segovia 0.4.1

A fast, chunked, memory-bounded Rust engine for electrophysiology (Neuropixels) signal processing, with Python bindings.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131

import numpy as np
import pytest
import segovia
from scipy import signal

FS = 30000.0
BAND = (300.0, 6000.0)
ORDER = 5


def _sos():
    return np.ascontiguousarray(
        signal.butter(ORDER, BAND, btype="band", fs=FS, output="sos"),
        dtype=np.float64,
    )


def write_recording(tmp_path, n_channels, n_samples, seed=0):
    rng = np.random.default_rng(seed)
    t = np.arange(n_samples) / FS
    data = np.zeros((n_samples, n_channels), dtype=np.float64)
    for c in range(n_channels):
        in_band = np.sin(2 * np.pi * (500 + 80 * c) * t)
        low = 0.7 * np.sin(2 * np.pi * 30 * t)
        high = 0.4 * np.sin(2 * np.pi * 11000 * t)
        noise = 0.3 * rng.standard_normal(n_samples)
        offset = 50 * (c - n_channels / 2)
        data[:, c] = 400 * (in_band + low + high) + 200 * noise + offset
    data = np.round(data).astype(np.int16)

    bin_path = tmp_path / "rec_g0_t0.imec0.ap.bin"
    meta_path = tmp_path / "rec_g0_t0.imec0.ap.meta"
    data.tofile(bin_path)
    meta_path.write_text(
        f"nSavedChans={n_channels}\n"
        f"imSampRate={int(FS)}\n"
        f"typeThis=imec\n"
        f"fileSizeBytes={bin_path.stat().st_size}\n"
    )
    return bin_path, meta_path, data


def reference_bandpass(raw, sos):
    return signal.sosfiltfilt(sos, raw, axis=0)


def reference_cmr(filtered):
    return filtered - np.median(filtered, axis=1, keepdims=True)


def reference_whitener(calib_cmr, eps, apply_mean):
    mean = calib_cmr.mean(axis=0) if apply_mean else np.zeros(calib_cmr.shape[1])
    centered = calib_cmr - mean
    cov = centered.T @ centered / centered.shape[0]
    evals, evecs = np.linalg.eigh(cov)
    w = evecs @ np.diag(1.0 / np.sqrt(evals + eps)) @ evecs.T
    return mean, w


def reference_chain(raw, sos, chunk_samples, calib_samples, eps, apply_mean):
    raw = raw.astype(np.float64)
    full = reference_cmr(reference_bandpass(raw, sos))
    calib = raw[:calib_samples]
    calib_cmr = reference_cmr(reference_bandpass(calib, sos))
    mean, w = reference_whitener(calib_cmr, eps, apply_mean)
    return (full - mean) @ w


def test_bandpass_only_matches_scipy(tmp_path):
    bin_path, meta_path, data = write_recording(tmp_path, 8, 30000)
    sos = _sos()
    reader = segovia.SpikeGlxReader(str(bin_path), str(meta_path))

    pre = reader.preprocess(
        sos,
        chunk_samples=6000,
        margin=4000,
        calib_samples=1,
        batch=4,
        whiten=False,
    )
    seg = np.concatenate(list(pre), axis=0)

    ref_bp = reference_bandpass(data.astype(np.float64), sos)
    ref_cmr = reference_cmr(ref_bp)

    scale = np.max(np.abs(ref_cmr))
    rel = np.max(np.abs(seg.astype(np.float64) - ref_cmr)) / scale
    assert rel < 1e-4, f"bandpass+CMR relative error {rel:.2e}"


def test_full_chain_matches_scipy(tmp_path):
    bin_path, meta_path, data = write_recording(tmp_path, 16, 30000)
    sos = _sos()
    reader = segovia.SpikeGlxReader(str(bin_path), str(meta_path))

    chunk_samples = 6000
    calib_samples = 12000
    eps = 1e-6
    margin = 4000

    pre = reader.preprocess(
        sos,
        chunk_samples=chunk_samples,
        margin=margin,
        calib_samples=calib_samples,
        eps=eps,
        apply_mean=True,
        batch=4,
    )
    seg = np.concatenate(list(pre), axis=0).astype(np.float64)

    ref = reference_chain(data, sos, chunk_samples, calib_samples, eps, True)

    assert seg.shape == ref.shape
    scale = np.max(np.abs(ref))
    rel = np.max(np.abs(seg - ref)) / scale
    assert rel < 1e-3, f"full-chain relative error {rel:.2e}"


def test_output_is_float32_and_bounded_chunks(tmp_path):
    bin_path, meta_path, data = write_recording(tmp_path, 8, 20000)
    sos = _sos()
    reader = segovia.SpikeGlxReader(str(bin_path), str(meta_path))
    pre = reader.preprocess(
        sos, chunk_samples=5000, margin=2000, calib_samples=10000
    )
    chunks = list(pre)
    assert all(c.dtype == np.float32 for c in chunks)
    assert all(c.shape[1] == 8 for c in chunks)
    assert sum(c.shape[0] for c in chunks) == data.shape[0]