Dynamic Forecasting and Charting

Lately I have been evaluating several web-based charting packages. Each has their own unique advantages and disadvantages in terms of price, performance, functionality, and licensing constraints. At the same time, I have also had a few discussion with aquaintances about the 'ideal' methodology for building web apps. Most of the discussions centered around the age-old debate of thin-client vs rich-client. Though new technology acronyms come and go, we have been through this debate in all sorts of past products (POS, kiosks, gas pumps, games, as well as web). I put together this quick forecasting and charting demo as a simple example of a browser-based rich-client using Flash and Javascript as well as a test for different charting tools. (Click to skip straight to the demo)

In the extreme case, a thin-client is a device or application that only does trivial I/O: display preformatted data and send raw user responses back to the server. On the other extreme, a rich-client could be an interactive application that does advanced processing and rendering: the server is nothing more than a remote database service. Unfortunately, the real world is a bit more complicated and most systems fall somewhere in-between.

In the case of interactive web applications, there are several choices of technologies including standard HTML as well as plug-ins like Flash/Flex, ActiveX, and .NET Smart Clients. The latter provide more flexibility, but most still favor traditional HTML and Javascript due to their widespread use and platform independence. Still, the HTML/AJAX route has some pitfalls due to the variances/quirks in the different browser's handling of (X)HTML, Javascript, and the XML/HTML DOM's. There are also many ways to implement and segment an N-tier system so a long line of decisions need to be made:

• Which technology acronym to use on the server: ASP, JSP, PHP, RoR, etc?
• Is the HTML rendered server side or client?
• If client, is the data transformed to HTML using XSLT or Javascript/AJAX?
• Was the data sent as XML or JSON?
• Was the data shipped with the original request or a separate service call?
• If service, was the request SOAP or REST?
• Does validation and business logic happen on the client, server, or both?
• The list goes on...

Personally, I love the scalability and performance of pushing work onto the client. I also love the responsiveness and interactivity of technologies like AJAX and Flash. At the same time, there is still some potential for compatability and support headaches, so my preference is to only use AJAX or Flash in targeted cases (not for the entire site). I'd reserve ActiveX and Smart Clients to intranets where compatability issues and objections have already been addressed.

Charting packages seem to run the gamut of dynamic client-side Flash to server-rendered static images (and even static Flash content). The presentation and business logic layers in an N-tier system are usually easily scaled by adding more servers/bandwidth/caches without too much of an additional cost. So while it is feasible to render charts on the server, you can't beat the interactivity of generating them on the client. At this point my preference is client-side Flash. You can pull data from a static file, a web service, or generate it on the fly using Javascript.

This demo incorporates three basic forecasting methods done in Javascript with a Flash based charting control:

• It works much like a simple spreadsheet application would.
• If you edit the history data, new forecasts are generated on the fly and the graph is updated.
• A checkbox on each row allows you to add or remove that data from the graph.
• For those double-checking my math: the forecast is done using 5 weeks of history and no data cleansing.
• After calculation, all numbers are rounded to the nearest integer.
• The trend is done using the least squares method.
• The seasonally adjusted average is the five week average + (prior year's value - prior year's average).

Series

Month

Show

JAN

FEB

MAR

APR

MAY

JUN

JUL

AUG

SEP

OCT

NOV

DEC

Actuals

Prior

Current

Forecast

Trend

Average

Seasonal