PNSQC Slides and Paper Are Up

My thesis defense is tomorrow which is why I haven’t posted in a couple of weeks.  If all goes well, I’ll be posting a link to that in the next week.

This is just a quick post to say that my paper and  presentation have been added to the PNSQC web-site, along with everyone else’s paper and presentation: click here and have fun exploring.

Since I don’t read from powerpoint slides, you won’t find much in the way of explanatory verbiage in the slides, but I’m happy to answer questions if there’s something you’d like me to clarify.  Ideally, I would download the paper, and read the paper while you have the slides up.  They told me to make all of the pictures for the paper extremely small and grayscale…sigh.  Kind of kills the whole visualization aspect of my paper, but I understand they had good reasons for asking.  This is exactly why Edward Tufte took out a 2nd mortgage on his house, and self-published his first book.  I would write more about that because it’s a post in and of itself, but my recursion ain’t workin’, gotta go!

Visualizing Defect Percentages with Parallel Sets

Prof. Robert Kosara’s visualization tool, Parallel Sets (Parsets) fascinates me. If you download it and play with the sample datasets, you will likely be fascinated as well. It shows aggregations of categorical data in an interactive way.

I am so enamored with this tool, in particular, because it hits the sweet spot between beauty and utility. I’m a real fan of abstract and performance art. I love crazy paintings, sculptures and whatnot that force you to question their very existence. This is art that walks the line between brilliant and senseless.

When I look at the visualizations by Parsets, I’m inclined to print them off and stick them on my cube wall just because they’re “purty.” However, they are also quite utilitarian as every visualization should be. I’m going to show you how by using an example set of defects. Linda Wilkinson’s post last week was the inspiration for this. You can get some of the metrics she talks about in her post with this tool.

For my example, I created a dataset for a fictitious system under test (SUT). The SUT has defects broken down by operating system (Mac or Windows), who reported them (client or QA) and which part of the system they affect (UI, JRE, Database, Http, Xerces, SOAP).

Keeping in mind that I faked this data, here is the format:

DefectID,Reported By,OS,Application Component
Defect1,QA,MacOSX,SOAP
Defect2,Client,Windows,UI
Defect3,Client,MacOSX,Database

The import process is pretty simple. I click a button, choose my csv file, it’s imported. More info on the operation of Parsets is here. A warning: I did have to revert back to version 2.0. Maybe Prof. Kosara could be convinced to allow downloads of 2.0.

I had to check and recheck the boxes on the left to get the data into the order I wanted. Here is what I got:

See the highlighted defect.

So who wants to show me their piechart that they think is perfectly capable of showing this??? Oh wait, PIE CHARTS WON’T DO THIS.  Pie Charts can only show you one variable.  This one has 4.

This is very similar to the parallel coordinate plot described by Stephen Few in Now You See It and shows Wilkinson’s example of analyzing who has reported defects. She was showing how to calculate a percentage for defects.  See how the QA at the top is highlighted?  There’s your percentage.  Aside from who has reported the defects, Parsets makes it incredibly easy to see which OS has more defects and how the defects are spread out among the components.  If I had more time, I would add a severity level to each defect.  Wouldn’t that tell a story.

Parallel Sets is highly interactive.  I can reorder the categories by checking and unchecking boxes.  I can remove a category by unchecking a box if I wish.

I took away the individual defects.

By moving the mouse around, I can highlight and trace data points.  Here I see that Defect 205 is a database defect for Mac OS X.  Although I didn’t do it here, I bet that I could merge the Defect ID with a Defect Description and see both in the mouse over.

See the highlighted defect.

Parallel Sets is still pretty young, but is just so promising.  I’m hoping that eventually, it will be viewable in a browser and easier to share.  Visualizations like this one keep me engaged while providing me with useful information for exploratory analysis.  That’s the promise of data viz, and Parallel Sets delivers.

Plagues aren’t just for blog posts

Vibrio cholerae with a Leifson flagella stain ...
Image via Wikipedia

For the past couple of months, James Whittaker has been writing about the “plagues of testing.” As he’s been posting, I’ve been reading through a book about a real plague.

