Figure 1: Proportion of outputs shown for 20 example neurons. This feature measures the proportion of a neuron’s outputs which go onto various different classes of downstream neuron. Colors denote different broad cell type classes (from table aibs_metamodel_mtypes_v661_v2).
Feature distances from final
To simplify things, we can consider these features in terms of how similar they are to that feature computed at the end of proofreading.
Math
Let be the feature for a given neuron at time index , and let be defined similarly for the final state of the neuron after every proofreading edit has been applied. Now, we can compute
where is some suitable distance function/metric, for instance, euclidean distance, Jensen-Shannon divergence, etc.
864691135082840567
864691135132887456
864691135135922201
864691135213953920
864691135292201142
864691135359413848
864691135502190941
864691135503003997
864691135518510218
864691135561619681
864691135564974423
864691135586352764
864691135660772080
864691135697284250
864691135808473885
864691135919630768
864691135995786154
864691136066728600
864691136618908301
864691136903387826
Figure 2: Distance from final neuron state over time in the historical ordering for the proportion of inputs metric.
Figure 3: Euclidean distance from final neuron state plotted for all 219 neurons. Red line shows the mean +/i 95% bootstrapped confience intervals.
TODO: Some kind of vertical histogram - just not sure what the normalization here should be
Clean-and-merge ordering
864691135082840567
864691135132887456
864691135135922201
864691135213953920
864691135292201142
864691135359413848
864691135502190941
864691135503003997
864691135518510218
864691135561619681
864691135564974423
864691135586352764
864691135660772080
864691135697284250
864691135808473885
864691135919630768
864691135995786154
864691136066728600
864691136618908301
864691136903387826
Figure 4
864691135082840567
864691135132887456
864691135135922201
864691135213953920
864691135292201142
864691135359413848
864691135502190941
864691135503003997
864691135518510218
864691135561619681
864691135564974423
864691135586352764
864691135660772080
864691135697284250
864691135808473885
864691135919630768
864691135995786154
864691136066728600
864691136618908301
864691136903387826
Figure 5
Outputs by distance metric
TODO: same treatment as the above
Summary
Neuron gallery
864691135082840567
864691135132887456
864691135135922201
864691135213953920
864691135292201142
864691135359413848
864691135502190941
864691135503003997
864691135518510218
864691135561619681
864691135564974423
864691135586352764
864691135660772080
864691135697284250
864691135808473885
864691135919630768
864691135995786154
864691136066728600
864691136618908301
864691136903387826
Figure 6: Summary gallery (click images to zoom in and scroll through) of various metrics and edit application methods, here showing the distance to the final state (euclidean distance).
