<p>CSV parsing can be split into three parts: headers, x-value and
y-values.</p>

<h4>Headers</h4>
<p>If you don't specify the <i>labels</i> option, dygraphs will look at the
first line of your CSV data to get the labels. If you see numbers for series
labels when you hover over the dygraph, it's likely because your first line
contains data but is being parsed as a label. The solution is to either add
a header line or specify the labels like this:</p>

<code>
  new Dygraph(el,
              "2009/07/12,100,200\n" +
              "2009/07/19,150,201\n",
              { labels: [ "Date", "Series1", "Series2" ] });
</code>

<h4>x-values</h4>
<p>Once the headers are parsed, dygraphs needs to determine what the type of
the x values is. They're either dates or numbers. To make this
determination, it looks at the first column of the first row ("2009/07/12"
in the example above). Here's the heuristic: if it contains a '-' or a '/',
or otherwise doesn't parse as a float, the it's a date. Otherwise, it's a
number.</p>

<p>Once the type is determined, that doesn't mean all the values will parse
correctly. The general rule is:<p>

<ul>
  <li>For dates, your strings have to be parseable by <i>Date.parse</i>.
  <li>For numbers, your strings have to be parseable by <i>parseFloat</i>.
</ul>

<p>You can manually verify this using a JavaScript console. If a value
doesn't parse, dygraphs will put a warning about it on your console. But
beware: different browsers support different date formats!</p>

<p>Here are some valid date formats:</p>
<ul>
  <li>2009-07-12</li>
  <li>2009/07/12</li>
  <li>2009/07/12 12</li>
  <li>2009/07/12 12:34</li>
  <li>2009/07/12 12:34:56</li>
</ul>

<p>If you specify the <i>xValueParser</i> option, then all this detection is
bypassed and your function is called instead. Your parser function takes in
a string and needs to return a number. For dates/times, you should return
milliseconds since epoch. You may also want to specify a few other options
to make sure that everything gets displayed properly.<p>

<p>Here's code which parses a CSV file with unix timestamps in the first
column:</p>

<code>
  new Dygraph(el,
              "Date,Series1,Series2\n" +
              "1247382000,100,200\n" +
              "1247986800,150,201\n",
              {
                xValueFormatter: Dygraph.dateString_,
                xValueParser: function(x) { return 1000*parseInt(x); },
                xTicker: Dygraph.dateTicker
              });
</code>

<h4>y-values</h4>
<p>Dependent (y-axis) values are simpler than x-values because they're
always numbers. The complexity here comes from the various ways that you can
specify the uncertainty in your measurements.<p>

<p>If your y-values are just numbers, then they need to be parseable by
JavaScript's parseFloat function. Acceptable formats include:</p>

<ul>
  <li>12
  <li>-12
  <li>12.
  <li>12.3
  <li>1.24e+1
  <li>-1.24e+1
</ul>

<p>If you have missing data, just leave the column blank (your CSV file will
probably contain a ",," in it).</p>

<p>If your numbers have uncertainty associated with them, then there are
three basic ways to express this: using fractions, standard deviations or
explicit ranges.</p>

<h5>Fractions</h5>
<p>If you specify the <i>fractions</i> option, then your data will all be
interpreted as ratios between zero and one. This is often the case if you're
plotting a percentage.</p>

<code>
  new Dygraph(el,
              "X,Frac1,Frac2\n" +
              "1,1/2,3/4\n"+
              "2,1/3,2/3\n"+
              "3,2/3,17/49\n"+
              "4,25/30,100/200",
              { fractions: true });
</code>

<p>Why not just divide the fractions out yourself? There are two attractive
reasons not to:</p>

<ul>
  <li>If you set both <i>fractions</i> and <i>errorBars</i>, then the
denominator is interpreted as a sample size and dygraphs will plot <a
  href="http://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval">Wilson
  binomial proportion confidence intervals</a> around each point.

  <li>If you set <i>showRoller</i>, then dygraphs will combine the values as
  fractions. If two point are <i>a/b</i> and <i>c/d</i>, it will plot
  <i>(a+b) / (c+d)</i> rather than <i>(a/b + c/d) / 2</i>, which is what
  you'd get if you divided the fractions through. This will also shrink the
  confidence intervals.</li>
</ul>

<h5>Standard Deviations</h5>
<p>Often you have a measurement and also a measure of its uncertainty: a
standard deviation. If you specify the <i>errorBars</i> option, dygraphs
will look for alternating value and standard deviation columns in your CSV
data.  Here's what it should look like:</p>

<code>
  new Dygraph(el,
              "X,Y1,Y2\n" +
              "1,10,5,20,5\n" +
              "2,12,5,22,5\n",
              { errorBars: true });
</code>

