• data mining/machine learning/statistics
• data visualization
• database management
• programming

There are some free online courses that cover many of these areas, and these courses are usually part of a degree or a certificate program in data mining. Those who are new or interested in this field can learn a whole lot without paying a dime. Here is the list:

• Intro to Probability and Statistics (Carnegie Mellon)
• Machine Learning 101/102
• GovData (MIT/Harvard)
• STATS 120: Information Visualisation (The University of Auckland)
• R Programming (UCLA)
• CS 229: Machine Learning (Stanford) (videos)
• Linguistics 420: Statistical Natural Language Processing (Georgetown)
• SI 508: Networks: Theory and Application (University of Michigan)
• CS 591: Data Mining (West Virginia University)
• STATS 782: Computing for Statisticians (The University of Auckland)
• 6.867: Machine Learning (MIT)
• Andrew Moore’s Slides on Statistical Data Mining Tutorials
• Lots of tutorials (Data Mining Tools)
• Capstone project:  kaggle or kdd (for a bigger list see kdnuggets)

Some free text books:

• The Elements of Statistical Learning by Hastie, Tibshirani, and Friedman
• Mining of Massive Datasets by Rajaraman and Ullman

In addition, there is an excellent thread on quora on how to become a data scientist that covers lot of things and is a very good resource on the practice of analytics.

We thought that it was a table size issue, may be the driver could not handle huge data. However, the problem was with the “unique record identifier” while linking the table, as JM reported here. I did not have this problem any longer when I did not select any fields for the unique record identifier while linking the table.

Although my syntax was right, MS Query kept throwing exception about the CASE statement. I read some discussions on-line, and some people had the wrong syntax, such as including the alias of the case statement in the group by statement. I did not have that problem. Finally, I found the solution: It was the order by clause. I was using the alias of the case statement in the order by clause. I removed it and the query ran just fine.

I was reading about knowledge management here, which says that knowledge management is nonsense. I agree to a certain degree, not because of the field, but because of its name. How do you manage knowledge? Isn’t knowledge derived? Wasn’t information “science” good enough? (I have problem with “business intelligence” as well…) As the author of that article says, it is a new term coined to attract attention. He does provide some evidence, but I was left unsatisfied.

I thought of performing text mining on publications database, and citeseer has this great resource. I downloaded the data (72 XML files), performed some clean-up, and ran a script to pull citeseer ID, author addresses, and publication dates where the abstract contained the term “knowledge management”. I was interested in seeing the trend of publication and places of publication.

Have a look at this chart:

There is a definite growth in this area, at least in research and publications. It is startling to see a paper published in 1970, and a peak in 2002. As citeseer data ends in 2004, it is possible that it doesn’t have complete publication history of 2004.

Geographic location wise, the US and Europe leads the way in number of publications:

The data came from this report came from a complex query using SQL server and Access that had this format:

Some options I investigated and tried (be sure to read this page, you might find many more):

• Google’s Chart API (http://code.google.com/apis/chart/types.html): you can embed these in Google spreadsheets, create HTML pages using Excel via VBA, or embed them in Excel sheets. Alas, none would work the way I would like them to work: In-cell graphics
• Use Sparklines for Excel add-in: this add-in will create great in-cell charts (bar, bullet graphs, sparklines, etc), but copying them down is difficult and resource intensive, and any change you make in the column size will alter the shape of that chart. In addition, I had more than 400 rows to populate–these would be too many objects in a spreadsheet for Excel to handle. My machine froze when I tried to copy it for 10 rows.
• Use R’s implementation by Jason Dieterle (search on this page for Jason Dieterle (email), January 28, 2008. Works very good. I modified the code to print max and mins only, but the function generates a graphic, which needed to be embeded in LaTex file. It did not work nicely. Charts were too big to fit in a cell of a table.
• Create bar charts in Excel using REPT function. I tried different font sizes and styles, but it didn’t look pretty, for the range of data varied. (Hint: make the alignment of text 90, use pipe signs, create columns and graphs for each data value, remove gridlines, keep the columns very close. It almost worked.)
• Use spark package for LaTex, doesn’t work in pdflatex, and you have to play a lot with the settings
• Use sparklines package for LaTex. Problem was that data needed to be normalized (or scaled) from 0 to 1, and needed extra parameters for min and max points. Solution: create a normalize function in Excel, and write a big formula to produce the exact needed string for the sparklines to work i.e.:
  \begin{sparkline}{5}
\sparkdot 1 1 blue
\sparkdot 0.2 0 red
\spark 0.2 0 0.4 0.0625 0.6 0.5625 0.8 0.75 1 1 /
\end{sparkline}

Here’s the normalize function:

Public Function Normalize(cell2Normalize As Range, WholeRng As Range)
       Normalize = (cell2Normalize.Value - WorksheetFunction.Min(WholeRng)) / (WorksheetFunction.Max(WholeRng) - WorksheetFunction.Min(WholeRng))
End Function

Here’s the big formula to convert the range for data values from 2005 to 2009 (in the range E2:I2) to the sparkline LaTex environment.

