DARTsort 0.5.16__tar.gz → 0.5.17__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {dartsort-0.5.16 → dartsort-0.5.17}/.github/workflows/ci.yml +1 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/PKG-INFO +1 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/pyproject.toml +4 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/__init__.py +0 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/agglomerate.py +0 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/clustering.py +1 -6
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/clustering_features.py +2 -4
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/density.py +16 -12
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/merge.py +1 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/mixture.py +6 -7
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/refine_util.py +6 -59
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/config.py +3 -3
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/detect/detect.py +10 -9
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/comparison.py +83 -82
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/hybrid_util.py +8 -6
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/localize/localize_torch.py +6 -5
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/main.py +11 -9
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/grab.py +6 -4
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching.py +2 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching_util/compressed_upsampled.py +3 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching_util/drifty.py +6 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching_util/matching_base.py +3 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/peel_base.py +26 -27
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/peel_lib.py +19 -17
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/subtract.py +2 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/amortized_localization.py +12 -13
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/decollider.py +1 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/whiten.py +4 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/drift_util.py +3 -3
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/internal_config.py +7 -5
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/interpolation_util.py +17 -16
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/motion.py +12 -12
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/noise_util.py +11 -61
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/spiketorch.py +5 -4
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/testing_util/matching_debug_util.py +10 -6
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/testing_util/mixture_testing_util.py +0 -3
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/waveform_util.py +17 -23
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/mixture.py +25 -24
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/vismain.py +7 -5
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort.egg-info/PKG-INFO +1 -1
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_dartsort.py +10 -5
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_matching.py +14 -11
- {dartsort-0.5.16 → dartsort-0.5.17}/.github/workflows/ci_windows.yml +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/.github/workflows/deploy.yml +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/.github/workflows/docs.yml +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/.gitignore +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/README.md +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/docs/index.md +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/docs/main_api.md +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/docs/toolkit_api.md +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/environment.yml +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/notebook/.keep +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/notebook/ibl_dandi_example.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/setup.cfg +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/cli.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/cluster_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/forward_backward.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/kmeans.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/clustering/modes.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/detect/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/detect/detect_filters.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/analysis.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/noise_recording_tools.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/sim_template_tools.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/simkit.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/evaluate/simlib.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/localize/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/localize/localize_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching_util/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching_util/pairwise.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/matching_util/pairwise_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/reduction_template.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/shaver.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/peel/threshold.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/default_temporal_kernel.npy +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/detect_np1.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/detect_np2.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/glasbey1024.npz +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/offset_trained_denoiser.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/offset_trained_denoiser_added_temp.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/offset_trained_denoiser_added_temp_boost.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/original_single_chan_denoiser.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/pretrained/single_chan_denoiser.pt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/get_templates.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/postprocess_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/realignment.