Speaker: Roy Williams
Institute: University of Edinburgh, Edinburgh, UK
Title: The UK Alert Broker
Abstract: Lasair is unique among the community brokers because its primary focus is not classification of active objects, but rather allowing users to build filters that select interesting alerts from the whole stream. Filters are defined by SQL-like language that acts on sky context, light-curve features, watchlists, sky area, and the classifications of other brokers. The system is extensible: users can build an annotator on an external system that can compute and classify, then push their results back to Lasair, for use by others in filtering.
Was talking about uh Les thank you all for being here I understand coffee break is next um I’d like to uh thank Matthew Graham if he’s still here uh for reminding us that uh it’s been uh 20 years since we started thinking about the Deluge of events and
Or at least it will be uh when it actually starts um Lassa is uh our broker um which is Edinburgh and Belfast and Oxford and it’s it’s designed to have an easy onramp uh so that it’s easy for an astronomer to to get started you know if
For example you’re interested in uh some kind of object you know uh Aman ven or something like that and you you know who they all are where they all are you can just ask tell me when any of my guys have has an outburst you can ask very simple
Questions so we’ve tried to make the on-ramp easy but we’ve also tried to make things competent now speaking of competent does this thing do ah there we go um let’s see we had a few uh jewels in the last month or so uh we’ve had a
Couple of uh uh uh rather fast uh tdes uh that we found uh since uh Christmas with uh Matt nickel and his collaborators in Belfast um we’ve had some uh uh new uh fresh Killin overs from Michael Fulton and uh his uh scheme which I’ll
Explain uh and uh there are some uh uh these white dwarf mergers and uh calcium rich supern noi and uh uh all sorts of interesting things are happen happening what Lassa does is to find the jewels it doesn’t just sort of produce huge stamp collections of supernova it produces the
Jewels the things you actually want um recently ztf has been putting force photometry in with their stream and uh the reason for that is uh that when you see a single five Sigma Alert in the actual stream you can ask for Force photometry what was it before and
That gives you a handle on the rise time and so you can detect these very fast Rises uh so now we have that uh available through through Lassa um you know I was at the Lego store in in Edinburgh the other day and um I realized that when I was a boy we
Used Lego in a different way uh these days it’s as if people buy these kits that make something up for you that you pick one of the classifi that’s already been built and then you can you know get the results from it but but with Lassa it’s more like a whole
Set of bricks and you just sort of put them all together in whatever way you want we’ve used um SQL as the glue between our bricks that ever since the SLO survey astronomers have become used to using SQL and so we thought this was a good channel to bring them on board
And uh once you’ve built your SQL query you can share it with other people you can use it as a springboard for more complicated analysis and annotations uh so the bricks are like her features Sky context watch lists watch Maps which is like a Gra gravitational wave uh uh uh uh Sky Map
The transit name server and the annotations that you yourself and other people have uh produced uh so okay this is a bit nerdy isn’t it um there’s some actual SQL up here uh see where it says uh objects. gmag and objects. JD Max the latest uh Julian day as latest detection and uh
You want the GM to be not null and you want it to be less than 15 so this is a way to filter out if if you want bright things it’s a very simple way of getting bright things and there’s all sorts of other stuff there’s the schema browser
There’s you can save these queries you can share these quer uh sorry filters we’re supposed to call them filters you can share them and save them and all of that kind of stuff uh so once you’ve made a filter of course you can click on on it we also
Have a very competent API through which you can run the uh uh uh the query so that’s where the human says okay give me the results of the query but you can also say I want this as a streaming query uh and that means that as the data
Comes in it runs your filter and uh immediately notifies you um you can get results for the 20th century people you can get email and for the 21st century people you can get Kafka and uh we have a marshall notebook connected up to the other end of the CFA
So that you can uh go through what’s happened and decide whether you like it or not and so on and so forth and right at the bottom here you see it says public uh if you want to allow other people to see your uh filter other people can then duplicate