As software testers, we see a system from a perspective that developers and business types rarely and may never see.  We know our tests, we know how well they ran.  We know our system under test and which components are picky.  If you are like me, and have access to the code base, you also know the code.  In the case of both the system and it’s code you know what should be better.  Sometimes this is not a big deal, but sometimes it is a warning that it’s time to polish the resume. I have not seen this, personally, but I know that there are testers who find themselves in this position.

John Snow was a doctor in this position. He was an expert in the new art of anesthesiology in 1850’s London. He was also deeply involved in the study of cholera and concluded that it was a water born illness. This was very much counter to the prevailing theory of the time that cholera was passed through a sheer volume of stench or miasma (I can hear Dr. Evil saying this word). Health authorities in London, were so convinced that miasma, or extreme smellyness, was the reason for disease that they passed a law in 1848 requiring Londoners to drain their waste into a sewage system that would deposit into the Thames river. Unfortunately, the Thames was also a main source of drinking water for the city. Snow knew all of this and could see a health crisis in the making.

In 1854 when a major outbreak of cholera erupted in a neighborhood very close to where Snow lived, he conducted a thorough investigation in order to prove this theory that cholera was passed through water. He had a list that detailed the names and addresses of 83 people who had died from cholera.  He also had an invaluable resource of detailed information about the neighborhood’s residents in the form of local clergyman Henry Whitehead.  While Whitehead tracked down and questioned everyone he possibly could about their drinking habits, Snow analyzed this data to form patterns of who had been drinking the water, who had died and, just as importantly, who had NOT died.  Not only were Snow and Whitehead able to convince the local parish board to remove the handle of a contaminated water pump, they knew their data so well that they were able to figure out the index, or original, case that had started the outbreak.

John Snow Map

After the outbreak had subsided, Snow put the analysis from his investigation together with a very famous map into a monograph that circulated among London’s health professionals.  His monograph slowly but effectively turned the tide of thinking among health professionals away from the miasma theory.  This and a very smelly Thames convinced authorities to build a new sewer system that drained into the sea.

Frequently, when I talk to people about data visualization, they always ask me how I know what to visualize.  The Ghost Map, by Steven Johnson, illustrates this perfectly.  John Snow went from having a theory to proving his theory with visualization in a convincing way.  There’s no wizard or easy-button for this one.  It takes knowing what you are trying to say and knowing the data, inside and out, you are using to prove your theory.  For testers, this means digging into test runs and testcase organization.  Where are the tests that failed?  How many times did they fail? How are they grouped together?  Does an object that make 1 test fail make others fail as well?  If you know which tests are failing, what do you know about the code you were excercising?  How complex is it?  I know this awful, but I would so go here if I thought it made a difference…who coded it and do I respect their coding talent? Even if I think they are solid, did they know what they were supposed to be doing? You have to know exactly why you are telling the PM you think there is a serious problem and have a way to show it. I lump business people in with having the same short attention span as doctors and politicians. My blog post probably lost them at “James Whittaker.”

Johnson and Whitehead could not drag the dead people’s relatives in front of London’s public health community and force the doctors to listen.  These days doctor’s short attention span is because of insurance, but I’m sure there were other reasons back in the day.  A good visualization does not take a long time for the viewer to process.  That is their special power.  Snow’s map is much more concise than a table of 83 names and addresses along with their individual stories.  Visualization can quickly show your groups of tests that are failing.  It can show that severe defects are increasing, and not decreasing, over time.  Business may drive the ship/do not ship decision, but a good tester will know why a seriously ailing system is in so much trouble.  A great tester can effectively communicate this to a business team.

Reblog this post [with Zemanta]

First Attempt at Visualizing Tests and Defects

This post is about a visualization I created to show test execution status with related defects using data obtained from HP Quality Center. I’m using Excel to create this chart and deliberately stayed away from “fancy tricks.” If you want to recreate this some steps will be different if you don’t use Quality Center or Excel. In that case, you get to figure it out.

