BaseSorting.select_periods() drops properties, annotations and the registered recording

未关闭
  • 哪里不对？
#4,915 3 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@AxelNoun 已经在做这个了。

开始于 2026年11月5日。

  • #4921 来自 @AxelNoun —— 未关闭

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
20/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
停滞
技术栈
python
领域
backend

调研方向

The issue points to src/spikeinterface/core/sorting_tools.py around select_sorting_periods and describes a reproduction using BaseSorting.select_periods(). Read that function and the related tests in test_basesorting.py; verify that the selected sorting retains properties, annotations, its registered recording, and correct returned spike times. An open pull request, #4921, is already linked to this issue.

由索引模型根据 Issue 内容生成。

描述

bug core

select_sorting_periods overwrites sorting with the new object, then copies the metadata from this object onto itself:

sorting = NumpySorting(
    sliced_spike_vector, sampling_frequency=sorting.sampling_frequency, unit_ids=sorting.unit_ids
)
sorting.copy_metadata(sorting)
return sorting

So after select_periods, all properties and annotations are lost. The recording is also no longer registered, so get_unit_spike_train(..., return_times=True) ignores the start time or time vector of the recording.

Reproduction

main at 9ef464e66, Linux (WSL Ubuntu 24.04), Python 3.12.3, numpy 2.5.4:

import numpy as np
import spikeinterface.core as sc
from spikeinterface.core.base import unit_period_dtype

fs = 1000.0
rec = sc.generate_recording(num_channels=2, sampling_frequency=fs, durations=[10.0], seed=0)
rec.shift_times(500.0)  # the recording starts at t = 500 s
sorting = sc.NumpySorting.from_unit_dict(
    {"a": np.array([10, 2000, 5000]), "b": np.array([30, 3000])}, sampling_frequency=fs
)
sorting.set_property("quality", np.array(["good", "mua"]))
sorting.annotate(probe_name="NP2.0")
sorting.register_recording(rec)

periods = np.array([(0, 0, 4000, 0), (0, 0, 4000, 1)], dtype=unit_period_dtype)
selected = sorting.select_periods(periods)

print("properties:   ", sorting.get_property_keys(), "->", selected.get_property_keys())
print("annotations:  ", sorting.get_annotation_keys(), "->", selected.get_annotation_keys())
print("has_recording:", sorting.has_recording(), "->", selected.has_recording())
print("spike times of 'a' (s):", sorting.get_unit_spike_train("a", return_times=True)[:2],
      "->", selected.get_unit_spike_train("a", return_times=True))

properties:    ['quality'] -> []
annotations:   ['probe_name'] -> []
has_recording: True -> False
spike times of 'a' (s): [500.01 502.  ] -> [0.01 2.  ]

The spike times are off by 500 s.

Possible fix
-        sorting = NumpySorting(
+        new_sorting = NumpySorting(
             sliced_spike_vector, sampling_frequency=sorting.sampling_frequency, unit_ids=sorting.unit_ids
         )
-        sorting.copy_metadata(sorting)
-        return sorting
+        sorting.copy_metadata(new_sorting)
+        if sorting.has_recording():
+            new_sorting.register_recording(sorting._recording)
+        return new_sorting

With this, the properties, annotations and spike times are preserved in the example above. test_basesorting.py, test_baserecording.py, test_binaryfolder.py, test_binaryrecordingextractor.py and test_numpy_extractors.py still pass (33 tests).

I found no existing report, and no open PR changes this line. It comes from #4316. #4796 fixed a similar loss of start times in NumpySorting.from_sorting. I can open a PR with this change and a test.

主要语言
Python
星标
870
派生
284
平均合并
3 天 1 小时
30 天内合并 PR
41

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

SpikeInterface/spikeinterface 的其他 Issue

查看 SpikeInterface/spikeinterface 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。