Visualizing Complex Data Sets? 180
markmcb writes "A year ago my company began using SAP as its ERP system, and there is still a great deal of focus on cleaning up the 'master data' that ultimately drives everything the system does. The issue we face is that the master data set is gigantic and not easy to wrap one's mind around. As powerful as SAP is, I find it does little to aid with useful visualization of data. I recently employed a custom solution using Ruby and Graphviz to help build graphs of master data flow from manual extracts, but I'm wondering what other people are doing to get similar results. Have you found good out-of-the-box solutions in things like data warehouses, or is this just one of those situations where customization has to fill a gap?"
perhaps worth looking at? (Score:3, Informative)
Abstract: We propose a method for characterizing large complex networks by introducing a new matrix structure, unique for a given network, which encodes structural information; provides useful visualization, even for very large networks; and allows for rigorous statistical comparison between networks. Dynamic processes such as percolation can be visualized using animations. Applications to graph theory are discussed, as are generalizations to weighted networks, real-world network similarity testing, and applicability to the graph isomorphism problem.
Re:perhaps worth looking at? (Score:5, Funny)
Re: (Score:2)
Abstract: We propose a method for characterizing large complex networks by introducing a new matrix structure, unique for a given network, which encodes structural information
I should probably go follow your link, but on the face of it, this sounds like a 60's paper about the adjacency matrix :)
Re: (Score:2)
Truly nice way of getting high-level information from graphs. I'll probably use it sometime in the future, as I actually work with graph viz, and was looking for ways to visually compare related graphs.
And for those too lazy to actually open the PDF and look at the pictures, the idea is to draw histograms of 1st order vertex degree up to nth order vertex degree (where order-l vertex degree gets defined as how many other vertices can be reached in exactly 'l' forward hops, as found by a breadth-first search
get rich slow (Score:1, Insightful)
Unfortunately, this means that it is much too utilitarian (and ultimately, why products like Peoplesoft are making headway).
If you find that you have developed a good product to help with operating SAP, you can sell it as a third party add on. Many of the popular add on's were created out of a sense of frustration with the "mother product".
Re: (Score:1, Informative)
American. But I did not intend to impugn German anything.
SAP's "engineering"
There, fixed that for me.
Re: (Score:1)
http://autos.msn.com/advice/CRArt.aspx?contentid=4023544 [msn.com]
The American cars are apparently more reliable than you remember.
Re: (Score:2)
Let us not forget that the cars that Germans drive didn't used to be, and probably still are not exactly the cars sold under the same nameplate in the US. And the differences are not all because the Teutonic version is better. Further, I remember visiting Germany on business trips in the 1980s and trying to deal with what appeared to be basically a third world phone system -- not up to the levels of even GTE's then awful service in Southern California and certainly not up to the levels of phone service in
Re: (Score:3, Funny)
Their priorities are different. I come from three generations of German VW assembly line workers (Wolfsburg plant). I grew up in the US. Last time I saw my now-retired uncle, he asked me about the (then) new VW Bug and why the Americans kept putting flowers in the dashboard-mounted gun rack. Explanation was pointless.
Re: (Score:2)
Re: (Score:2)
One of my work colleagues was in the states at Christmas visiting family. As far as he is concerned most mid range cars made for the US market are utter rubbish.
He put it down to the fact that over there your roads are very different. You have an awful lot of straight lines and much less traffic due to increased road capacity. He also pointed out that since you have loads cheaper Gas you will quite happily buy large engined monsters that take ages to reach a reasonable speed.
We have cluttered roads, incessa
Re: (Score:2)
And please don't underestimate the torque of those "large engine monsters", a 425 (that's 7 l for the SI crowd) can make a 2 t Cadillac accelerate rather decently.
Re: (Score:2)
Re: (Score:2)
It is, but it only likes to integrate with itself.
PtolemyPlot (Score:2, Informative)
Great source of data visualization inspiration (Score:3, Interesting)
http://visualcomplexity.com
Have fun!
Re: (Score:2)
http://visualcomplexity.com
Have fun!
Seems to be excellent inspiration on visualizing data..
canâ(TM)t open the page âoehttp://visualcomplexity.com/â because it could not connect to the server âoevisualcomplexity.comâ
it apparently advocates the RIAA approach.. pretend its not there at all.
Visualizing via OpenGL (Score:2)
These can be incorporated into other general shelf visualization tools or just be used standalone on any major platform as long as the machine has the horsepower, including, not suprisingly, a powerful GPU.
the first computer i started doing visualzation on was a SGI. imagine that.
