exg 0.0.5

EXG (EEG/ECG/EMG) preprocessing — native Rust DSP + FIF reader, numerical parity with MNE-Python
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
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255

/// Integration tests for the FIF reader.
///
/// All numerical assertions compare against reference vectors generated by
/// `scripts/gen_fiff_vectors.py` (MNE ground truth).  Each test documents
/// the exact MNE call used to produce the reference.
mod common;
use common::load_vectors_f64;
use approx::assert_abs_diff_eq;
use exg::fiff::raw::open_raw;
use std::path::Path;

const FIF_PATH: &str = concat!(env!("CARGO_MANIFEST_DIR"), "/data/sample1_raw.fif");

fn fif_available() -> bool {
    Path::new(FIF_PATH).exists()
}

// ── Header / info tests ──────────────────────────────────────────────────

#[test]
fn info_nchan() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let expected = vecs.get("nchan").unwrap()[[0]] as i32;
    let raw = open_raw(FIF_PATH).unwrap();
    assert_eq!(raw.info.n_chan as i32, expected);
}

#[test]
fn info_sfreq() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let expected = vecs.get("sfreq").unwrap()[[0]];
    let raw = open_raw(FIF_PATH).unwrap();
    assert_abs_diff_eq!(raw.info.sfreq, expected, epsilon = 1e-6);
}

#[test]
fn info_first_last_samp() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let expected_first = vecs.get("first_samp").unwrap()[[0]] as i64;
    let expected_ntimes = vecs.get("ntimes").unwrap()[[0]] as i64;
    let raw = open_raw(FIF_PATH).unwrap();
    assert_eq!(raw.first_samp as i64, expected_first);
    assert_eq!(raw.n_times() as i64, expected_ntimes);
    // last_samp = first_samp + ntimes - 1
    assert_eq!(raw.last_samp as i64, expected_first + expected_ntimes - 1);
}

#[test]
fn info_channel_names() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    // ch_names is [C, 16] stored as U8 → decoded as f64 (byte values 0..255).
    let names_raw = vecs.get("ch_names").unwrap();
    let raw = open_raw(FIF_PATH).unwrap();
    let n_ch = raw.info.n_chan;

    let ref_names: Vec<String> = (0..n_ch)
        .map(|c| {
            let bytes: Vec<u8> = (0..16_usize)
                .map(|b| names_raw[[c, b]] as u8)
                .take_while(|&b| b != 0)
                .collect();
            String::from_utf8_lossy(&bytes).into_owned()
        })
        .collect();

    for (i, (got, exp)) in raw.info.ch_names().iter().zip(ref_names.iter()).enumerate() {
        assert_eq!(*got, exp.as_str(), "ch[{i}] name mismatch");
    }
}

#[test]
fn info_calibrations() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let ref_cal   = vecs.get("ch_cal").unwrap();
    let ref_range = vecs.get("ch_range").unwrap();
    let raw = open_raw(FIF_PATH).unwrap();
    for (i, ch) in raw.info.chs.iter().enumerate() {
        assert_abs_diff_eq!(ch.cal   as f64, ref_cal  [[i]], epsilon = 1e-6_f64);
        assert_abs_diff_eq!(ch.range as f64, ref_range[[i]], epsilon = 1e-6_f64);
    }
}

#[test]
fn info_channel_kinds() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let ref_kind = vecs.get("ch_kind").unwrap();
    let raw = open_raw(FIF_PATH).unwrap();
    for (i, ch) in raw.info.chs.iter().enumerate() {
        assert_eq!(ch.kind, ref_kind[[i]] as i32, "ch[{}] kind", i);
    }
}

#[test]
fn info_channel_locations() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let ref_locs = vecs.get("ch_locs").unwrap();  // [C, 12] f32
    let raw = open_raw(FIF_PATH).unwrap();
    for (c, ch) in raw.info.chs.iter().enumerate() {
        for i in 0..12 {
            let got = ch.loc[i] as f64;
            let exp = ref_locs[[c, i]];
            if exp.is_nan() {
                assert!(got.is_nan(), "ch[{}] loc[{}]: expected NaN, got {}", c, i, got);
            } else {
                assert_abs_diff_eq!(got, exp, epsilon = 1e-6_f64);
            }
        }
    }
}

// ── Data reading tests ────────────────────────────────────────────────────

#[test]
fn data_shape_matches_mne() {
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_data");
    let ref_data = vecs.get("data").unwrap();   // [C, T] f64 from MNE
    let raw = open_raw(FIF_PATH).unwrap();
    let data = raw.read_all_data().unwrap();
    assert_eq!(data.nrows(), ref_data.shape()[0], "n_channels");
    assert_eq!(data.ncols(), ref_data.shape()[1], "n_times");
}

