06/13 - 06/19:
This was my first week of research with Professor Hullman at the iSchool at University of Washington. I spent the week familiarizing myself with the campus and with Professor Hullman's past work. I read her past papers and also went through brainstorming ideas and related papers for the Expectation Visualization project. By the end of the week, I had had multiple meetings with Professor Hullman and her Ph.D. student Yeaseul Kim. We started to sketch out initial tasks and goals for the coming weeks. Professor Hullman generously considered my interest in learning more coding for graphical interfaces. So my first task was to start looking for resources to learn the Javascript package, D3.
06/20 - 06/26:
Into my second week at UW, I had familiarized myself with the Data Science lab and the work culture. I had been added to the iSchool's webserver, and was learning to use it for testing my own code. With Professor Hullman's help and resources from her own lectures of teaching D3, I had started my D3 lessons. I used many online resources, like Scott Murray's D3 tutorials. Using open source code, including Mike Bostock (the creator)'s examples, I had started writing my own simple D3 code.
The style of D3 coding is very different from any other programming language I have ever used. It was very frustrating when I couldn't understand why I couldn't get something, that would have been very simple in any other language, to work in D3. It took me a long time to get used to the nesting and chaining methods in D3, but I slowly started to get used to it. I started out with a static bar chart created by binding data to svg elements. With guidance from Yeasuel, by the end of the week I had created a bar chart that users could interact with.
06/27 - 07/02:
07/05 - 07/10:
07/11 - 07/17:
My talk at the weekly lunch went really well. This was the first time I got to showcase my work and talk about my progress in front of an audience outside of the immediate members of my lab. I was proud to be able to hold the attention of a room full of experienced researchers in Information Visualization. I was also proud to be presenting to Jeffrey Heer, whose lab developed D3, the main programming package I am learning to code these interfaces. He also advised Mike Bostock, who is one of the main developers of D3. I was especially delighted when Jessica congratulated me for giving a good talk. At the end of the talk, we got constructive feedback for ways to improve our final study and also our current interface to find the best drawing design to use in our final study.
The rest of the week I worked on integrating all the feedback. This was the first week I started getting really frustrated with the work. I had to change the overall size of the chart, which ended up rendering the mechanisms of the drawing options faulty. I had to recode most of the measurements and mechanisms of the drawing options. I have been making changes to the interface and am currently on my 5th major revised version. Still, I've come to really appreciate the importance of iteration, especially in interactiuve design from work I've done in the past. I know that the end result of all these tweaks and changes will be well worth all the work. I relieved the frustration by reading books on visualization and working on the overall layout of the interface, to get a change of pace and work. It really helped to take a break from the monotonous coding of the drawing designs. I am really grateful for how patient and considerate Jessica has been about the pace of project. She has been very helpful and motivational throughout the process.
The final layout for the interface looks as follows. Some of the changes I've made since last week are changing the sizes of circles in the 25 count chart to correspond to 4 times those in the 100 circles chart. I also changed the feedback method to rankings instead of picking options, in order to get more well-rounded feedback. I also added prompts to alert the user when they enter an invalid ranking or don't enter a ranking for each option. The prompt prevents the user from moving on to the next page until they fulfill the requirements. I am still adding features like randomizing the order of the drawing designs and an algorithm to validate whether or not the users have tried all the designs before raking them/moving on to the next page.
Given that we have made so many improvements on the interface, we are now planning on sending it out to a wider audience within the UW campus to get a better sense of which design will work best.
______________________________________________________
______________________________________________________
______________________________________________________
______________________________________________________
07/18 - 07/24:
We had another meeting about the progress of the interface with our collaborator, and decided to make more changes. Given the amount of iterations and work that has gone into the interface, we are going to write an independent paper about the process and the results of this survey. We are making the interface more focused on getting the users to use the designs to replicate a distribution and rate the designs on how satisfying and efficient each design is to complete the task.
We also decided to add more drawing options such that each drawing design has a 20 and a 100 count version for the discrete distribution interfaces. For the continuous distribution format, there will be a version with a line graph and another with a line graph that has the area under the graph filled in.
07/25 - 07/31:
08/01 - 08/07:
This week we spent a lot of our time analyzing the results we got from running our survey with the DUB list members. There were some definite choices for the least favorite, and an obvious winner for the continuous format of representation. But we couldn't decide which one clearly was the most liked and efficient at the same time for the discrete format. So we decided to run the survey again on Amazon's Mechanical Turk. We ran the study with the master turkers, who have a lot of experience and a good reputation with tasks on Turk. My job then was to set up the survey such that it was ready to go on the Turk platform. This was my opportunity to use the reading I had done on how the Turk platform works during the early weeks of my work with Jessica.
Since the Turk platform doesn't directly allow us to grab information (other than worker information), we had to set it up so that their platform would redirect to our server and then back to theirs again. Since I'm completely new to php, it took me a while to get the transfer of information to work. I used a lot of Yeaseul's old studies on Turk as an example to get my version to work. When it was finally ready, we put the study up on Turk. This was my first time running a survey on such a large platform, and I was both nervous and excited. I was surprised by the quick output we got; we had 80 results in less than half a day. I parsed the results using the same code I'd written for our survey last week and Jessica analyzed both the survey results together.
While Jessica was working on analyzing the data, I started working on our next project, where we use Expectation Visualization on uncertainty. The project is to run a study using the best drawing tool that we decide on from the survey. The users will be allowed to predict the probability distribution for a future replication of a specific study that has been run before. Once they make their prediction using the drawing tool we developed, we would overlay the true distribution over their drawing so they can compare and contrast the two. My first task was to create the true distribution using our interfaces and then find a way to have it overlaid on their drawing once they're done predicting. Since we had only decided on the continuous version, I worked on getting the continuous version to have the true distribution show up when users hit "Continue".
______________________________________________________
08/08 - 08/14:
This week I worked on the discrete outcomes interface for our second study. After the analysis was complete, we came to the conclusion that the 20-circle version of the pull-up interface was the best drawing option for discrete techniques. I used the same method as I used for the continuous version to have the true distribution show up overlain on the user's predicted distribution. Since the discrete version consisted of individual circles, I had to modify the code I used for the continuous version.
This week I also started putting together my version of the paper summarizing the work I did over the summer. We had collectively been working on a paper on ShareLatex that will cover both of our studies. The paper will be submitted to the CHI conference later this year.
08/15 - 08/19:
This was my last week so I helped put finishing touches to the interface for the second survey. I also helped make the images that will be used in our paper. I also spent come time on my version of the paper.