Talk:Lesson06

Hello everyone,

I would like to share my thoughts on time series graphs that we have seen in this week's content, especially on possible improvements one can do for the Slope Graph and Horizon Graph. Feel free to give your feedback or counterpoints!

Reason being is that I do see them being useful in giving an intuitive insight, but I am not so sure they can handle data of higher dimensions. For example, the Slope Graph shows change between two time periods well, but if one adds more intervals, it would start to amass lines, similar to how Parallel Coordinates (PC) graphs look. However, PC could filter/exclude without losing the meaning because they are looking at relationships. If we Filter off the Slope Graph, it would lose some meaning, as the relationships have to be present in order to compare the rankings between countries for example.

A possible improvement is to add color to represent the change, or rate of change between intervals, and making it interactive, such as brushing + highlighting effect to track the progress of a singular record. This way, adding more time intervals could be resolved, and allows the user to handle complexity better.

On Horizon Graphs, I believe it has its merits in juxtaposing both the positive and negative values linearly for a large number of records, and having additional dimensions for cross-analysis may become excessive for the user to handle due to the lack of space and color. For example when plotting Horizon Graphs for stock data, price is the main attribute, or perhaps total assets of the company. However, looking at this one dimension alone may not be sufficient. What if we need to add more - how would a Horizon Graph be able to handle this?

Well, we could use additional colors I suppose, but perhaps 2-3 dimension after, it would start looking like a graph of a rainbow. Instead, I thought of combining the Horizon Graph together with the Stacked Area chart concept, then swapping the y-axis to a logarithmic scale - to represent a change, rather than absolute values. Of course, this use case is limited to looking at change over time, and one may still need to see the absolute values. In this case, perhaps using animations to swap between the scales, or even generating charts to represent the dimensions on mouse hover could work.

Please let me know on your thoughts - looking forward to possibly implementing one of these! - Thomas Thio

Visualising Change: An Innovation in Time-Series Analysis

This article firstly describe the reasons why humans are so innately interested in visualization. We can easily see patterns presented in certain ways, but if they are presented in other ways, they become invisible. We can present our data in such a way that the important and informative patterns stand out.

Typical Time-Series Analysis and Its Limitations Lines do a wonderful job of displaying the ups and downs of change through time. Notice how much easier it is to see the shape of change in the graph above, compared to the bar graph below. The strength of a bar graph is its ability to feature individual values and support comparison of one value to another, but the overall shape of change gets lost in the forest of bars.

Meaningful Characteristics of Change through Time Patterns of change through time can take many forms, many of which are meaningful. These patterns are formed primarily by combinations of four characteristics:

• Magnitude of change
• Shape of change
• Velocity of change
• Direction of change

- Alson Tan

Introduction to Cycle Plots

http://www.perceptualedge.com/articles/guests/intro_to_cycle_plots.pdf
This source introduces us the Cycle Plots which not only shows the overall trend but also the day-of-the-week or the month-of-the-year effect. Let's compare it with normal line chart to have a clearer picture about Cycle Plot.

The line chart shows the sales of a company over an eight week period and different colors of the line stand for different weeks. The line chart is able to show a sales trend along with each week,however, it is very hard for us to identify the trend over the eight weeks.