Vis for showing the samples that contain tokens which maximally activate a particular e.g. neuron or direction:
This vis is very similar to TextNeuronActivations, with the difference being that this vis contains a different set of samples for each layer/neuron selection, whereas TextNeuronActivations keeps the samples fixed and shows different activation values for each layer/neuron.
(Note, this PR uses functionality introduced in #21 and #34. All the commits from those PRs are included here, so it would be cleaner to merge those PRs first.)
Vis for showing the samples that contain tokens which maximally activate a particular e.g. neuron or direction:
This vis is very similar to TextNeuronActivations, with the difference being that this vis contains a different set of samples for each layer/neuron selection, whereas TextNeuronActivations keeps the samples fixed and shows different activation values for each layer/neuron.
(Note, this PR uses functionality introduced in #21 and #34. All the commits from those PRs are included here, so it would be cleaner to merge those PRs first.)