My weekly status meeting drove me to create this chart. Every week, I sit in this meeting with some pretty important people. Whenever I’m in a testing phase, I have to show what it is I’ve been working on. Previously, I’ve used the report templates from Quality Center, but they really are crap. Wait, that’s not big enough…They really are C-R-A-P. Not only is it difficult to jostle the correct data into place, but they are ugly and none of my superiors actually trusts the Quality Center visual. Since I’ve been doing all this reading about data viz, I realized that I could easily create something much better using Excel in less time.

There are several steps to creating any visualization, and I’ll break down the creation of this chart into the following steps. They are from Ben Fry’s excellent book, Visualizing Data: Exploring and Explaining Data with the Processing Environment.
Acquire – getting your data
Parse – structuring, and sorting your data
Filter – remove stuff you don’t need from your data
Mine – apply mathemagic in the form of statistics, data mining, whatever to show patterns in your data
Represent – choose the type of graph or chart you will be using
Refine – polish your chart so that it has clarity and makes people want to look at it
Interact – add ways for the user to explore your chart

Acquire
Looking at your test set in quality center’s test lab, right click, and QC will show you an option at the bottom for “save as.” Click and this and choose excel sheet. Html would work too since excel can parse through QC’s html.

Parse
For this chart, I like to create 3 groupings: passed, failed and no run. You can have other groupings, but you probably want to make sure that the “other” groupings are together. In Excel, I sort the data by status into the 3 groups. If you are using a test set from the past, you might only have 2 groupings.

Filter
This will change depending on how large your team is and how the work is divided. Since I’m the only tester, we all know who executed the tests in my chart. I remove the columns for Attachment, Planned Host, Responsible Tester, Execution Date, Time and Subject. This leaves only 2 columns, Test Name and Status.

Mine
No mining in this one, it’s just straight up data.

Represent
This chart will be using a stacked bar format which looks similar to a stemplot, but isn’t really a stemplot.

Refine
Here’s where the magic is really happening. See that column for status? Add a column between “Test Name” and “Status.” Each cell in the new column gets a color according to that test’s status: Passed=green, Failed=red, No Run and anything else is gray. If you use the basic colors in Excel, your colors will be too bright. In Excel 2003 you can change this by navigating to Tools>Options>Colors. Here you can choose some less saturated versions of red and green. You’ll still need the very bright red, so don’t save another color over it.

Use fill color to change the colors in the empty column next to the status column. Now you can see each test and you can see how much is passed and how much is failed. Since the status is now indicated by color, you can get rid of the column with the status as text. If you’d rather keep it, you could have one column with the fill and the font color set the same. This would mask the text of the status inside the cell.

Gridlines can be very distracting, so clear all of them. To do this navigate to Tools>Options>View>Window options and clear the Gridlines check box. Doing this should immediately relax your eyes when you look at the chart.

To make it obvious which test has which status, right justify the column for test name.

If you want to show defects on your chart, you can do this using color. Remember where I said not to save over the bright red? If you have defects that you need to show on the chart, change the color of the related test case to bright red and add the I.D. and title of the defect to column on the right of the status. This works when you typically have 1 defect per test case. If you have more, I would just color that test bright red, and have a list of the defects elsewhere. Showing defects this way highlights the need for accurate and concise defect descriptions. I was reading in How We Test Software at Microsoft that their testers work very hard at creating good defect descriptions. In fact, a friend of mine had a great post that received excellent comments about this. Here is where the defect description can make a big difference. Ditto the test names. Everyone in the status meeting sees these and asks questions.

Edward Tufte would probably say that I should print this chart out on a really big piece of paper, but he doesn’t work in my group. My status meeting is paperless, so this will be displayed on a big fancy conference room flat screen, ‘cause that’s how we roll. Actually, I’m not joking about the rolling part. The person who leads the meeting has a tendency to compulsively scroll up and down, so I designed my chart to be displayed within the width of his laptop screen. That’s why the summary is out to the side. Also, I made the text of the title and the summary much larger than the font of the individual tests. I’m using Trebuchet MS for the font.

