In the following cases, search returns non-relevant results:

• cse8: Zero results, expected CSE 8A/CSE 8B.
• cse 8a: First result is LIGN 8, expected CSE 8A.
• dsc 40: First result is DSC 10, expected DSC 40.
• quar: Zero results, expected courses with Quarfoot, David James.

Hi, I'm Shuhua, and I noticed that on your website you said that you'll be making a course scheduler. I wasn't sure how else to message you, but I'm actually working on one too - in C. I just started recently. Perhaps we could work together?

The seascape.app website is currently missing a website icon. Consider designing one so that bookmarks and tabs for Seascape look nicer :)

What it looks like right now:

It would make the website more convenient to use if I was able to start a new search from within the stats menu for a professor, instead of having to go back or go to the home page before searching again.

Dear @dcao,

I'm a current UCSD student. I really love how aesthetically pleasing seascape is.

I hope to update seascape's data with the latest CAPE datasets to help more UCSD students.

Most of the code in this repo is written in Haskell, and I was wondering if I could have your explanations on the framework you used and explanations for the methods.

I believe you're very busy with other things currently. And I'm wondering if I could have access to the blog post originally posted on this page (https://cao.st/posts/seascape).

Thank you!

Sincerely,
Rebecca

as more and more data is getting scraped into seascape, it's becoming increasingly unwieldy trying to manage all the scraped data from these ad-hoc data files. we can split the codebase into two parts:

• a data backend/api: written in python, with scraping scripts called regularly that write to a central sqlite/other sql database
• a distinct frontend which reads data from this api - we can use the existing haskell web server or use something else like svelte, phoenix/liveview, etc.

by doing this, we'll have one source of truth for all the data, instead of having to manage several different data files scraped from different places which can have overlapping data. this might also improve frontend dev.

we can also combine these somewhat - e.g. a regular phoenix or haskell app which reads from a database regularly scraped to with scraping scripts.

I'm not entirely sure what it is, but I haven't been able to get it to work.

• Basic prereq info
• Option/button to show transitive dependencies?

• TA or Prof grading?
• Plain-English translation of percentiles

• Have user rank preferences (not a "how much do you care about this 1-10," but a "rank how important each of these are to you")
• Have user input classes they're considering

For classes where the professor has only taught for one quarter, improve the visual representation instead of showing a single dot (put something saying not enough data to draw graph)

https://www.evanmiller.org/how-not-to-sort-by-average-rating.html

Right now everything we do is based on averaging out values. However, we currently obscure the underlying uncertainty about classes if they don't have enough reviews. Thus, we should start incorporating Wilson confidence intervals across the whole website. This will involve multiple changes:

• Refer to implementation in biohazard library, or the provided article.
• When using Wilson metrics, our sample size is the number of evaluations. Our approval percentages are our p-hat.
• Change rankings to use Wilson metric
• Clearly present the number of evaluations on the section page, since it's such a core part of our uncertainty evaluation.
• Change the box-and-whiskers to show 95% confidence intervals for each prof instead of Q1/Q3.
• In our sparklines, show the confidence interval in the sparklines?
• Add an "Uncertainty" section with details on our confidence intervals for all of our data, along with graphs of how our confidence has changed over time
• Or, in the report card, just add a plus/minus afterwards indicating our uncertainty, or a parenthesized (between x and y)
• for hours/gpa, we can calculate confidence intervals for sample mean using t dist

And a backend in Haskell? Appreciate the acknowledgement for the scraping code.

Author of Smarter CAPEs.

• Searching is broken currently - ordering is incorrect
• Ability to sort by arbitrary criteria (last taught, rec instr, etc.)
• By default, sort by "relevance" (search ranking)

To implement: implement fn to convert list of search results to a map from course to list of sections
fn takes arbitrary sort fn

• Write new scraping script for individual section pages
• Integrate data in section page