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/template_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/templates.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/templates/templib.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/_multichan_denoiser_kit.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/all_transformers.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/amplitudes.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/denoising_scorer.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/enforce_decrease.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/fixed_prop.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/interp.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/localize.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/matching_denoiser.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/mixture_classifier.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/pipeline.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/reduction.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/single_channel_denoiser.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/supervised_denoiser.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/temporal_pca.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/transform/transform_base.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/cli_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/data_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/job_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/logging_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/main_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/more_operators.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/multiprocessing_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/nn_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/peel_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/preprocess_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/py_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/registration_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/residual_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/sparse_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/spikeio.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/testing_util/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/torch_optimization_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/util/torch_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/__init__.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/analysis_plots.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/colors.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/gt.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/layout.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/metrics_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/recanim.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/scatterplots.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/sorting.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/unit.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/unit_comparison.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/versus.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort/vis/waveforms.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort.egg-info/SOURCES.txt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort.egg-info/dependency_links.txt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort.egg-info/entry_points.txt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort.egg-info/requires.txt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/src/dartsort.egg-info/top_level.txt +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/conftest.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_agglomerate.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_alignment.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_cluster_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_clustering.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_config.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_data_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_density.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_detect.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_drift_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_drifty_matching.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_enforce_decrease.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_grab_and_featurize.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_kmeans.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_linalg.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_localize_torch.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_mixture_classifier.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_modeling.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_simkit.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_spiketorch.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_subtract.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_templates.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_threshold.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_torch_optimization_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_transform.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/tests/test_waveform_util.py +0 -0
- {dartsort-0.5.16 → dartsort-0.5.17}/zensical.toml +0 -0
|
@@ -75,7 +75,10 @@ line-length = 88
|
|
|
75
75
|
|
|
76
76
|
[tool.ruff.lint]
|
|
77
77
|
extend-select = ["E", "F", "I", "N", "B", "SIM", "NPY", "RUF", "PTH", "PD", "FURB"]
|
|
78
|
-
ignore = ["F722", "N812", "SIM102", "SIM108", "E501", "N806", "N803", "N815"]
|
|
78
|
+
ignore = ["F722", "N812", "SIM102", "SIM108", "E501", "N806", "N803", "N815", "RUF005", "SIM300", "C408"]
|
|
79
|
+
|
|
80
|
+
[tool.ruff.lint.per-file-ignores]
|
|
81
|
+
"**/__init__.py" = ["F401", "F403"]
|
|
79
82
|
|
|
80
83
|
[tool.jupytext]
|
|
81
84
|
"formats" = "ipynb,py:percent"
|
|
@@ -159,12 +159,7 @@ class Clusterer:
|
|
|
159
159
|
mask_ix = np.flatnonzero(features.keep_mask)
|
|
160
160
|
else:
|
|
161
161
|
mask_ix = None
|
|
162
|
-
if self.sampling_cfg is None:
|
|
163
|
-
if mask_ix is not None:
|
|
164
|
-
return True, mask_ix
|
|
165
|
-
else:
|
|
166
|
-
return False, slice(None)
|
|
167
|
-
