Classification probability

Support Lasair
#1

Hi,

I would like to ask if there is a way to get the probability of the classification for an object (e.g. if it wass classified as NT what was the probability for it to be an NT).

Thank you,
Lydia

#2

Hi Lydia,

Sherlock (the underlying codebase providing Lasair’s contextual classifications) does not provide an associated probability with the classifications. If a transient has been flagged as an NT, then this is Sherlock’s ‘best guess’ at identifying its nature.

Kind Regards,
Dave

#3

Hi Dave,

Thank you for the reply. Do you know if retrieving the history of classification for each detected epoch is possible?

Thank you,
Lydia

#4

From the web interface, no. But if you were connected to a stream containing the transient, every alert packet would have the epoch-specific Sherlock classification. @roy can you confirm this is the case?

#5

Hi @daveyoung, @roy, just following up on @lydiam’s question here on whether the classification history is stored for each epoch.

#6

Hi @ryanlau, it is not stored within Lasair presently. Can you share with us how you would use this history of Sherlock classifications?

#7

Hi @daveyoung. Thank you for clarifying. Personally, I want the history to test how fast we can have an accurate classification for follow-up. I am looking into already spectroscopically classified transients but I would like to check what was the earlier classifications with Sherlock. For example, I need to be able to separate between NT and SN as early as possible.

Thank you,
Lydia

#8

Hi @daveyoung , just following up with you on @lydiam’s question. Based on the definitions of “SN” and “NT” from the Sherlock web page:

  1. Nuclear Transient (NT) if the transient falls within the synonym radius of the core of a resolved galaxy,
  2. Supernova (SN) if the transient is not classified as an NT but is found within the magnitude-, morphology- or distance-dependant association radius of a galaxy

It seems like an object classified as an NT could very well be a supernova but just happens to be near (1.5’’) the core of a resolved galaxy – I am not sure if the classifications are updated after more information is obtained, but I would be curious to the answer on this as well.

#9

Hi,

This is the exact issue. There is high contamination of SN in the NT classification even for late-time lightcurves. @daveyoung @ryanlau

Best,
Lydia

#10

Hi both.

@ryanlau, you correctly surmise that an NT could indeed be an SN.

You must understand that Sherlock is a contextual classifier and does not consider time-series data. We all know that lightcurve fitting on the first 1-2 data points is close to useless at differentiating between various flavours of transients simply because there is limited information. However, there is much more spatial/contextual information immediately available in the form of catalogues. This is where Sherlock comes in.

The core purpose of Sherlock is to provide a blunt, low-resolution filter on streams of transients. Think of it as helping to provide you with a first pass triage on the 1000s of transients discovered every night with Rubin. So if you are interested in supernovae, then create a filter with Lasair to search for NT, SN and Orphans. After this, it is up to you to decide what you want to do with the filtered transients … wait for more data and pass to your favourite lightcurve classifier, or cherry pick a few transients for spectroscopic classification, etc.

Does that help answer your queries?

cheers,
Dave

#11

Hi @daveyoung,

Thanks for the reply, that makes sense to me. @lydiam, does Dave’s response answer your question as well? Specifically, the suggested approach to filter for NTs and then do your own light curve analysis or obtain follow up spectra sounds like the best route.

#12

Hi all,

Sorry for the delayed reply. Yes, it does make sense.

Thank you,
Lydia