I have a question for you (Score:5, Insightful)
How are you supposed to handle the data if you do not understand it? Sure, there can be too much to see/think about at one time, but if you don't understand it, how can you visualize it usefully?
I am asking because I have a problem: Where I work, I understand the data and I make efforts to visualize it for others. The trouble starts when they don't understand the data and it's sources and limitations, so what they see in my visualization is all they know of it, and they make assumptions about it. I've even had people worry that the network is down because there were holes in the collected data which then showed up in the visualizations.
If anyone has some good URLs for such thinking, I'd be grateful.
I simply do not understand how you can visualize data for people if you yourself do not understand it.
Re: (Score:3, Interesting)
the lat, lon and depth values are courtest of NOAA, freely available. this is a screenshot of a real time frame in openGL of the world with each vertex pair colored by depth. you can rotate it, probe it and a few other things.
link [rimfrost.org]
Re: (Score:2)
If give you a scatterplot of X vs Y you would instantly be able to see what kind of relationship exists (if any). You might notice an exponential relation, linear relation, or that they are the exact same. For more statistical things, you might notice two distinct groupings, correlations, or completely uncorrelated date.
So your boss doesn't know what the relationship between X and Y are, but he wants to save $ and increase profits. You tell him that based on this relationship that you proved exists (whate
Re:I have a question for you (Score:4, Interesting)
What you are doing in your post is investigating the data until you UNDERSTAND what is usefull and then presenting (visualising) it for you're boss, who probably adds another layer of "visualization" for his boss, etc. (ie: You are acting as human visualisation tool that the boss can use to visualise the output of silicon visualisation tools)
To scale up you're simple X/Y plot of two variables to corporate size you propose using a visualization tool that UNDERSTANDS database structures and UNDERSTANDS the fact that to plot strings against integers you need a default transform, etc, etc. You are handed a bunch of DB's with hundereds of tables, thousands of columns and countless transaction transforms ferrying data from one DB to the other.
So you start with all possible pairs to see if there is a nice easy curve that can relate them. You get 10,000 statistically significant relationships - the problem posed in TFS is how do you now visualize all those graphs to find the relevant relationships without UNDERSTANDING the data.
As to TFS, visualization relies on data minning which will never be "solved" because given enough data you can always add one more level of UNDERSTANDING (see: Godel [miskatonic.org]). This is not to say that trying to solve it is pointless. On the contrary, google news is excellent and accessible example of how far things have progressed in the last couple of decades.
Simply presenting multiple known facts/relationships in an easily accessible format takes a deep UNDERSTANDING of the data. Even if you do UNDERSTAND the facts/relationships, creating the format is an art that has few masters [wikipedia.org].
Re: (Score:2)
You should do some 'Exploratory Data Visualization'. Use GGobi (http://www.ggobi.org/) it's free, works fast, can handle lots of data, lots of different formats.
There are a bunch of different options, but when I'm trying to figure out what some basic data relationships, I use the scatterplot matrix (2D matrix of 2D plots) or the 2D tour.
R language (Score:5, Informative)
Re:R language (Score:5, Informative)
http://cneurocvs.rmki.kfki.hu/igraph/ [rmki.kfki.hu]
Processing is another package that is geared towards data visualization which java developers might find easier use
http://www.processing.org/ [processing.org]
Re: (Score:1)
To many times I've seen people jump to conclusions about a small part of a data set, without looking at large sets of data.
Some people don't understand it until you put their fingers on the chart and make them follow the line.
Re: (Score:2)
Also agree about R. Also, consider hiring a statistical consultant. They do this kind of thing all the time.
Re: (Score:2)
I'd also suggest R. One of the problems with visualizing complex data sets is that, almost by definition, the prepackaged graphics tools don't allow you to create custom-designed graphics that suit the particular data-set you're working with. But with a bit of programming in R you can get amazing results.
There are some R packages that can help too -- I write about one of them, ggplot [revolution-computing.com], here. (Disclaimer: I work for a company that provides support for R.)
Try the InfoVis community (Score:5, Informative)
The infovis community has been dealing with these subjects for years. There's many different visualisation techniques around. Here's a list of the past conferences and the papers:
http://conferences.computer.org/Infovis/ [computer.org]
Plenty of good products out there, but the one that I like most is from Tableau Software (http://www.tableausoftware.com/).
Re: (Score:2, Informative)
Xgobi (Score:2)
I used Xgobi (http://www.research.att.com/areas/stat/xgobi/) for a lot of things back in the day. It gave me the ability to 'see' and understand high dimensional data sets quite easily when I was looking at computer vision research.