elif features.n <= self.sampling_cfg.n_waveforms_fit:
|
|
162
|
+
if self.sampling_cfg is None or features.n <= self.sampling_cfg.n_waveforms_fit:
|
|
168
163
|
if mask_ix is not None:
|
|
169
164
|
return True, mask_ix
|
|
170
165
|
else:
|
|
@@ -1,5 +1,4 @@
|
|
|
1
1
|
from typing import Self, cast
|
|
2
|
-
from logging import WARN
|
|
3
2
|
|
|
4
3
|
import h5py
|
|
5
4
|
import numpy as np
|
|
@@ -131,7 +130,6 @@ class SimpleMatrixFeatures:
|
|
|
131
130
|
samp = amp * np.sign(v)
|
|
132
131
|
|
|
133
132
|
if clustering_features_cfg.use_signed_amplitude:
|
|
134
|
-
samp *= clustering_features_cfg.amp_scale
|
|
135
133
|
features.append(clustering_features_cfg.amp_scale * samp[:, None])
|
|
136
134
|
|
|
137
135
|
do_pcs = bool(clustering_features_cfg.n_main_channel_pcs)
|
|
@@ -170,7 +168,7 @@ class SimpleMatrixFeatures:
|
|
|
170
168
|
pcs,
|
|
171
169
|
raise_for_numerics=clustering_features_cfg.raise_for_numerics,
|
|
172
170
|
)
|
|
173
|
-
|
|
171
|
+
_erp, pcs = interpolate_by_chunk(
|
|
174
172
|
mask=mask,
|
|
175
173
|
dataset=pcs,
|
|
176
174
|
geom=motion.geom,
|
|
@@ -190,7 +188,7 @@ class SimpleMatrixFeatures:
|
|
|
190
188
|
else:
|
|
191
189
|
assert sorting.parent_h5_path is not None
|
|
192
190
|
with h5py.File(sorting.parent_h5_path, "r", locking=False) as h5:
|
|
193
|
-
|
|
191
|
+
_erp, pcs = interpolate_by_chunk(
|
|
194
192
|
mask=mask,
|
|
195
193
|
dataset=h5[clustering_features_cfg.pca_dataset_name],
|
|
196
194
|
geom=motion.geom,
|
|
@@ -105,7 +105,10 @@ def get_smoothed_density(
|
|
|
105
105
|
)
|
|
106
106
|
nbins = np.ceil(dextents / bin_sizes).astype(int)
|
|
107
107
|
nbins = nbins.clip(min_n_bins, max_n_bins)
|
|
108
|
-
bin_edges = [
|
|
108
|
+
bin_edges = [
|
|
109
|
+
np.linspace(e[0], e[1], num=nb + 1)
|
|
110
|
+
for e, nb in zip(extents, nbins, strict=True)
|
|
111
|
+
]
|
|
109
112
|
|
|
110
113
|
# compute histogram and figure out how big the bins actually were
|
|
111
114
|
raw_histogram, bin_edges = np.histogramdd(infeats, bins=bin_edges)
|
|
@@ -178,7 +181,7 @@ def get_smoothed_density_ratio(
|
|
|
178
181
|
max_n_bins=max_n_bins,
|
|
179
182
|
min_n_bins=min_n_bins,
|
|
180
183
|
)
|
|
181
|
-
for sigma, sigma_low in zip(sigmas, sigma_lows)
|
|
184
|
+
for sigma, sigma_low in zip(sigmas, sigma_lows, strict=True)
|
|
182
185
|
]
|
|
183
186
|
if len(sigmas) == 1:
|
|
184
187
|
return dens[0]
|
|
@@ -233,7 +236,7 @@ def nearest_higher_density_neighbor(
|
|
|
233
236
|
def remove_border_points(
|
|
234
237
|
labels, density, kdtree, search_radius=1.0, n_neighbors_search=5, workers=1
|
|
235
238
|
):
|
|
236
|
-
|
|
239
|
+
_distances, indices = kdtree.query(
|
|
237
240
|
kdtree.data,
|
|
238
241
|
k=1 + n_neighbors_search,
|
|
239
242
|
distance_upper_bound=search_radius,
|
|
@@ -249,7 +252,7 @@ def remove_border_points(
|
|
|
249
252
|
|
|
250
253
|
units = np.unique(labels)
|
|
251
254
|
new_labels = labels.copy()
|
|
252
|
-
for
|
|
255
|
+
for u in units[units > 0]:
|
|
253
256
|
in_unit = np.flatnonzero(labels == u)
|
|
254
257
|
in_unit_border = in_border[labels_in_border == u]
|
|
255
258
|
if not in_unit_border.size:
|
|
@@ -266,7 +269,7 @@ def sort_density(
|
|
|
266
269
|
sigma0: float,
|
|
267
270
|
sigma1: float,
|
|
268
271
|
max_sigma: float = 3.0,
|
|
269
|
-
device: torch.device =
|
|
272
|
+
device: torch.device | str = "cpu",
|
|
270
273
|
batch_size: int = 2048,
|
|
271
274
|
col_batch_size: int = 1024,
|
|
272
275
|
show_progress: bool = False,
|
|
@@ -295,6 +298,7 @@ def sort_density(
|
|
|
295
298
|
np.ndarray
|
|
296
299
|
The density ratio for each point in X.
|
|
297
300
|
"""
|
|
301
|
+
device = torch.device(device)
|
|
298
302
|
X = torch.asarray(X, device=device)
|
|
299
303
|
order = torch.argsort(X[:, 0])
|
|
300
304
|
X = X[order]
|
|
@@ -313,7 +317,7 @@ def sort_density(
|
|
|
313
317
|
n1 = X.new_tensor(-0.5 * dim * np.log(2.0 * np.pi * sigma1 * sigma1))
|
|
314
318
|
|
|
315
319
|
if show_progress:
|
|
316
|
-
iter = progrange(0, n, batch_size, desc=f
|
|
320
|
+
iter = progrange(0, n, batch_size, desc=f"SortDens:{device.type}")
|
|
317
321
|
else:
|
|
318
322
|
iter = range(0, n, batch_size)
|
|
319
323
|
|
|
@@ -384,7 +388,7 @@ def kdt_density(
|
|
|
384
388
|
max_sigma: float = 3.0,
|
|
385
389
|
n_threads=8,
|
|
386
390
|
):
|
|
387
|
-
n
|
|
391
|
+
n = X.shape[0]
|
|
388
392
|
max_dist = max_sigma * sigma_regional
|
|
389
393
|
|
|
390
394
|
jobs = range(0, n, batch_size)
|
|
@@ -506,7 +510,7 @@ def density_peaks(
|
|
|
506
510
|
leafsize=24,
|
|
507
511
|
density_strategy="sort",
|
|
508
512
|
workers=-1,
|
|
509
|
-
device: torch.device =
|
|
513
|
+
device: torch.device | str = "cpu",
|
|
510
514
|
):
|
|
511
515
|
"""Density peaks clustering as described by Rodriguez and Laio, but...
|
|
512
516
|
|
|
@@ -584,7 +588,7 @@ def density_peaks(
|
|
|
584
588
|
has_nhdn = np.flatnonzero(nhdn < n).astype(np.intc)
|
|
585
589
|
data = np.ones(has_nhdn.size)
|
|
586
590
|
graph = coo_array((data, (nhdn[has_nhdn], has_nhdn)), shape=(n, n))
|
|
587
|
-
|
|
591
|
+
_ncc, labels = connected_components(graph)
|
|
588
592
|
|
|
589
593
|
if remove_borders:
|
|
590
594
|
labels = remove_border_points(
|
|
@@ -724,7 +728,7 @@ def gmm_density_peaks(
|
|
|
724
728
|
|
|
725
729
|
Xi = X[choices]
|
|
726
730
|
nc = len(Xi)
|
|
727
|
-
inliers,
|
|
731
|
+
inliers, _kdtree = kdtree_inliers(
|
|
728
732
|
Xi,
|
|
729
733
|
n_neighbors=outlier_neighbor_count,
|
|
730
734
|
distance_upper_bound=outlier_radius * np.sqrt(X.shape[1]),
|
|
@@ -735,7 +739,7 @@ def gmm_density_peaks(
|
|
|
735
739
|
|
|
736
740
|
Xi = X[inliers]
|
|
737
741
|
ni = len(Xi)
|
|
738
|
-
|
|
742
|
+
_uchans, cchans = np.unique(channels[inliers], return_counts=True)
|
|
739
743
|
comps_per_chan = np.minimum(
|
|
740
744
|
max_components_per_channel,
|
|
741
745
|
np.ceil(cchans / min_spikes_per_component).astype(int),
|
|
@@ -1211,7 +1215,7 @@ def _density_peaks_clustering_uhd_implementation(
|
|
|
1211
1215
|
)
|
|
1212
1216
|
|
|
1213
1217
|
graph = coo_array((np.ones(2 * has_nhdn.size), rc), shape=(n, n))
|
|
1214
|
-
|
|
1218
|
+
_ncc, labels = connected_components(graph, directed=False)
|
|
1215
1219
|
|
|
1216
1220
|
if remove_borders:
|
|
1217
1221
|
labels = remove_border_points(
|
|
@@ -185,7 +185,7 @@ def calculate_merge_distances(
|
|
|
185
185
|
np.fill_diagonal(dists, 0.0) # sometimes numerical 0 is -1e-6.
