Week 5 (July 26th - July 29th)

July 26th:

In our meeting today we discussed running basic tests on CLIP Model B/32, as we found it gave the best results, to see if it will be able to recognize politicians. We decided that I should run three different kinds of tests a Easy, Medium, and Hard test. After completing these tests I was tasked with trying to brainstorm more test ideas based on the articles I had read.

The basis of the Easy, Medium, and Hard tests where that I would be giving CLIP different text prompts to run on each different type of Politician which would get increasingly difficult each test. I would run these tests seperately on each set of Politicians, Senate, House, and Mayors. I was given the task to come up with the different text prompts for each tests given a suggestion of some words and 3-6 words per test. For the next couple of days I began trying to write different Python scripts that would accomplish this.

July 28th:

When I met with Vicente today I showed him the results of the tests I ran based on a text script, however, it was hard to analyze these results. For this reason, Vicente suggested that I place the results in tables to have the capability of analyzing them. Furthermore, he helped me figure out a table layout and gave suggestions on which words I should keep using or which words I could switch out. In addition to suggesting this we also talked about me creating data report that included how many images I had, what size they were, of who they were, etc. Moreover, Vicente was very helpful in giving me some pseudocode for making webpages in github and how to resize all the images in a folder.

Today and the day after I began running the tests again this time creating several tables with aggregated results. Here is an example of one I finished.

Upcoming

• Create the webapge with individual results
• Create a table with aggregated results
• Data Report
• Looking for demographic information about the senators (gender, age, ethnicity, race) — lower priority