Spotfire (Score:3, Informative)
Am I missing something or... (Score:5, Informative)
Wouldn't any everyday cube browser along with any tool to detect base dimentions in a datawarehouse schema do the trick? You may have to add a few custom dimentions on your own depending on how shitty the master data is (I don't think that can be helped, no matter the solution, if a dimention is "these two fields multiplied together times a magic number appended to the value of another table", you need to know, no tool will guess), but aside that?
Thats usually what I do anyway. I dump my data in a datawarehouse, use whatever built in wizard can auto-generate dimensions, then play with them in a cube browser. Works for even pretty archaic home-made multi-thousand-tables-without-normalization ERP systems I had to work with in the past anyhow.
IMG is your friend (Score:1)
Not sure what you can use to create a visualization, but the information you need is in the IMG.
I don't have a need to develop a visualization of the whole of our SAP implementation, just my little FI-CO corner of it, and that's a big enough pain
Business Intelligence (Score:3, Informative)
Your ERP isn't supposed to directly analyze the data. You're supposed to use a Business Intelligence software package for that. This being SAP, I believe they'll try to sell you Hyperion.
Re: (Score:2)
They'll push you towards SAP BW. A beast at best. Very strong tool, but like anything SAP, difficult to master.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
Honest attempt at an answer... (Score:2)
I'll try to answer your question without the key info needed: "What is the data your modeling?"
You're on the right track...
Either way, from experience i'd say you're answer is "this just one of those situations where customization has to fill a gap"
Be warned though, out of the box solutions do exactly what's on the box. Anything else is going to be modeled by you, or customized (usually at a high rate), by the vendor.
That being said, I've used oracles' solution http://www.oracle.com/solutions/business_int [oracle.com]
Re: (Score:1)
Hey, you managed to use both "your" and "you're" both correctly and incorrectly in one post!
I think you must be taking the idea of stochastic grammar somewhere it doesn't belong...
Just take the first 65k rows (Score:2, Funny)
Just take the first 65k rows and dump them into excel and create a pivot table.
Re: (Score:2)
Re: (Score:1, Funny)
65k rows should be enough for everybody.
Re: (Score:2)
[in Excel 2007 that's] ~2M rows!
I can't help but notice that 2M ~= 2^21. So the index fits perfectly into one word of three 7-bit bytes.
Are Microsoft programmers optimizing for weird 60's architectures? ;)
Re: (Score:2)
Actually, Excel 2007's limit is 1,048,576 rows or 2^20 which is pretty round in binary, but still somewhat arbitrary.
http://office.microsoft.com/en-us/excel/HA100778231033.aspx [microsoft.com]
Re: (Score:1)
Re: (Score:1)
Centruflow (Score:1)
Project is on RubyForge... (Score:2)
...which I just took offline for a quick database upgrade. Er, sorry, will be back online soon!
Re: (Score:2)
...and, back online; here's the Ruby/ASP project [rubyforge.org] which also hosts ruby-graphviz.
Take a look at Prefuse (Score:2)
Just pipe it (Score:3, Funny)
Re: (Score:2)
I'm not sure if you're jesting about that, but I wrote a patch for xscreensaver to allow just that functionality for the xmatrix hack. Looks great for things like 'ps -eo command'. See any version from 5.04 onwards.
I wrote this after noticing the DNA encoding and thinking, "hey. Wouldn't it be great if I could feed this with real DNA sequence?"
Unfortunately, I had to send the same stream to each of the feeders on the screen, which means it can only show one vertical line of data, rather than 40 or so.
Re: (Score:2)
Good idea, by the way.
General Purpose InfoViz Tool (Score:1)
I've had really good success using an information visualization tool called Starlight on a number of projects like this. Everything from process modeling to military intelligence. It's a commercial spin-out from the DOE PNL lab information visualization research in Washington State.
http://www.futurepointsystems.com/
Four dimensions (Score:2)
Re: (Score:2, Insightful)
I'm sure it's my mathematics background, but when I saw the headline I assumed the author would be discussing something involving the square root of negative one, to which my response was, "Silly author, you can't visualize four dimensions. (Sober.)"
You have a mathematical background and can not visualize four dimensions? Here is how you do it: Just visualize the problem in n dimensions and then set n=4.
IBM data explorer (Score:3, Interesting)
Re: (Score:2)
Here's a community that looks to be more recently active (that is, it has been dormant for a shorter period) than the original IBM site, having taken the last IBM build and worked on it - http://www.opendx.org/news.html [opendx.org].