="\begin{sparkline}{5} " & "\sparkdot " & CHOOSE(MATCH(MAX(E2:I2),E2:I2,0),0.2,0.4,0.6,0.8,1) & " " & 1 & " blue " & "\sparkdot " & CHOOSE(MATCH(MIN(E2:I2),E2:I2,0),0.2,0.4,0.6,0.8,1) & " " & 0 & " red " & " \spark 0.2 " & Normalize(E2,E2:I2) & " 0.4 " & Normalize(F2,E2:I2) & " 0.6 " & Normalize(G2,E2:I2) & " 0.8 " & Normalize(H2,E2:I2) & " 1 " & Normalize(I2,E2:I2) & " / \end{sparkline}"

This last option worked beautifully. I dragged the formula down. I selected the data, and clicked on “Convert Table to LaTex” button (using this add-in). Copied the LaTex code to clipboard and pasted it in my LaTex editor. Manually merged the rows for colleges (using \multirow), and generated a beautiful looking pdf.

I was very happy. Printed it in color. Got a request back very soon that there should be total rows.

I forgot about the sparklines and created a report in Access with plain old numbers in less than 15 mins. Gave it back.

I tried to repeat this – Sparkline in Cognos -in Access with no luck.

(I later tried it one more time: Got the data in Excel using External data> Access, created a pivottable with rows and all, did some formatting, copied and pasted values and formats, inserted sparkline code, converted it to LaTex, copied and pasted in LaTex editor, and here’s the beautiful looking sample pdf of 16)

I wish there were simple reporting solutions that included awesome data visualization tools. (BTW, Excel 2010 will have sparklines: link) For this report, I did try Sweave, R, and LaTex, but because of the time constraints I could not investigate it further.

Please comment if you know any other way which meet (or don’t, Tableau is certainly one) these conditions: inexpensive (read free), efficient, and repeatable.

“The search key was not found in any record”

After some research, I found that either it was a bug caused by Microsoft Jet 4.0 Service Pack or it was the size of the database (should be not greater than 2GB). It turned out to be the later one. I fixed this problem by compacting and resizing the database. In Access 2007, you click on the Office Button (Top Left Corner), then Manage > Compact and Repair Database.

1. Used Yahoo Pipes (I created mine, but this one has more feeds)– this pipe aggregates feeds from different job web-sites, and gives the user unique job listing that you can subscribe via RSS: Job Feed Aggregator by Sean Dolan
2. Subscribed to the RSS feed for the keyword “data mining”
3. Copied the job descriptions and requirements of many jobs, and saved the text file
4. Got the python stemmer
5. Applied the python stemmer to the text file. Stemmer truncates words to their roots, so that we can combine variants of a word into a single word. (First or second step in text mining)
6. Created a tag cloud using the services of http://www.wordle.net/ . They use “stop words,” so I didn’t have to apply those. Stop words are common words, which necessarily don’t add any value for categorization, of a language.

Data Mining Jobs Tag Cloud

The most frequent word is: experience. Companies want people with experience in different data mining techniques. You’ll see that some other big words are: SAS (stemmed as sa), Excel, SQL, analytical skills, statistics, and quantitative skills.

And how do you master these skills, you ask?

1. Get a graduate degree in statistics, economics, mathematics, computer science, financial engineering, or industrial engineering with emphasis on databases, data mining, and marketing.
2. Successfully complete data mining projects using free, open-source data mining tools, such as Weka, R, Orange, Rapid-Miner.
3. Participate in data mining competitions. SAS’s data mining conference has a data mining competition every year.

Have a look at a detailed study by Pejic Bach, M: Creating profile of data mining specialist

Solution: Simply hit the left or the right arrow key, and it will take you to the left or the right cell of the field you were working on.

1. LaTex: a powerful typeset editor, where you focus on writing and not on formatting. It takes care of all the headings, page numbering, figure/table/equation numbering, TOC, bibliography/citation, and more. Although the learning curve is rather steep, once you get the hang of it, life becomes so easy. For windows: you need to get MikTex and any LaTex editor, such as LEd, LyX, or WinEd
2. R: awesome statistical package with wonderful graphics components. Producing stunning graphics and statistics has never been easy. It had me at summary. Any software that can do produce the following, just by giving summary(iris) command has to be great:

Summary produced by R of the iris data set

In my thesis, I had plenty of equations, and every time I made some significant changes, MS Word happily would turn those equations into empty white boxes — and then I had to rewrite them. In retrospect, I find it ridiculous that I was entering citations manually. So every time I added a new reference, I would manually change the bibliography page and the page where I cited that reference. With LaTex, it is just a breeze to do all this.

Descriptive stats, Box-plots, normal curves, neural network, charts with LaTex equations, and a lot of more stuff, all could be easily done using R, and the best part is “repeatability.” With simple commands, you could export all the charts as images for various data sets or for various training algorithms. No sweat! Try that with Excel. (I did some years ago).

• The Elements of Graphing Data, by William S. Cleveland
• Information Dashboard Design, by Stephen Few

Administrators/executives neither have the time nor the patience to understand complicated data mining algorithms and its results, and when they don’t understand them most probably they will never go in “production.” Simple, yet informative, designs and charts have better chances of going in production, which I am sure every data miner longs for.

I found a course web-site on Information Visualization: http://www.stat.auckland.ac.nz/~ihaka/120/lectures.html