Use logistic regression scorer to partition your patch of sky into physical objects

[MatthewJA]

Each object will have some number of TGSS and NVSS components.

[MatthewJA]

Look at some of these. Are you missing components? Are you adding extras? etc

[daccordeon]

Comparison below at 5 degree fov around J101436.8-270532.
I reckon that, as expected, we're adding lots more components, especially in the central large multi-object. I will find an example where we might be missing something (like some aged triple with the two "jet-like" blobs further apart and elongated).

torch_lr_partition.pdf

[MatthewJA]

Oh that's a super good plot. Great work!

(You're more likely to pick up too many components than too few. I suggest finding an example of a partitioning that is OK, and one that is not.)

[daccordeon]

Thanks! Will search for two blobs close together with nothing else nearby as an OK example.
See #32

[daccordeon]

This looks like a candidate for two aligned ellipses:
J121923+054926

Sometimes a blob is large enough to count as multiple sources, this is a not very interesting form of multi-component:
TGSS J153418.0+101644, etc.
NVSS J153418+101544, J153414+101647, J153417+101708, J153417+101821

Also, looked up the most distant radio galaxy (looks very boring):
TGSS J153049.9+104933, NVSS J153050+104931

[daccordeon]

Here's one such case of single match partition which seems alright.
TGSS J104025.2-162920, NVSS J104025-162920
Why it hasn't jumped to the nearby blobs, I don't know, but I would agree with the label.

[daccordeon]

Move to close this and issue #32