#[test]
fn data_all_samples_match_mne() {
    // MNE reference: `raw.get_data()` with `preload=True`.
    // Tolerance: f32 round-trip → max error ≤ 1 ULP of f32.
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_data");
    let ref_data = vecs.get("data").unwrap();
    let raw = open_raw(FIF_PATH).unwrap();
    let data = raw.read_all_data().unwrap();

    let n_ch = data.nrows();
    let n_t  = data.ncols();
    let mut max_err = 0_f64;
    let mut worst_c = 0;
    let mut worst_t = 0;

    for c in 0..n_ch {
        for t in 0..n_t {
            let err = (data[[c, t]] - ref_data[[c, t]]).abs();
            if err > max_err {
                max_err = err;
                worst_c = c;
                worst_t = t;
            }
        }
    }

    // Raw values are stored as f32 in the FIF file → round-trip error ≤ ~1e-10.
    assert!(max_err < 1e-9,
        "max error {max_err:.2e} at ch={worst_c} t={worst_t}  \
         (got={:.10e} ref={:.10e})",
        data[[worst_c, worst_t]], ref_data[[worst_c, worst_t]]);
}

#[test]
fn data_first_second_sample_exact() {
    // Bit-exact: raw buffer bytes decoded in Python and Rust must be identical.
    if !fif_available() { return; }
    let vecs_short = load_vectors_f64("fiff_sample_data_short");
    let ref_256 = vecs_short.get("data").unwrap();  // [12, 256] f64
    let raw = open_raw(FIF_PATH).unwrap();
    let data = raw.read_slice(0, 256).unwrap();

    for c in 0..12 {
        for t in 0..256 {
            let err = (data[[c, t]] - ref_256[[c, t]]).abs();
            assert!(err < 1e-9,
                "ch={} t={}: got={:.10e} ref={:.10e} err={:.2e}",
                c, t, data[[c, t]], ref_256[[c, t]], err);
        }
    }
}

#[test]
fn read_slice_equals_subarray_of_read_all() {
    if !fif_available() { return; }
    let raw = open_raw(FIF_PATH).unwrap();
    let all = raw.read_all_data().unwrap();
    let slice = raw.read_slice(100, 600).unwrap();   // 500 samples
    for c in 0..12 {
        for t in 0..500 {
            assert_abs_diff_eq!(slice[[c, t]], all[[c, t + 100]], epsilon = 1e-14);
        }
    }
}

#[test]
fn buffer_boundaries_match_mne() {
    // MNE: `raw._raw_extras[0]['bounds']`
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_info");
    let ref_bounds = vecs.get("buf_bounds").unwrap();  // int64
    let raw = open_raw(FIF_PATH).unwrap();

    // Build the cumulative boundary array from our buffer records.
    let mut rust_bounds: Vec<u64> = Vec::new();
    rust_bounds.push(raw.first_samp);
    for buf in &raw.buffers {
        rust_bounds.push(rust_bounds.last().copied().unwrap() + buf.n_samp as u64);
    }

    assert_eq!(rust_bounds.len(), ref_bounds.shape()[0],
        "number of boundaries: Rust={} MNE={}", rust_bounds.len(), ref_bounds.shape()[0]);
    for (i, (&rust_b, ref_b)) in rust_bounds.iter()
        .zip(ref_bounds.iter().map(|&v| v as u64))
        .enumerate()
    {
        assert_eq!(rust_b, ref_b, "bound[{}]: rust={} mne={}", i, rust_b, ref_b);
    }
}

#[test]
fn data_statistics_match_mne() {
    // High-level sanity: mean / std / min / max should match to float32 precision.
    if !fif_available() { return; }
    let vecs = load_vectors_f64("fiff_sample_data");
    let ref_data = vecs.get("data").unwrap();
    let raw = open_raw(FIF_PATH).unwrap();
    let data = raw.read_all_data().unwrap();

    let n = data.len() as f64;
    let rust_mean = data.iter().sum::<f64>() / n;
    let ref_mean  = ref_data.iter().sum::<f64>() / n;
    assert_abs_diff_eq!(rust_mean, ref_mean, epsilon = 1e-12);

    let rust_max = data.iter().cloned().fold(f64::NEG_INFINITY, f64::max);
    let ref_max  = ref_data.iter().cloned().fold(f64::NEG_INFINITY, f64::max);
    assert_abs_diff_eq!(rust_max, ref_max, epsilon = 1e-10);
}

#[test]
fn read_slice_boundary_wraps_buffers() {
    // A slice that spans exactly two buffer boundaries should still be correct.
    if !fif_available() { return; }
    let raw = open_raw(FIF_PATH).unwrap();
    let all  = raw.read_all_data().unwrap();
    // Sample1_raw: buffers are 256 samples each.
    // Slice from 200..320 crosses the 256-sample boundary.
    let slice = raw.read_slice(200, 320).unwrap();
    for c in 0..12 {
        for t in 0..120 {
            assert_abs_diff_eq!(slice[[c, t]], all[[c, t + 200]], epsilon = 1e-14);
        }
    }
}