<p>The "5" values are standard deviations. When each point is plotted, a
2-standard deviation region around it is shaded, resulting in a 95%
confidence interval. If you want more or less confidence, you can set the
<i>sigma</i> option to something other than 2.0.</p>

<p>When you roll data with standard deviations, dygraphs will plot the
average of your values in each rolling period and the RMS value of your
standard deviations: sqrt(std1 + std2 + std3 + ... + stdN)/N.</p>

<h5>Custom error bars</h5>
<p>Sometimes your data has asymetric uncertainty or you want to specify
something else with the error bars around a point. One example of this is
the "temperatures" demo on the <a href="http://danvk.org/dygraphs">dygraphs
  home page.</a>, where the point is the daily average and the bars denote
the low and high temperatures for the day.</p>

<p>To specify this format, set the <i>customBars</i> option. Your CSV values
should each be three numbers separated by semicolons ("low;mid;high").
Here's an example:</p>

<code>
  new Dygraph(el,
              "X,Y1,Y2\n" +
              "1,10;20;30,20;5;25\n" +
              "2,10;25;35,20;10;25\n",
              { customBars: true });
</code>

<p>The middle value need not lie between the low and high values. If you set
a rolling period, the three values will all be averaged independently.</p>


<a name="url"></a>
<h3>URL</h3>
<p>If you pass in a URL, dygraphs will issue an XMLHttpRequest for it and
attempt to parse the returned data as CSV.
</p>

<p><i>Common problems</i>. Make sure the URL is accessible and returns data
in text format (as opposed to a CSV file with an HTML header). You can see
what the response looks like by checking your JS console or by requesting
the URL yourself.</p>


<a name="array"></a>
<h3>Array (native format)</h3>
<p>If you'll be constructing your data set from a server-side program (or
from JavaScript) then you're better off producing an array than CSV data.
This saves the cost of parsing the CSV data and also avoids common parser
errors.</p>

<p>The downside is that it's harder to look at your data (you'll need to use
a JS debugger) and that the data format is a bit less clear for values with
uncertainties.</p>

<p>Here's an example of "native format":</p>

<code>
  new Dygraph(document.getElementById("graphdiv2"),
              [
                [1,10,100],
                [2,20,80],
                [3,50,60],
                [4,70,80]
              ],
              {
                labels: [ "x", "A", "B" ]
              });
</code>

<h4>Headers</h4>
<p>Headers for native format must be specified via the <i>labels</i>
option. There's no other way to set them.</p>

<h4>x-values</h4>
<p>If you want your x-values to be dates, you'll need to use specify a Date
object in the first column. Otherwise, specify a number. Here's a sample
array with dates on the x-axis:</p>

<code>
  [
    [ new Date("2009/07/12"), 100, 200 ],
    [ new Date("2009/07/19"), 150, 220 ]
  ]
</code>

<h4>y-values</h4>
<p>You can specify <i>errorBars</i>, <i>fractions</i> or <i>customBars</i>
with the array format. If you specify any of these, the values become arrays
(rather than numbers). Here's what the format looks like for each one:</p>

<code>
  <i>errorBars</i>: [x, [value1, std1], [value2, std2], ...]
  <i>fractions</i>: [x, [num1, den1], [num2, den2], ...]
  <i>customBars</i>: [x, [low1, val1, high1], [low2, val2, high2], ...]
</code>

<p>To specify missing data, set the value to null or NaN. You may not set a value
inside an array to null or NaN. Use null or NaN instead of the entire array. 
The only difference between the two is when the option
<a href="options.html#conectSeparatedPoints">connectSeparatedPoints</a>
true. In that case, the gaps created by nulls are filled in, and gaps
created by NaNs are preserved.
</p>

<a name="function"></a>
<h3>Functions</h3>

<p>You can specify a function that returns any of the other types. If
<i>x</i> is a valid piece of dygraphs input, then so is</p>

<code>
  function() { return x; }
</code>

Functions can return strings, arrays, data tables, URLs, or any other data type.

<a name="datatable"></a>
<h3>DataTable</h3>
<p>You can also specify a Google Visualization Library <a
  href="http://code.google.com/apis/visualization/documentation/reference.html#DataTable">DataTable</a>
object as your input data. This lets you easily switch between dygraphs and
other gviz visualizations such as the Annotated Timeline. It also lets you
embed a Dygraph in a Google Spreadsheet.</p>

<p>You'll need to set your first column's type to one of "number", "date"
or "datetime".</p>

<pre>
DataTable TODO:
- When to use Dygraph.GvizWrapper
- how to specify fractions
- how to specify missing data
- how to specify value + std. dev.
- how to specify [low, middle, high]
- walkthrough of embedding a gadget in google docs/on a web page
- walkthrough of using std. dev. in a spreadsheet chart
</pre>

<!--#include virtual="footer.html" -->