search

LSSTDESC / tomo_challenge

2020 Tomographic binning challenge
13 stars 18 forks source link

Entry SimpleSOM #39

Open AngusWright opened 4 years ago

AngusWright commented 4 years ago

Simple SOM is, as the name suggests, a simplistic SOM tomographic-bin constructor.

This method is one which was discussed at the DESC Collaboration Meeting. It constructs a high-resolution Self-organising map (SOM) of the colour/magnitude space, and then associates cells of the SOM based on their colour or redshift (default) properties into N distinct groups. These N associations are then rank-ordered by their mean redshift, and are combined into num_bins tomgoraphic bins.

Example metrics will be posted below.

AngusWright commented 4 years ago

Example stats: 

SimpleSOM run_5bins {'SNR_ww': 252.65442996279617, 'SNR_gg': 1351.9975410515556, 'SNR_3x2': 1352.764455490629, 'FOM_3x2': 28973.699568005108}
SimpleSOM run_10bins  {'SNR_ww': 254.59477957187315, 'SNR_gg': 1643.0217912858973, 'SNR_3x2': 1643.5489132300042, 'FOM_3x2': 55703.92224078966}

This method isn't optimised for SNR in any way. It is therefore reasonable to expect that the final bins do not encore the full information accessible to the method.