|
|
186
186
|
min_dist = dists.min()
|
|
187
187
|
if min_dist < -1e-3:
|
|
188
|
-
warnings.warn(f"Alarmingly negative min distance {min_dist}.")
|
|
188
|
+
warnings.warn(f"Alarmingly negative min distance {min_dist}.", stacklevel=2)
|
|
189
189
|
dists = np.maximum(dists, 0.0, out=dists)
|
|
190
190
|
|
|
191
191
|
return units, dists, shifts, template_snrs
|
|
@@ -1934,8 +1934,7 @@ class TruncatedSpikeData(BatchedSpikeData):
|
|
|
1934
1934
|
d[:n_units_new, :n_units_new] = distances
|
|
1935
1935
|
d[n_units_new:, n_units_new:].fill_diagonal_(0.0)
|
|
1936
1936
|
distances = d
|
|
1937
|
-
|
|
1938
|
-
self.update_adjacency(n_units=n_units_new + n_new)
|
|
1937
|
+
self.update_adjacency(n_units=n_units_new + n_new)
|
|
1939
1938
|
if pnoid and distances is not None:
|
|
1940
1939
|
_max_cand = self.candidates.amax()
|
|
1941
1940
|
assert _max_cand < distances.shape[0], f"{_max_cand} {distances.shape}"
|
|
@@ -3263,9 +3262,9 @@ class TruncatedMixtureModel(BaseMixtureModel):
|
|
|
3263
3262
|
max_group_size=max(1, self.p.split_k - 1),
|
|
3264
3263
|
)
|
|
3265
3264
|
if _stop_after:
|
|
3266
|
-
split_groups =
|
|
3265
|
+
split_groups = [
|
|
3267
3266
|
sg for _, sg in zip(range(_stop_after), split_groups, strict=False)
|
|
3268
|
-
|
|
3267
|
+
]
|
|
3269
3268
|
if show_progress:
|
|
3270
3269
|
split_groups = progbar(split_groups, desc="Split", smoothing=0.0)
|
|
3271
3270
|
|
|
@@ -4414,14 +4413,14 @@ def get_full_neighborhood_data(
|
|
|
4414
4413
|
):
|
|
4415
4414
|
in_unit = np.flatnonzero(sorting.labels == uid)
|
|
4416
4415
|
if in_unit.size > fit_count:
|
|
4417
|
-
fit_ixs = rg.choice(in_unit, size=fit_count)
|
|
4416
|
+
fit_ixs = rg.choice(in_unit, size=fit_count, replace=False)
|
|
4418
4417
|
fit_ixs.sort()
|
|
4419
4418
|
elif in_unit.size == fit_count:
|
|
4420
4419
|
fit_ixs = in_unit
|
|
4421
4420
|
else:
|
|
4422
4421
|
panic()
|
|
4423
4422
|
if fit_ixs.size > train_count:
|
|
4424
|
-
train_ixs = rg.choice(fit_ixs, size=train_count)
|
|
4423
|
+
train_ixs = rg.choice(fit_ixs, size=train_count, replace=False)
|
|
4425
4424
|
train_ixs.sort()
|
|
4426
4425
|
elif fit_ixs.size == train_count:
|
|
4427
4426
|
train_ixs = fit_ixs
|
|
@@ -5535,7 +5534,7 @@ def evaluate_group_demolitions(
|
|
|
5535
5534
|
if mean_eval_resp is None:
|
|
5536
5535
|
mean_eval_resp = mean_responsibilities(scores=eval_scores, n_units=mm.n_units)
|
|
5537
5536
|
ratio = mean_train_resp[group] / mean_eval_resp[group]
|
|
5538
|
-
can_demolish = ratio >
|
|
5537
|
+
can_demolish = ratio > mm.p.demolition_min_resp_ratio
|
|
5539
5538
|
|
|
5540
5539
|
if not can_demolish.any():
|
|
5541
5540
|
return GroupDemolition(unit_ids=group, improvement=0.0, demolished=None)
|
|
@@ -1,7 +1,6 @@
|
|
|
1
