Add spikeinterface automerge presets #199
Conversation
Awesome! I agree for the naming. In addition, we could actually reframe the similarity as a custom preset, which only runs the similarity step :) so all methods would use the spikeinterface machinery. What do you think? |
|
bravo |
|
I changed all the naming to match the "preset" language we use in spikeinterface. I like the idea of eventually making the similarity be a spikeinterface preset, but I really like the current similarity tools: you can try different similarity metrics, it's all cached nicely, it won't overwrite the |
|
@alejoe91 @samuelgarcia in mergeview, line 100 ( ) we store some merge info:for info_name in more_labels:
values = []
for unit_id1, unit_id2 in itertools.combinations(group_ids, 2):
unit_ind1 = unit_ids.index(unit_id1)
unit_ind2 = unit_ids.index(unit_id2)
values.append(self.merge_info[info_name][unit_ind1][unit_ind2])Questions:
|
Ah now I understand why @samuelgarcia had a list of supported labels...I thought we could generalize! |
|
@chrishalcrow can you test now? Should be fixed! |
|
It works - and finds some beautiful splits! 
|
| merge_info = self.merge_info[info_name] | ||
| if isinstance(merge_info, np.ndarray) and \ | ||
| merge_info.shape == (len(unit_ids), len(unit_ids)): | ||
| for unit_id1, unit_id2 in itertools.combinations(group_ids, 2): | ||
| unit_ind1 = unit_ids.index(unit_id1) | ||
| unit_ind2 = unit_ids.index(unit_id2) | ||
| values.append(merge_info[unit_ind1][unit_ind2]) | ||
|
|
||
| if max_group_size == 2: | ||
| row[info_name] = f"{values[0]:.2f}" | ||
| else: | ||
| min_, max_ = min(values), max(values) | ||
| row[f"{info_name}_min"] = f"{min_:.2f}" | ||
| row[f"{info_name}_max"] = f"{max_:.2f}" | ||
| else: | ||
| min_, max_ = min(values), max(values) | ||
| row[f"{info_name}_min"] = f"{min_:.2f}" | ||
| row[f"{info_name}_max"] = f"{max_:.2f}" | ||
| if info_name in labels: | ||
| labels.remove(info_name) | ||
| elif f"{info_name}_min" in labels: | ||
| labels.remove(f"{info_name}_min") | ||
| labels.remove(f"{info_name}_max") |
There was a problem hiding this comment.
For @samuelgarcia
Changed the logic here to be able to show any extra outs with shape (num_units, num_units), instead of pre-selecting potential labels
| num_channels = 32 | ||
| num_units = 16 | ||
| elif case == 'medium_split': | ||
| elif case == 'medium-split': |
There was a problem hiding this comment.
changed this to - to have this line work:
test_folder.stem.split('_')[-1]
3 participants
Compute potential automerges using presets inherited from spikeinterface (and still do similarity, since it's so nice and simple!)
Possible improvement: use the spikeinterface nomonclenture of presets instead of methods?