From CS294-10 Visualization Fa08
- Seth Horrigan
I started by looking at the dataset for which I created a visualization in the earlier assignments. It is fairly straightforward to visualize two variables using only position. Adding in color, texture, intensity, etc allows up to ten variables to be visualized in one graphic, although this may become very visually cluttered. I am interested in designing a robust technique to visualize substantially more variables simultaneously. My hope is that I could devise a way to view all 61 variables at once, but that may be over-ambitious. At the least, I wish to visualize a 15 of the ordinal variables together clearly while still maintaining the sense of individual respondents contributing to the whole.
The problem is fully visualizing multi-dimensional datasets in two dimensions. Many datasets, from features of digital cameras to attributes of Wall Street stocks, contain upwards of 20 dimensions that would be useful to visualize simultaneously. For a small number of entries and a mixture of nominal, ordinal, and interval data a table can serve quite well:
However, for a large number of data sources and a large number of dimensions, this quickly becomes unwieldy. Survey questionnaires often contain more than 20 questions, and can have thousands or tens of thousands of responses. Visualizing all of these data points can be very difficult. Basic scatterplot graphs can be very useful for visualizing a large number of entries in a small number of dimensions, but they need to be greatly improved to visualize many dimensions at once. The most common solution I have seen is represent each variable separately using a one-dimensional chart such as a bar graph. I hope to improve on that methodology.
Initial Problem Presentation