1
|
from threading import local
|
|
2
2
|
from typing import cast
|
|
3
3
|
|
|
4
|
-
import h5py
|
|
5
4
|
import numba
|
|
6
5
|
import numpy as np
|
|
7
6
|
import torch
|
|
@@ -12,7 +11,6 @@ from spikeinterface.core.baserecording import BaseRecording
|
|
|
12
11
|
|
|
13
12
|
from dartsort.clustering.mixture import Scores
|
|
14
13
|
|
|
15
|
-
from ..transform.temporal_pca import BaseTemporalPCA
|
|
16
14
|
from ..util import data_util, spiketorch
|
|
17
15
|
from ..util.internal_config import ComputationConfig, RefinementConfig
|
|
18
16
|
from ..util.logging_util import DARTSORTDEBUG, DARTSORTVERBOSE, get_logger, progbar
|
|
@@ -25,57 +23,6 @@ from .clustering_features import StableWaveformFeatures
|
|
|
25
23
|
logger = get_logger(__name__)
|
|
26
24
|
|
|
27
25
|
|
|
28
|
-
def get_noise_log_priors(noise, sorting, refinement_cfg):
|
|
29
|
-
from dartsort.templates import TemplateData
|
|
30
|
-
|
|
31
|
-
if not refinement_cfg.noise_fp_correction:
|
|
32
|
-
return None
|
|
33
|
-
|
|
34
|
-
h5_name = sorting.parent_h5_path
|
|
35
|
-
if h5_name is None:
|
|
36
|
-
return None
|
|
37
|
-
stem = h5_name.stem
|
|
38
|
-
|
|
39
|
-
if stem.startswith("matching"):
|
|
40
|
-
model_dir = h5_name.parent / f"{stem}_models"
|
|
41
|
-
templates_npz = model_dir / "template_data.npz"
|
|
42
|
-
if not templates_npz.exists():
|
|
43
|
-
raise ValueError(f"{templates_npz} is not there?")
|
|
44
|
-
|
|
45
|
-
with h5py.File(h5_name, "r", locking=False) as h5:
|
|
46
|
-
matching_labels = h5["labels"][:]
|
|
47
|
-
|
|
48
|
-
template_data = TemplateData.from_npz(templates_npz)
|
|
49
|
-
tpca = data_util.get_tpca(sorting)
|
|
50
|
-
assert isinstance(tpca, BaseTemporalPCA)
|
|
51
|
-
if (sl := getattr(tpca, "temporal_slice", None)) is not None:
|
|
52
|
-
temps_tpca = torch.asarray(template_data.templates[:, sl])
|
|
53
|
-
else:
|
|
54
|
-
temps_tpca = torch.asarray(template_data.templates)
|
|
55
|
-
|
|
56
|
-
n, t, c = temps_tpca.shape
|
|
57
|
-
temps_tpca = temps_tpca.permute(0, 2, 1).reshape(n * c, t)
|
|
58
|
-
temps_tpca = tpca._transform_in_probe(temps_tpca)
|
|
59
|
-
temps_tpca = temps_tpca.reshape(n, c, -1).permute(0, 2, 1)
|
|
60
|
-
|
|
61
|
-
noise_log_priors = noise.detection_prior_log_prob(temps_tpca)
|
|
62
|
-
logger.dartsortdebug(
|
|
63
|
-
f"Got log priors ranging {noise_log_priors.min()}-{noise_log_priors.max()}."
|
|
64
|
-
)
|
|
65
|
-
noise_log_priors = noise_log_priors[matching_labels]
|
|
66
|
-
|
|
67
|
-
return noise_log_priors
|
|
68
|
-
elif stem.startswith("subtract"):
|
|
69
|
-
noise_log_priors = noise.channelwise_detection_prior_log_prob()
|
|
70
|
-
logger.dartsortdebug(
|
|
71
|
-
f"Got log priors ranging {noise_log_priors.min()}-{noise_log_priors.max()}."
