Skip to content

improve tdc clustering options #3658

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 9 commits into from
Feb 19, 2025
Merged

improve tdc clustering options #3658

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
6055526
improve tdc clustering options
samuelgarcia Jan 30, 2025
84b82f1
Merge branch 'main' of github.com:SpikeInterface/spikeinterface into …
samuelgarcia Jan 31, 2025
73c4f71
finally remove the median removal in peak detection
samuelgarcia Jan 31, 2025
9a82a55
Merge branch 'main' into improve_split
samuelgarcia Feb 11, 2025
dff9146
clean
samuelgarcia Feb 11, 2025
749c0e9
more clean
samuelgarcia Feb 11, 2025
0e745ef
Merge branch 'main' into improve_split
samuelgarcia Feb 19, 2025
7aa36ae
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Feb 19, 2025
c03ce47
Merge branch 'main' into improve_split
yger Feb 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions src/spikeinterface/sorters/internal/tridesclous2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,16 @@ class Tridesclous2Sorter(ComponentsBasedSorter):
"selection": {"n_peaks_per_channel": 5000, "min_n_peaks": 20000},
"svd": {"n_components": 6},
"clustering": {
"recursive_depth": 3,
"split_radius_um": 40.0,
"clusterer": "hdbscan",
"clusterer_kwargs": {
"min_cluster_size": 10,
"min_samples": 1,
"allow_single_cluster": True,
"cluster_selection_method": "eom",
},
"do_merge": True,
"merge_radius_um": 40.0,
"threshold_diff": 1.5,
},
Expand Down
6 changes: 5 additions & 1 deletion src/spikeinterface/sortingcomponents/clustering/merge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,7 +99,7 @@ def merge_clusters(
fig, ax = plt.subplots()
clusterer = hdbscan.HDBSCAN(metric="precomputed", min_cluster_size=2, allow_single_cluster=True)
clusterer.fit(pair_values)
print(clusterer.labels_)
# print(clusterer.labels_)
clusterer.single_linkage_tree_.plot(cmap="viridis", colorbar=True)
# ~ fig, ax = plt.subplots()
# ~ clusterer.minimum_spanning_tree_.plot(edge_cmap='viridis',
Expand Down Expand Up @@ -294,6 +294,10 @@ def find_merge_pairs(
template_locs = channel_locs[max_chans, :]
template_dist = scipy.spatial.distance.cdist(template_locs, template_locs, metric="euclidean")

# print("template_locs", template_locs.shape, template_locs)
# print("template_locs", np.unique(template_locs[:, 1]).shape)
# print("radius_um", radius_um)

pair_mask = pair_mask & (template_dist <= radius_um)
indices0, indices1 = np.nonzero(pair_mask)

Expand Down
4 changes: 2 additions & 2 deletions src/spikeinterface/sortingcomponents/clustering/split.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,10 +205,10 @@ def split(
recursion_level=1,
debug_folder=None,
clusterer="hdbscan",
clusterer_kwargs={"min_cluster_size": 25, "min_samples": 5},
feature_name="sparse_tsvd",
neighbours_mask=None,
waveforms_sparse_mask=None,
clusterer_kwargs={"min_cluster_size": 25},
min_size_split=25,
n_pca_features=2,
minimum_overlap_ratio=0.25,
Expand Down Expand Up @@ -261,7 +261,7 @@ def split(
if clusterer == "hdbscan":
from hdbscan import HDBSCAN

clust = HDBSCAN(**clusterer_kwargs)
clust = HDBSCAN(**clusterer_kwargs, core_dist_n_jobs=1)
clust.fit(final_features)
possible_labels = clust.labels_
is_split = np.setdiff1d(possible_labels, [-1]).size > 1
Expand Down
90 changes: 55 additions & 35 deletions src/spikeinterface/sortingcomponents/clustering/tdc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,16 @@ class TdcClustering:
},
"svd": {"n_components": 6},
"clustering": {
"recursive_depth": 3,
"split_radius_um": 40.0,
"clusterer": "hdbscan",
"clusterer_kwargs": {
"min_cluster_size": 10,
"min_samples": 1,
"allow_single_cluster": True,
"cluster_selection_method": "eom",
},
"do_merge": True,
"merge_radius_um": 40.0,
"threshold_diff": 1.5,
},
Expand Down Expand Up @@ -142,18 +151,24 @@ def main_function(cls, recording, peaks, params, job_kwargs=dict()):
min_cluster_size = 50
# min_cluster_size = 10

