Information Visualization - Assignment 1: Critiquing Multivariate Visualization System | CS 7450, Assignments of Computer Science

Download Information Visualization - Assignment 1: Critiquing Multivariate Visualization System | CS 7450 and more Assignments Computer Science in PDF only on Docsity! CS 7450 Rahul Nair Assignment 1: Critiquing Multivariate Visualization Systems This assignment required that the students familiarize themselves with three different commercial information visualization tools and critique them. In order to do so we were asked to select two data sets and create some hypothesis about them. The tools were then to be used to perform those specific tasks and critiqued based on observation made while doing them. Selection of Data Sets In order to evaluate the system I had to choose 2 data sets that would best allow me to explore the various capabilities of the tools. I finally decided to choose the grocery store survey and the car model data for the following reasons: Grocery store data: This data set comprised of responses to a grocery survey. It had 5163 cases each of which had 8 individual variables of which 4 are nominal and 4 are quantitative. Since the data is mainly demographic and also pertains to common shopping habits it is relatively easy to form hypothesis to be explored tools. The large number of entries allowed testing of the resolution and scalability of the individual tools. Car performance data: This data set contains 406 entries about main performance criteria of cars sold between the years 1970 and 1982. Each entry contains the data about a single car and include the name, manufacturer, fuel efficiency (miles per gallon), number of cylinders, displacement, horsepower, weight, acceleration, model year and country of origin. Thus there are 5 quantitative variables and 3 nominal variables. The other 2 variables while being numeric have such a limited range of values that they may be considered nominal. This data set had the advantage of being from a well know topic so that several know hypotheses could be tested. Hypothesis Grocery store data Due to the simple every day nature of this data set I used commonly held truths as my hypothesis to find if there was any truth to them. The main aim of the hypothesis in this was to evaluate the tools capacity to handle large data sets. The hypothesis were- 1. Purchase amount is proportional to family size This is a commonly held assumption that stems from the logic that if a family is large but does grocery shopping at the same intervals as a small family then the larger family will have a higher purchase amount since they require more items. 2. Payment method varies with income This is an urban myth/fact that people with higher incomes pay using cash no matter what is the size of the bill. Since the data in this set contains both the customers payment method and average income it will be relatively easy to prove or disprove. 3. Purchase amount is proportional to the customers income This is another logical deduction since it can be assumed that customers with a greater income can usually afford to buy more expensive brands and other luxury items. CS 7450 Rahul Nair Car performance data The cars performance data set allowed the exploration of several tried and tested hypothesis to see if any of the visualization tools gave different view of the data. 1. Acceleration of a car is inversely proportional to the weight and directly proportional to the horsepower This basically measures the effect of the weight and horsepower of the car on the acceleration of the car. The acceleration is measured in seconds (0 to 60 mph) and is thus an inverse scale with a lower time signifying better acceleration. The value for acceleration should be proportional to the horsepower and inversely proportional to the weight of the car. 2. Japanese and European cars are more fuel efficient than American cars This hypothesis was set up to test the commonly held belief that European and Japanese cars are more fuel efficient than their American counterparts. This could be due to several factors such as weight and type of the vehicle which must also be explored. Tools Spotfire Spotfire is a multivariate information visualization tool made by Spotfire, Inc. Its primary visualization tool is a scatterplot technique which can be extended from 2D to 3D with several other types of information that can be encoded in things like size, colour and orientation. It also supports several techniques to allow the user to easily understand the data such as line connections and jitter. The interface is user modifiable and incorporates several features like alpha sliders, range sliders, checkboxes, etc… The system also allows the users to add new columns or data according to any expression they choose. The navigation through the system is rather simple and simple tasks like filtering can be accomplished using the intuitive range sliders. CS 7450 Rahul Nair Payment method varies with income This was another disproved hypothesis as the data set showed no correlation between the users method of payment and their income. SeeIt also showed that the average payment amount for cash purchases was lower than the average for checks, debit cards and store cards. “Other” payment methods showed a wide fluctuation through the income range but that is probably due to insufficient samples. Eureka also intuitively displayed the relative proportions of each payment type with ease. Spotfire was unable to show average values for payment methods. SeeIt: Purchase amount Vs Family size SeeIt: Purchase method Vs Income Spotfire: Purchase amount Vs Family size Eureka: Purchase method Vs Income CS 7450 Rahul Nair Purchase amount is proportional to the customers income The tools also showed that there was no relationship between the customers income and their purchase amount, the average family size did vary much either. There was no significant difference in using the tools. Car performance data Acceleration of a car is inversely proportional to the weight and directly proportional to the horsepower This hypothesis produced an interesting result since it showed that the heavier cars usually had better acceleration. Those cars did however have very high horsepower figures. Further study showed that the cars with the best acceleration figures were all American and made in the early 70’s. This leads to the conclusion that they were probably a by product of the American muscle car generation. There was a marked difference between the performances of the tools on this task. Eureka was unable to show any clear relations between the data while Spotfire showed the individual cases in great detail. However the best tool for this hypothesis was SeeIt which allowed the easy depiction of trends over the data set using 2 variables in their respective axes while encoding acceleration and model year as colour and height respectively. Spotfire : Purchase amount Vs Income CS 7450 Rahul Nair Japanese and European cars are more fuel efficient than American cars This hypothesis was easily proved using SeeIt to display model year, origin, and average MPG per model year. Encoding colour into the data set also showed that acceleration has stayed reasonably constant over the years apart from the American muscle cars of the early 70’s. While Spotfire also showed the data it was considerably more difficult and requires the user makes judgments about the status of the display. Eureka was totally unable to show the relationships between the variables. SeeIt: Acceleration Vs Horsepower & weight SeeIt: Fuel efficiency Spotfire: Acceleration Vs Horsepower & weight Spotfire: Fuel efficiency