|
|
72
|
-
)
|
|
73
|
-
noise_log_priors = noise_log_priors[sorting.channels]
|
|
74
|
-
return noise_log_priors
|
|
75
|
-
else:
|
|
76
|
-
return None
|
|
77
|
-
|
|
78
|
-
|
|
79
26
|
@databag
|
|
80
27
|
class PCMergeResult:
|
|
81
28
|
sorting: data_util.DARTsortSorting
|
|
@@ -186,7 +133,7 @@ def pc_merge(
|
|
|
186
133
|
)
|
|
187
134
|
sf = means.square_().sum(dim=1).sqrt_() * counts.sqrt()
|
|
188
135
|
sf = sf.numpy(force=True)
|
|
189
|
-
sorting,
|
|
136
|
+
sorting, _new_ids = reorder_by_depth(
|
|
190
137
|
sorting, motion=motion, spatial_footprints=sf, geom=motion.rgeom
|
|
191
138
|
)
|
|
192
139
|
return PCMergeResult(sorting=sorting)
|
|
@@ -505,15 +452,15 @@ def gmm_isolation_scores(
|
|
|
505
452
|
if dd is None:
|
|
506
453
|
continue
|
|
507
454
|
|
|
508
|
-
assert return_kdes
|
|
509
|
-
assert kde is not None
|
|
455
|
+
assert return_kdes
|
|
456
|
+
assert kde is not None
|
|
510
457
|
if domain is None:
|
|
511
458
|
domain = dd
|
|
512
|
-
assert kdes is None
|
|
459
|
+
assert kdes is None
|
|
513
460
|
kdes = np.full((unit_ids.size, domain.shape[0]), np.nan)
|
|
514
461
|
else:
|
|
515
|
-
assert np.array_equal(domain, dd)
|
|
516
|
-
assert kdes is not None
|
|
462
|
+
assert np.array_equal(domain, dd)
|
|
463
|
+
assert kdes is not None
|
|
517
464
|
kdes[j] = kde
|
|
518
465
|
|
|
519
466
|
return GMMIsolationScores(isolation=isolation, kde_domain=domain, kdes=kdes)
|
|
@@ -138,14 +138,14 @@ class DARTsortUserConfig:
|
|
|
138
138
|
"""Threshold in standardized (SNR) voltage units for initial detection;
|
|
139
139
|
peaks or troughs larger than this value will be grabbed."""
|
|
140
140
|
|
|
141
|
-
matching_threshold: Annotated[float, Field(gt=0)] =
|
|
141
|
+
matching_threshold: Annotated[float, Field(gt=0)] = 6.0
|
|
142
142
|
"""Template matching threshold. If subtracting a template leads
|
|
143
143
|
to at least this great of a decrease in the norm of the residual,
|
|
144
144
|
that match will be used. This is in the same units as the corresponding
|
|
145
145
|
threshold in Kilosort and other sorters, and it represents reduction in
|
|
146
146
|
Euclidean norm of standardized data due to matching a new event."""
|
|
147
147
|
|
|
148
|
-
initial_threshold: Annotated[float, Field(gt=0)] =
|
|
148
|
+
initial_threshold: Annotated[float, Field(gt=0)] = 7.0
|
|
149
149
|
"""Initial detection's neural net matching threshold. Same as
|
|
150
150
|
matching_threshold, except that a neural net is trying to guess
|
|
151
151
|
the true waveforms here, rather than using cluster templates."""
|
|
@@ -208,7 +208,7 @@ class DARTsortUserConfig:
|
|
|
208
208
|
agg_kind: Literal["none", "template_distance", "qda"] = "qda"
|
|
209
209
|
"""Final distance or GMM-based merge type."""
|
|
210
210
|
|
|
211
|
-
spikeinterface_merge_preset: str | Literal["none"]
|
|
211
|
+
spikeinterface_merge_preset: str | Literal["none"] = "none"
|
|
212
212
|
"""Call out to SpikeInterface's auto_merge() for a final merge using timing / RP information.
|
|
213
213
|
Setting this is slightly different' from calling auto_merge() externally, since the internal
|
|
214
214
|
version will make use of dartsort's templates and template distances.