Disclaimer: I looked at OpenDX for a project, but never went anywhere with it.
Re: (Score:2)
It has its own website here too:
OpenDX [opendx.org]
You need a Powerwall (Score:1)
Both tools combined allow you to easily visualize large data sets and adjust the resolution of your data.
Pentaho (Score:2)
Traditionally... (Score:4, Funny)
Essbase (Score:2)
Take a look at Essbase http://en.wikipedia.org/wiki/Essbase [wikipedia.org]. It is now owned by Oracle and is used by finance departments at most Fortune 100 companies.
As you have do doubt discovered, SAP is great for transaction level detail, but kinda sucks at the big picture and doing "what ifs". Essbase's tight integration with MS Excel and very cool reporting tools makes it a much easier to analyze your data than looking at spending reports from SAP.
Mainly implemented by budgeting and finance groups, Essbase is not
Application? (Score:1)
DAD software [dad.net.au] has the ability to customize data types, multiple inheritance of objects, and to define different relationship types.
You can then trace along object relationships bringing back a dynamic graphic depending on what you want to show (and spit out to PDF).
Processing (Score:1)
Business Intelligence (Score:1)
Re: (Score:2)
Normally, BI and data warehousing tools in general are used for visualising things like sales of beer vs temperature or cost of toilet roll relative to the same quarter last year. Basically, transactional data and trends.
I think what the OP is looking for is either something to generically draw interrelationships between tables, or between the defferent views on the material master, BoMs and the like. In either case the problem is that when you do it with any non-trivial data you'll need to hire a basketb
I used to work for (Score:2)
The solution you probably want is to make sure your SAP is set up to use a common relational database, then use another tool (Crystal Reports, Seagate, etc.) to visualize your data in ways that are not already built-in to your ERP system.
Re:I used to work for - on the right track (Score:2)
I'm a SAP consultant and have "cleaned-up" several data sets over the years. I'm lucky in that all of my customers are running it on Unix with DB2.
I wrote a series of PHP scripts that go through everything and present inside a somewhat simplified web interface. I also use Crystal Reports to provide "cleaner" copies.
But, at the end of the day, it's more of a brute-force exercise then anything. Providing a simpler interface then R/3 is the first step, but you have to have users that are willing to use. What I
Re: (Score:2)
Our biggest customer complaint was that we did not use a "common" relational database (SQL Server, Oracle, etc.). Therefore, we had to provide all the data views the customer might desire which is not feasible.
The simplest solution to this dilemma was to modify our system to use a database that the customer could access themselves... read-only of course. Of course the relational databases were slower, and the modifica
Tableau All The Way (Score:1)
I first saw a video of Hans Rosling [youtube.com], who had some very unique ways of visualizing data that would otherwise be useless to a simple mind such as mine.
After I watched that, I found a piece of software called Tableau [tableausoftware.com]. I downloaded the trial version, and really liked how easy it made visualizing data for me. I can take the data I have, and Tableau will see how it's connected and allow you to generate visual reports of the data. I'm not saying that it'll work for everything, but it certainly does what I need i
StarLight (Score:2)
depending upon the problem domain, a very useful (albeit expensive) set of tools is StarLight, written for the US Government: http://starlight.pnl.gov/ [pnl.gov]
highly recommended if you've got tough visualization problems. this tends to get used for the *really* interesting visualization challenges.
PivotLink (Score:2)
I am pimping my own employer's product here, and I'm admittedly biased, but we've got a phenomenal web-based/SaaS solution to this exact problem. We've done work for clients with billions and billions of rows of data (like 50+GB) and we've got a unique database that can generate reports in seconds that could take upwards of fifteen minutes on a SQL-backed solution. You can take any report, drill down arbitrarily into the data below, flip through the datasets, arbitrarily flip axes, filter out unwanted da
Business Intelligence Cubes (Score:1)
Cytoscape (Score:5, Informative)
I had a similar situation to yours recently, except I was trying to detangle a horridly complex product substitution graph for a logistics company.
I used a bunch of Perl to crunch the raw databases into various abstract graph structures, but instead of graphviz or something created by/for developers, I found that the best software for graph visualisation is the stuff that the genetics and bio people use.
The standout for me was a program called Cytoscape [cytoscape.org] which can import enormous graph datasets and then gives you literally dozens of different automated layout algorithms to play with (most of which I'd never heard of, but it's easy to just go through them one at a time till something works)
It's got lots of plugins for talking to genetics databases and such, but if you ignore all that and use Perl/Ruby/whatever for the data production part of the problem, it's a great way to visualise it.
