.Astronomy 11 Recap

I had the opportunity to be a part of the 11th edition of .Astronomy over the past week. Apart from being my first unconference (where large parts of the schedule were left open for discussions and hacks proposed that morning), it introduced me to topics that I haven’t spent nearly enough time thinking about, including mentoring, outreach-vs-inreach (where you’re approaching people vs the other way around), inclusivity, the carbon footprints of conferences and the acknowledgement of indigenous contributions. A large part of the conference was also active on Twitter, which had previously been on the fringe of my radar at best. Seeing the amount of engagement that the astro community shows on the platform made me reconsider a bit.

The sessions had a roughly even split between topical astronomy talks ranging from adaptive optics and gravitational wave detectors to breakout sessions on how to use different plotting packages in python and addressing the problems with how science is portrayed in the media.

I would like to go into more detail but in the interest of time here’s a short list of things that stuck with me post-conference:

1. Plotting in python: Nathaniel Starkman put together a great resource on plotting in python that can be found in this github repo.

2. Intro to Gaussian Processes: Mubdi Rahman gave a great introduction to using gaussian processes implemented using GPy that can be found at this repo. Although I haven’t used GPy before (preferring Dan Foreman-Mackey’s george and celerite) it might be worth investigating in the future if its convergence is faster. There was also an interesting discussion towards the end on how fast different samplers are at optimising the GP hyperparameters, and if a hamiltonian MC or nested sampling would faster than, for example, emcee. This blog post does a great job of comparing a bunch of different methods.

3. Fixing all the things: A session on lifehacks in astronomy yielded a bunch of interesting tricks, ranging from how to open a Jupyter notebook by double-clicking on it, to starting a new google doc by typing docs.new in the address bar of your browser, pip installing missing modules while running a script, the onetab extension for chrome, text/code editors like sublime, atom, and VScode, saving code in case of a fire with git fire and finding out-of-maintenance webpages with the wayback machine.

4. Glue: is an incredibly useful visualization software that allows the creation of powerful multidimensional or multipanel plots linking different datasets (by gluing them together, hence the name). Glupyter extends this by providing a front-end for jupyter notebooks. Alyssa Goodman and Erik Tollerud gave wonderful demos of this during one of the breakout sessions.

5. Notable biases in media-portrayals of current science include clickbait (sensationalizing headlines), the lone-wolf syndrome (eg. NASA or Einstein), discrediting the audience by oversimplifying, a disconnect between science and science reporting (careless language, lack of science literacy, or not acknowledging uncertainties) or creating artificial conflict. Solutions include better engagement by scientists with the public, providing context and narrative alongside science results, talking about future steps and portraying science as an incremental process, and better crediting the junior members and the large collaborative efforts that are responsible for significant results.

6. The current and previous .astronomy hacks are collected and can be found at the .Astronomy hacks collector. My own hack was a way for me to play around with the word2vec that translates words or groups of words to vectors in a large-dimensional space. This can then be used to probe associations and generate semantic maps between different word-clusters. I used doc2vec on a set of ~ 38k astro-ph.GA abstacts from ArXiv to learn context from papers and use it to generate a list of similar papers given an input set of keywords or an ArXiv id, which can be found at the chaotic_neutral repo. The advantage of this is that it allows for adding together or subtracting papers and keywords to uniquely specify the context in which to search for similar vectors, providing a useful complement to ADS, VoxCharta, Google Scholar and Arxivsorter while performing a literature survey. Combining this with author affiliations over the last year allows us to extend this to finding the places in the world that are actively researching a certain topic or set of topics - particularly useful for folks looking for places to apply for grad school or postdocs. I’ll try to go over this in a more detailed post at a later date.