|
|
@@ -84,16 +84,16 @@ def detect_and_deduplicate(
|
|
|
84
84
|
Tp = slen * nrep
|
|
85
85
|
key = (traces.device.type, traces.device.index, Tp, dedup_temporal_radius)
|
|
86
86
|
if key in _pepper:
|
|
87
|
-
|
|
87
|
+
dedup_time_pepper = _pepper[key]
|
|
88
88
|
else:
|
|
89
89
|
# favor earlier times when deduplicating identical values (at float16 prec)
|
|
90
|
-
|
|
90
|
+
dedup_time_pepper = torch.linspace(
|
|
91
91
|
dedup_salt_eps, 0.0, steps=slen, device=traces.device
|
|
92
92
|
)
|
|
93
|
-
|
|
94
|
-
_pepper[key] =
|
|
93
|
+
dedup_time_pepper = dedup_time_pepper.repeat(nrep)
|
|
94
|
+
_pepper[key] = dedup_time_pepper
|
|
95
95
|
else:
|
|
96
|
-
|
|
96
|
+
dedup_time_pepper = None
|
|
97
97
|
|
|
98
98
|
for i0 in range(0, T, batch_size):
|
|
99
99
|
i1 = min(T, i0 + batch_size)
|
|
@@ -131,6 +131,9 @@ def detect_and_deduplicate(
|
|
|
131
131
|
tmp = peak
|
|
132
132
|
del peak
|
|
133
133
|
|
|
134
|
+
if detection_mask is not None:
|
|
135
|
+
detect.logical_and_(detection_mask[i00:i11].T)
|
|
136
|
+
|
|
134
137
|
# check if deduping
|
|
135
138
|
if not will_dedup:
|
|
136
139
|
all_peaks[i0:i1] = detect[:, istart:iend].T
|
|
@@ -139,11 +142,9 @@ def detect_and_deduplicate(
|
|
|
139
142
|
# -- deduplicate peaks
|
|
140
143
|
if dedup_chans_salt is not None:
|
|
141
144
|
Xdd[:-1] += dedup_chans_salt
|
|
142
|
-
if
|
|
143
|
-
Xdd[:-1] +=
|
|
145
|
+
if dedup_time_pepper is not None:
|
|
146
|
+
Xdd[:-1] += dedup_time_pepper[i00:i11]
|
|
144
147
|
mask_out = torch.logical_not(detect, out=tmp)
|
|
145
|
-
if detection_mask is not None:
|
|
146
|
-
mask_out.logical_and_(detection_mask[i00:i11].T)
|
|
147
148
|
Xdd[:-1].masked_fill_(mask_out, 0.0)
|
|
148
149
|
|
|
149
150
|
# no-threshold max pool for deduplication
|
|
@@ -136,7 +136,7 @@ class DARTsortGroundTruthComparison:
|
|
|
136
136
|
return res
|
|
137
137
|
|
|
138
138
|
def soft_assignment_key(
|
|
139
|
-
self, to_try=
|
|
139
|
+
self, to_try=("merged_responsibilities", "gmm_responsibilities")
|
|
140
140
|
) -> str | None:
|
|
141
141
|
if hasattr(self, "_soft_assignment_key"):
|
|
142
142
|
return self._soft_assignment_key
|
|
@@ -147,87 +147,88 @@ class DARTsortGroundTruthComparison:
|
|
|
147
147
|
self._soft_assignment_key = None
|
|
148
148
|
return None
|
|
149
149
|
|
|
150
|
-
def brier_score(self, gt_unit: int) -> tuple[float, float | None]:
|
|
151
|
-
|
|
152
|
-
|
|
153
|
-
|
|
154
|
-
|
|
155
|
-
|
|
156
|
-
|
|
157
|
-
|
|
158
|
-
|
|
159
|
-
|
|
160
|
-
|
|
161
|
-
|
|
162
|
-
|
|
163
|
-
|
|
164
|
-
|
|
165
|
-
|
|
166
|
-
|
|
167
|
-
|
|
168
|
-
|
|
169
|
-
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
|
|
173
|
-
|
|
174
|
-
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
|
|
178
|
-
|
|
179
|
-
|
|
180
|
-
|
|
181
|
-
|
|
182
|
-
|
|
183
|
-
|
|
184
|
-
|
|
185
|
-
|
|
186
|
-
|
|
187
|
-
|
|
188
|
-
|
|
189
|
-
|
|
190
|
-
|
|
191
|
-
|
|
192
|
-
|
|
193
|
-
|
|
194
|
-
|
|
195
|
-
|
|
196
|
-
|
|
197
|
-
|
|
198
|
-
|
|
199
|
-
|
|
200
|
-
|
|
201
|
-
|
|
202
|
-
|
|
203
|
-
|
|
204
|
-
|
|
205
|
-
|
|
206
|
-
|
|
207
|
-
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
213
|
-
|
|
214
|
-
|
|
215
|
-
|
|
216
|
-
|
|
217
|
-
|
|
218
|
-
|
|
219
|
-
|
|
220
|
-
|
|
221
|
-
|
|
222
|
-
|
|
223
|
-
|
|
224
|
-
|
|
225
|
-
|
|
226
|
-
|
|
227
|
-
|
|
228
|
-
|
|
229
|
-
|
|
230
|
-
|
|
150
|
+
# def brier_score(self, gt_unit: int) -> tuple[float, float | None]:
|
|
151
|
+
# """Compute something like a Brier score in our context
|
|
152
|
+
|
|
153
|
+
# There are edge cases in spike sorting that normal Brier score doesn't deal with.
|
|
154
|
+
# First of all, the GT and tested event counts are different, as are the GT and tested
|
|
155
|
+
# class counts. This means we have FPs and FNs for both events and classes. So we have
|
|
156
|
+
# to handle those cases.
|
|
157
|
+
|
|
158
|
+
# For handling the class mapping from GT to tested:
|
|
159
|
+
# - Use the best matching sorted unit's soft score as the probability of the
|
|
160
|
+
# true class.
|
|
161
|
+
# - If a GT unit has no matched sorted unit, its Brier score is 2, the worst
|
|
162
|
+
# possible score.
|
|
163
|
+
|
|
164
|
+
# That's how we deal with classe. Now for the spikes.
|
|
165
|
+
|
|
166
|
+
# Else, we go spike by spike through the GT spikes and get a score for the
|
|
167
|
+
# ith spike as brier[i], and output mean(brier).
|
|
168
|
+
# The cases are: TP, FP, FN.
|
|
169
|
+
|
|
170
|
+
# - For a true positive, it's standard. Spike i's resp[i, 0] is its true class
|
|
171
|
+
# prob, so...
|
|
172
|
+
# brier[i] = (1 - resp[i, 0])**2 + (resp[i, 1:]**2).sum()
|
|
173
|
+
|
|
174
|
+
# - Now, for a false positive. This spike might come from another class, but we
|
|
175
|
+
# don't know which one here. It's debatable what to do -- you could try to figure out
|
|
176
|
+
# what the best matching unit is for the true class... but we don't even know if the
|
|
177
|
+
# spike exists in the GT sorting, much less which one it is. So, we just set
|
|
178
|
+
# brier[i] = 2
|
|
179
|
+
# The worst possible value.
|
|
180
|
+
|
|
181
|
+
# - For a false negative, it's the same situation. Maybe the spike got picked up by another
|
|
182
|
+
# unit and has some mass on the best matching unit's component, but we don't
|
|
183
|
+
# know that, so we set
|
|
184
|
+
# brier[i] = 2
|
|
185
|
+
# """
|
|
186
|
+
# in_gt_unit, matched_gt_indices, only_gt_indices = self.matched_and_missed(
|
|
187
|
+
# gt_unit
|
|
188
|
+
# )
|
|
189
|
+
|
|
190
|
+
# # hard brier score
|
|
191
|
+
|
|
192
|
+
# tpmask = gt_is_tp[in_gt]
|
|
193
|
+
# fnmask = np.logical_not(tpmask)
|
|
194
|
+
# hard_brier = 2.0 * fnmask.mean()
|
|
195
|
+
# brier_records.append(
|
|
196
|
+
# dict(
|
|
197
|
+
# gt_unit_id=gt_unit_id,
|
|
198
|
+
# matched=matched,
|
|
199
|
+
# brier=hard_brier,
|
|
200
|
+
# kind="hard",
|
|
201
|
+
# true_class_mean_prob=tpmask.mean(),
|
|
202
|
+
# )
|
|
203
|
+
# )
|
|
204
|
+
|
|
205
|
+
# # soft...
|
|
206
|
+
# tested_ix = tested_index_of_gt[in_gt]
|
|
207
|
+
# uresp = resp[tested_ix]
|
|
208
|
+
# ucand = cand[tested_ix]
|
|
209
|
+
# # handle subtraction of true class one hot
|
|
210
|
+
# isbest = ucand == tested_best_match
|
|
211
|
+
|
|
212
|
+
# true_class_probs = uresp[isbest]
|
|
213
|
+
# uresp[isbest] = 0.0
|
|
214
|
+
# sbrier = np.square(uresp).sum(axis=1)
|
|
215
|
+
# sbrier += np.square(noiseresp[tested_ix])
|
|
216
|
+
# sbrier += np.logical_not(isbest.any(1))
|
|
217
|
+
# pass
|
|
218
|
+
|
|
219
|
+
# def brier_scores(self) -> tuple[np.ndarray, np.ndarray | None]:
|
|
220
|
+
# has_soft = self.soft_assignment_key() is not None
|
|
221
|
+
# hard_scores = np.zeros(self.n_gt_units)
|
|
222
|
+
# if has_soft:
|
|
223
|
+
# soft_scores = np.zeros_like(hard_scores)
|
|
224
|
+
# else:
|
|
225
|
+
# soft_scores = None
|
|
226
|
+
# for j, gt_unit in enumerate(self.unit_ids):
|
|
227
|
+
# hard_scores[j], ss = self.brier_score(gt_unit)
|
|
228
|
+
# if ss is not None:
|
|
229
|
+
# assert soft_scores is not None
|
|
230
|
+
# soft_scores[j] = ss
|
|
231
|
+
# return hard_scores, soft_scores
|
|
231
232
|
|
|
232
233
|
def _unit_info_dataframe(self, force_distances=False, perf_only=False):
|
|
233
234
|
firing_rates = self.gt_analysis.firing_rates()
|