Looks like SAP tricked another sucker (Score:1)
Looks like SAP tricked another sucker.
A company I worked at several years ago migrated to SAP. It took several hundred million dollars, 6 years, AND the companies main branch was already using SAP. All to replace an MVS system that cost under $5M a year to run, did more, and was much faster.
SAP is NOT a business application. It's a programming environment where you get to build and customize your own. Then those German Wunderkids break your customizations every time there is an SAP change.
A "good" business
Re: (Score:3, Insightful)
Looks like you work for another company that tried to reimplement their old system word for word and step by step in SAP.
And customization has a specific meaning in SAP that doesn't involve any coding. It appears you don't know that, which doesn't improve your credibility.
If you want that, don't use a
Sql-fairy! (Score:1)
A good start (Score:1)
Data Mining? (Score:2)
Have you looked at data mining solutions? Someone mentioned Pentaho already, but there's also:
Rapid Miner [rapidminer.com]
Orange Data Miner [ailab.si]
all of which are packed with enterprisey features. But you may have to learn some stats. Once you get past what you can do with the pre-packaged stats methods, then head for R [r-project.org], or write a RapidMiner plugin in python.
Check out Stephen Few and Tableau Software (Score:1)
JMP (Score:2)
There's a product called JMP [jmp.com]. It's relatively inexpensive. It's great at visualizing (especially statistical) data. They've got a 30-day trial. Check it out.
__
disclosure: I have an association JMP's parent company.
Re: (Score:2)
It also runs on Linux MacOS and Windows. It works great with large data sets and supports a lot of data sources (ranging from XLS files to SAS databases)
SAP MDM ? BI ? (Score:2, Insightful)
Re: (Score:3, Funny)
Bargain and SAP are not two words to be used together!
Tool in Hans Rosling's infamous TED presentation? (Score:2)
Re: (Score:2)
CoMotion (Score:2)
General Dynamics offers a product called CoMotion that allows you to visually explore your data and find interesting patterns and trends.
http://www.gdc4s.com/content/detail.cfm?item=32341561-76f9-40f8-8ad5-0f0d66dd240e [gdc4s.com]
CoMotion is a commercial fork of Visage, a collaborative visualization platform designed at Carnegie Mellon University and MAYA Design [maya.com]:
Palantir (Score:2)
I don't use their product but http://palantirtech.com/ [palantirtech.com] makes a data visualization tool and has a good blog about it, with some interesting Java dev tips thrown in. It might be overkill for the data discussed in the article summary, but sounds pretty badass.
One really interesting blog article http://blog.palantirtech.com/2008/12/12/vizweek-2008-report/ [palantirtech.com] talks about something called the "VAST Interactive Challenge", which as near as I can tell is a competition for data visualization tools to go head-to-head a
Check out OpenDX (Score:2)
Check out OpenDX [opendx.org], its visualization capabilities are way beyond Graphviz's and it provides a GUI. It's an open source version of IBM's famous Visualization Data Explorer (initially released in 1991), which IBM converted into an open source project a couple of years ago.
Quoting the site: "OpenDX is a uniquely powerful, full-featured software package for the visualization of scientific, engineering and analytical data: Its open system design is built on familiar standard interface environments. And its sophis
Re: (Score:2)
BTW, I suppose you'd like to have a look at Data Mining [opendx.org] examples gallery first.
Re: (Score:3, Insightful)
Re: (Score:1)
Colourmapping (Score:2)
The one thing that OpenDX still has, that for some reason the others have not, is an excellent interactive colourmap editor. I use OpenDX to get the colours correct and then export to use in other toolkits.
Re: (Score:2, Interesting)
Tufte's ideas are good for presenting simple information. He gets many things right (eg if the visualisation doesn't work in black and white, adding colour won't fix it). However, many in the infovis community are outright sceptical, if not dismissive of his ideas for analysing high dimensional datasets.
Where his ideas really work is once you have "the answer" that you want to present to someone else. However, the basic exploration of the data to find interesting keypoints, is not what he specialises in. Th
Re: (Score:2)
Tufte's work is a great foundation for people who don't know anything about presenting data effectively. I know several people who have been to his seminars, one of whom is a CIO who has since made Tufte's books required reading for most of the people who report to him.
Like the other poster here said, Tufte's stuff isn't that useful for really complex multidimensional data. His stuff is more oriented toward presenting relatively simple stuff in ways that are readable and aren't misleading. On those two subj
Re: (Score:2)