Interact
At the most basic level, this chart allows the user to interact with the chart by focusing on the smaller, individual tests if they choose to do that. They can also look at the defect information if they would rather. As an improvement, I might figure out how to add some information for each test when it is moused over.

This chart probably won’t scale if you have multiple testers, multiple projects or copius amounts of tests and defects. I know that’s most testers, but I really made this chart specifically for my needs. This shows the very busy and important people what they need to see in a way that they can trust.

As always, comments are welcome. Love it, hate it? Let me know. I’m still learning about this stuff and welcome the feedback.

Integrity in Data Visualization: Part 2 of 2

In the second part of this series about data visualization (Part 1 is here), I will show how, according to Edward Tufte, information visualizations can trick the viewer. Since I read through this chapter in his book, The Visual Display of Quantitative Information I’ve noticed a compromise of graphical integrity in the most surprising of places. Some of these are included in my post as examples.

Currently, I’m writing a paper for my metrics class on software visualization. After the class, it will be expanded to include how software visualization can work for testers. Part of knowing how to use visualization for any project is understanding what constitutes a good or bad visualization. I’m guessing that if you found my blog you might be a tester, and in that case, understanding the basics of data visualization will help you understand where I’m going with some of my upcoming posts. Outside of testing, understanding complex visualization is a skill we all need to have because we live in an age of data.

Labeling should be used extensively to dispel any ambiguities in the data. Explanations should be included for the data. Events happening within the data should be labeled.

For this example, I’d like to use a graphic that was in a post on TechCrunch as I was in the process of writing this post. It was posted by Vic Gundotra who is VP of Engineering for Mobile and Developer Products at Google. You can read his full post here. I’m calling out a couple of his graphics because I was pretty shocked that he would post these. I’ve seen some of his presentations on YouTube, and they were awesome presentations. Using graphics as bad as what he posted on TechCrunch only diminishes an otherwise strong message and will make me think twice about any visuals he presents in the future.

Here is the first of Vic Gundotra’s graphics. Notice how he does not label the totals on the graph, but separately at the bottom. There is no way a viewer can compare the data he’s representing.

In times series graphics involving money, monetary units should be standardized to account for inflation.

Ok, this might be a controversial example but please read the explanation before flaming me. I freaking love this award winning interactive graphic about movies created by the storied New York Times graphic department…BUT I have a bone to pick with it. The graphic shows movies from 1986 to 2008 and it does not account for inflation. There are some other things going on with this graphic as well, but since it’s about movies and not our budget crisis, I give it’s lack of adjustment for inflation a “meh.” It just goes to show that nothing is perfect.

All graphics must contain a context for the data they represent.

Here’s another Vic Gundotra graphic. Notice how there is no total for the number of users represented, yet Gundotra is trying to say that 20 times more are using the T-Mobile G1. By not including this number, Gundotra is not providing an accurate picture of how many people are using either. It could be 20 people using the G1 or it could be 20,000. There’s no way to know. The fact that he didn’t include this number says, to me, that maybe the number of people is embarrassingly small, but that’s another post for another blog.

Numbers represented graphically should be proportional to the numeric quantities they represent.

This is all about scale in graphics. If you are looking at a graphic, the pieces you are looking at should be to scale. Tufte actually has a formula for what he calls “The Lie Factor.” This link has a couple of illustrations and also shows how this formula is calculated.

Number of dimensions carrying information should not exceed dimensions in the data.

You know all of those 3d pie chart and bar graph templates in the Microsoft Excel chart wizard? Don’t use them anymore, and yes, I’ve used them myself in the not-too-distant past. They qualify as “chart junk” from Tufte’s perspective.

Variation should be shown for data only and not the design of the graphic.

I looked but couldn’t find a good example of this on the web. (If you see one let me know.) One of the graphs that Tufte uses to illustrate this point shows a bar graph where the bars for years that are deemed, “more relevant,” are popped out in a separate larger section using a really heinous 3d effect. It’s on page 63.

As always, comments are welcome.

Reblog this post [with Zemanta]