The goal of this project is to create interactive visualizations for a more complex and real - that is, not fabricated - dataset, implementing novel interactions. The data were acquired from UCI Machine Learning Repository , and more specifically, the Computer Hardware data .
As mentioned above, the data we are using is called "Computer Hardware" This dataset looks into 209 models of processors that are manufactured by 30 vendors. They look into 6 numerical attributes for each processor (Thus, there are 8 attributes in total), which are as follows:
We believe that we can use two interactive visualizations to answer the
three questions above.
We can answer the first question using a bipartite
visualization. Here is the example of the bipartite visualization:
Here, we believe that The states can be replaced by vendors, and the
percentages of state incidents with the number of models. The channel
variable can be replaced by the PRP range or ERP range. The user can hover
over a specific score range or a specific model to filter out - that is,
only emphasize the bar on which the mouse is hovered - the rest of the
data. We may be able to further tweak the visualization to show specific
models, not just the vendors, when the user prompts by clicking on the
bar.
The second and third questions can be visualized using a "difference chart"
graph. Here is an example:
Here, we believe that we can put appropriate attribute, most likely the
numerical attributes, as the x-axis and map both PRP and ERP to those
attributes. We can see the relationship and correlation between the PRP or
ERP and these attributes. We can also observe how well ERP represents PRP
by comparing the scores themselves, by mapping both of them at once. The
main form of interaction we can think of is changing the attribute that
will fall into the x-axis. The y-values will also change as the numerical
attributes are not necessarily bound to the models. We may be able to make
the visualization further informative by showing the difference b/w PRP
and ERP, which would find the corresponding PRP given an ERP point, and
vise versa, and give us the disparity between the two.
The new concept that we will be looking into would be drop-down menu and that will allow the user to choose which attribute he/she wants to look into. The drop-down menu wil be used for the "difference chart," where the user can dictate the x-axis variable to see different correlations. As for the bipartite visualization, we will be furthering our understanding of transitions.
For both visualizations, we hope to use the example code as the skeleton of the project. We will be tweaking the code - most of which is provided as in .js link - to fit the dataset we have and the visualization of our preferance. As for the bipartite visualization, the main changes applied will be the categorizing, to fit the data format. As for the difference chart, we will first get a firm and complete graph using one attribute, then attempt to implement the drop-down menu to allow the user to change the attribute being used for the x-axis. Once we know how to change the attribute being plugged into the graph, we can make the graph transition to be re-drawn given the new attribute.