Participants (22):

• Mickael R, Mat S, Young-Lo K, Martin B (Lyon), Joel J, Ariel G, Suhail D, Mattia B, Ana S-C (Stockholm), Maxime D (Trinity), Melissa A, Philippe R, Manu G (Cleremont), Jakob N, Nicolas M, Simeon R, Valery B (Berlin), Nicolas R (Paris), Bastien C, Ben R, Julian B (Marseille), Eleni (!)

• It snowed last week.
• Classifications are now getting routinely reported to TNS
  • It's also annoyingly overwriting the redshifts
  • ZTF21aantxma is a prime example of this
• Can we get a SEDm spectral series for objects caught early (e.g. ZTF21aaomuka): Joel to talk to Shri and Matthew G
• Can we do something with 2018 SNe that were actually in 2019. Pre-explosion epochs? These are likely poor subtractions

• Looking at how well the SEDm pipeline is producing classifications.
  • What is the effect of CR rejection and background removal?
• Looking at objects in 2018 and comparing with BCS. Has downloaded all data
  • There are some issues with the contour-separation: the WCS solution is off so the wrong area is excluded. Need to fix this.
• Looking at two effects: CR rejection (byecr) and Background subtraction (contsep)
• Needs to quantify fit quality: considering SNID rlap, and fraction of templates.
• Breaking things down in to monthly segments.
  • The cosmic ray removal is improving all of the fits (inferred through the mean rlap over all objects).
  • The 'failure fraction' (fraction of unclassified spectra) is similar.
  • Need to improve the contsep solutions.
• In comparing to BCS: compare all subtypes or just Ia's? Maybe also compare with snIascore?
• Could also add host galaxy information into the fits: combine with Jeremy's code.
• Questions include: how secure is the cosmic ray selection? 5sigma
• Drafting a paper

Nicolas on Machine Learning / Photo-typing:

• Working with Jakob N in Berlin
• Looking at Sequential Deep Learning and Early Detection for Spectroscopy.
• Sequential Deep Learning:
  • How can we automatically categorize a transient? This is an ill-posed question: what's the classification metric?
    • Looking at classification of bright sources (<18.5) into types; Ia/non-Ia for faint SNe (>19), and subtyping.
  • Trained on alert photometry, with RCF 2018 sample + AGNs/CVs as truth.
    • No host redshift considered.
    • The dataset is heterogeneous.
    • Fitting in flux-space with a 60 day light-curve.
  • Looking into the RAPID software. Works on sims (Plasticc); not on ZTF data.
    • Written a function to input the data formats, etc.
    • Also adapting the hyperparameters.
    • The training sample is highly unbalanced. Resampling helps.
    • Bright transients are better classified. AGN seem to be well classified.
  • Looking to classify faint objects. Developed a method to simulate faint SNe from bright ones.
  • Can combine classifications to get an ensemble classification: works well, but of course ~40% of objects are then unclassified.
  • All available on github!
  • Thinking about retraining on full ZTF-I sample.

Suhail on DESI:

• Spoken to Maayanne about a Citizen Science project.
• The plan is to target the hosts of all transients in DESI-SV.
• Jakob: the objects are not in the DESI target list.... A change in policy with MOUs. They might get targeted, but we might not know!
• Need to find some clarity with this!

ZTF Ia Phone-con: 2021-Mar-17

Meeting Agenda:

1. SNe of the week [Joel J]

2. The status of DESI [Suhail D]

3. Machine Learning with ZTF [Nicolas]

4. Spectroscopic Classifications [Young-Lo K]

5. AOB / Close

Participants (XX):

• Mickael R, Mat S, Young-Lo K, Martin B (Lyon), Joel J, Ariel G, Suhail D, Mattia B, Ana S-C(Stockholm), Kate M, Jacco T, Luke H, Maxime D, Mark M (Trinity), Melissa A, Philippe R, Manu G (Cleremont), Bastien C, Benjamin R (Marseille), Nicolas R (Paris)

Notes:

Transients of the Week:

Suhail on DESI:

Nicolas on Machine Learning:

Young-Lo on Spectroscopy:

Closing Remarks:

• Sign up for a talk!