clusterer = params["clustering"].get("clusterer", "hdbscan")
clusterer_kwargs = params["clustering"].get("clusterer_kwargs", {})

post_split_label, split_count = split_clusters(
original_labels,
recording,
features_folder,
method="local_feature_clustering",
method_kwargs=dict(
clusterer="hdbscan",
clusterer_kwargs={
"min_cluster_size": min_cluster_size,
"allow_single_cluster": True,
"cluster_selection_method": "eom",
},
clusterer=clusterer,
clusterer_kwargs=clusterer_kwargs,
# clusterer="hdbscan",
# clusterer_kwargs={
# "min_cluster_size": min_cluster_size,
# "allow_single_cluster": True,
# # "cluster_selection_method": "eom",
# "cluster_selection_method": "leaf",
# },
# clusterer="isocut5",
# clusterer_kwargs={"min_cluster_size": min_cluster_size},
feature_name="sparse_tsvd",
Expand All @@ -164,39 +179,44 @@ def main_function(cls, recording, peaks, params, job_kwargs=dict()):
n_pca_features=3,
),
recursive=True,
recursive_depth=3,
recursive_depth=params["clustering"]["recursive_depth"],
returns_split_count=True,
debug_folder=clustering_folder / "figure_debug_split",
**job_kwargs,
)

merge_radius_um = params["clustering"]["merge_radius_um"]
threshold_diff = params["clustering"]["threshold_diff"]

post_merge_label, peak_shifts = merge_clusters(
peaks,
post_split_label,
recording,
features_folder,
radius_um=merge_radius_um,
# method="project_distribution",
# method_kwargs=dict(
# waveforms_sparse_mask=sparse_mask,
# feature_name="sparse_wfs",
# projection="centroid",
# criteria="distrib_overlap",
# threshold_overlap=0.3,
# min_cluster_size=min_cluster_size + 1,
# num_shift=5,
# ),
method="normalized_template_diff",
method_kwargs=dict(
waveforms_sparse_mask=sparse_mask,
threshold_diff=threshold_diff,
min_cluster_size=min_cluster_size + 1,
num_shift=5,
),
**job_kwargs,
)
if params["clustering"]["do_merge"]:
merge_radius_um = params["clustering"]["merge_radius_um"]
threshold_diff = params["clustering"]["threshold_diff"]

post_merge_label, peak_shifts = merge_clusters(
peaks,
post_split_label,
recording,
features_folder,
radius_um=merge_radius_um,
# method="project_distribution",
# method_kwargs=dict(
# waveforms_sparse_mask=sparse_mask,
# feature_name="sparse_wfs",
# projection="centroid",
# criteria="distrib_overlap",
# threshold_overlap=0.3,
# min_cluster_size=min_cluster_size + 1,
# num_shift=5,
# ),
method="normalized_template_diff",
method_kwargs=dict(
waveforms_sparse_mask=sparse_mask,
threshold_diff=threshold_diff,
min_cluster_size=min_cluster_size + 1,
num_shift=5,
),
**job_kwargs,
)
else:
post_merge_label = post_split_label.copy()
peak_shifts = np.zeros(post_split_label.size, dtype="int64")

# sparse_wfs = np.load(features_folder / "sparse_wfs.npy", mmap_mode="r")

Expand Down
5 changes: 3 additions & 2 deletions src/spikeinterface/sortingcomponents/peak_detection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -678,8 +678,8 @@ def __init__(
random_data = get_random_data_chunks(recording, return_scaled=False, **random_chunk_kwargs)
conv_random_data = self.get_convolved_traces(random_data)
medians = np.median(conv_random_data, axis=1)
medians = medians[:, None]
noise_levels = np.median(np.abs(conv_random_data - medians), axis=1) / 0.6744897501960817
self.medians = medians[:, None]
noise_levels = np.median(np.abs(conv_random_data - self.medians), axis=1) / 0.6744897501960817
self.abs_thresholds = noise_levels * detect_threshold
self._dtype = np.dtype(base_peak_dtype + [("z", "float32")])

Expand All @@ -693,6 +693,7 @@ def compute(self, traces, start_frame, end_frame, segment_index, max_margin):

assert HAVE_NUMBA, "You need to install numba"
conv_traces = self.get_convolved_traces(traces)
# conv_traces -= self.medians
conv_traces /= self.abs_thresholds[:, None]
conv_traces = conv_traces[:, self.conv_margin : -self.conv_margin]
traces_center = conv_traces[:, self.exclude_sweep_size : -self.exclude_sweep_size]
Expand Down