That filter and modify it in ways that they uh wish to a watch map is uh well thank you thank you to the taxpayers of France for um funding the mock project uh these wonderful files that uh Express a subset of the sky and uh these uh you can
Upload one of these mock files on the left you see the SLO uh footprint uh which is uh uh if you want to get alerts that fall within the SLO footprint on the right is the real reason for uh this functionality which is uh to do uh Sky
Maps and uh we’re about to uh release our system for dealing with uh gravitational wave alerts uh we also have uh an intelligent Sky context system called Sherlock that’s been developed in Belfast over many years and given a position in the sky it tells you what’s there
Uh it picks from a number of these cataloges uh that you can read um you know the CVS and the uh the agns and and all the rest of it and does an intelligent classification uh I won’t go into the details of what that means and it makes a classification from uh those
That you see on the right uh so when it says uh Supernova it actually means that uh there’s a host Galaxy and when it says nuclear transient it means that uh the transient is close to the center of a host Galaxy and you can see uh the little uh
Smoking pipe up at the top which uh probably not very not very woke these days is it a tobacco pipe anyway that’s the Sherlock symbol uh we’ve built a number of L curves of of light curve features in addition to what Ruben has provided already Ruben has done a great great job
With periodic uh features and uh stochastic features and we’ve tried to look at light curves uh where you don’t have very many points and uh one for example is uh the uh flux jump that’s this guy who says I want to know when one of my sources suddenly gets bright
What they can do is just monitor the flux jump over on the right there’s this idea of we have six filter instead of six filters with Z with lsst whereas we only had two with ztf and so maybe we can think of a surface of time and wavelength uh rather than treating uh
The the filters independently and you know here you can see uh trying to fit exponentials trying to fit uh uh Basin multiplied by a black body in in wavelength so it’s a 2d fit and let’s see which of them does the best job uh let’s see here’s how to find a
Title disruption events with Lassa this is uh a bit of SQL is that okay to show a bit of actual code it’s all right isn’t it um so first you ask Sherlock um does it have a host Galaxy and that’s this second line the classification is in either Supernova or nuclear transient
And you ask for the separation in AR seconds to be less than half uh an arcc uh the third line is about we want more than two uh points in the light curve and then G minus r the color and then uh uh we want things brighter than
20th magnitude and we want uh something which is away from the Milky Way the galactic uh latitude then we uh combine that SQL on the light curve with uh this watch list uh in this case um there’s a paper by these people French and zukov uh saying that they think these
55,000 galaxies are especially liable to have nuclear transients and so you combine the SQL with the watch list and out pops The tdes annotation is where um uh the CFA stream from a prefilter goes into your own code that runs on your own computer and it does some computations on the
Light curve or other aspects and then pushes that back into the Lasser system and uh so one of them for example is fast finder which is made to find fast rising sources and it produces a piece of Json as its uh output such as that shown um here’s the uh workflow uh you
Would have an active filter one of these streaming filters uh which you see uh at the top uh in purple and then that goes over to uh your own machine which does the processing and light curves and stuff and uh then uh makes a decision about whether to annotate or not and
Pushes it back into Lasser uh so here’s uh that uh cycle where you have a pre-filter at the top left then there’s this fitting by uh the fastfinder system which then causes a Lassa annotation a discussion on slack and then off to Gemini to get a
Spectrum um we we’ve also sort of rather naughtily uh chose stolen stuff oh no uh no we have with permission taken uh kfka feeds from alers and think and um so you could for example uh use different systems to find kovi and say I only want the ones where um fast finder thinks
It’s uh uh a kill andova and think thinks it’s a kill andova so let’s hope we can do more um working together in this kind of way uh okay so this is it um as we heard yesterday I like this quote lsst variability greatly exceeds the parameter space of training sets from
Richard Anderson is he here yes thank you thank you for that um and and that’s why we’ve gone with this approach of computing features because that’s uh an unbiased way of um characterizing a light curve rather than building the classifiers which sort of tells you what you might expect to get before you’ve
Actually seen the stuff so please give it a try um l. ls. ac.uk and are there are any questions yeah any