osf-rs 0.0.1

OSF Sleep Foundation Model — inference in Rust with Burn ML
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

/// Model and data configuration for OSF inference.
///
/// `ModelConfig` mirrors the Python OSF ViT hyperparameters stored in
/// the `metadata` dict of `osf_backbone.pth`.

// ── ModelConfig ───────────────────────────────────────────────────────────────

#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
pub struct ModelConfig {
    /// Model variant name (e.g. "vit_base").
    #[serde(default = "default_encoder_name")]
    pub encoder_name: String,

    /// Number of input channels (default 12).
    #[serde(default = "default_num_leads")]
    pub num_leads: usize,

    /// Temporal patch size in samples (default 64).
    #[serde(default = "default_patch_size_time")]
    pub patch_size_time: usize,

    /// Channel patch size (default 4).
    #[serde(default = "default_patch_size_ch")]
    pub patch_size_ch: usize,

    /// 0 = 1D patchify (all channels), 1 = 2D patchify (channel groups).
    #[serde(default = "default_lead_wise")]
    pub lead_wise: usize,

    /// Sampling rate (Hz, default 64).
    #[serde(default = "default_sample_rate")]
    pub sample_rate: usize,

    /// Window size in seconds (default 30).
    #[serde(default = "default_window_size_sec")]
    pub window_size_sec: usize,

    /// Sequence length = sample_rate * window_size_sec (default 1920).
    #[serde(default = "default_seq_len")]
    pub seq_len: usize,

    /// Transformer hidden dimension (default 768).
    #[serde(default = "default_width")]
    pub width: usize,

    /// Number of transformer blocks (default 12).
    #[serde(default = "default_depth")]
    pub depth: usize,

    /// Number of attention heads (default 12).
    #[serde(default = "default_heads")]
    pub heads: usize,

    /// MLP hidden dimension (default 3072 = 4 * width).
    #[serde(default = "default_mlp_dim")]
    pub mlp_dim: usize,

    /// Attention head dimension (default 64 = width / heads).
    #[serde(default = "default_dim_head")]
    pub dim_head: usize,
}

fn default_encoder_name() -> String { "vit_base".to_string() }
fn default_num_leads()     -> usize { 12 }
fn default_patch_size_time() -> usize { 64 }
fn default_patch_size_ch() -> usize { 4 }
fn default_lead_wise()     -> usize { 1 }
fn default_sample_rate()   -> usize { 64 }
fn default_window_size_sec() -> usize { 30 }
fn default_seq_len()       -> usize { 1920 }
fn default_width()         -> usize { 768 }
fn default_depth()         -> usize { 12 }
fn default_heads()         -> usize { 12 }
fn default_mlp_dim()       -> usize { 3072 }
fn default_dim_head()      -> usize { 64 }

impl Default for ModelConfig {
    fn default() -> Self {
        Self {
            encoder_name:    default_encoder_name(),
            num_leads:       default_num_leads(),
            patch_size_time: default_patch_size_time(),
            patch_size_ch:   default_patch_size_ch(),
            lead_wise:       default_lead_wise(),
            sample_rate:     default_sample_rate(),
            window_size_sec: default_window_size_sec(),
            seq_len:         default_seq_len(),
            width:           default_width(),
            depth:           default_depth(),
            heads:           default_heads(),
            mlp_dim:         default_mlp_dim(),
            dim_head:        default_dim_head(),
        }
    }
}

impl ModelConfig {
    /// Number of time patches per channel row.
    pub fn num_patches_time(&self) -> usize {
        self.seq_len / self.patch_size_time
    }

    /// Number of channel rows (lead groups).
    pub fn num_lead_rows(&self) -> usize {
        if self.lead_wise == 0 { 1 } else { self.num_leads / self.patch_size_ch }
    }

    /// Total number of patches (excluding CLS token).
    pub fn num_patches(&self) -> usize {
        self.num_lead_rows() * self.num_patches_time()
    }

    /// Build config for a specific variant.
    pub fn for_variant(name: &str) -> Self {
        match name {
            "vit_nano" => Self {
                encoder_name: "vit_nano".into(),
                width: 128, depth: 6, heads: 4, mlp_dim: 512, dim_head: 32,
                ..Default::default()
            },
            "vit_tiny" => Self {
                encoder_name: "vit_tiny".into(),
                width: 192, depth: 12, heads: 3, mlp_dim: 768, dim_head: 64,
                ..Default::default()
            },
            "vit_small" => Self {
                encoder_name: "vit_small".into(),
                width: 384, depth: 12, heads: 6, mlp_dim: 1536, dim_head: 64,
                ..Default::default()
            },
            "vit_middle" => Self {
                encoder_name: "vit_middle".into(),
                width: 512, depth: 12, heads: 8, mlp_dim: 2048, dim_head: 64,
                ..Default::default()
            },
            "vit_base" | _ => Self::default(),
        }
    }
}

// ── PSG Channel definitions ─────────────────────────────────────────────────

/// The 12 standard PSG channels used by OSF, in canonical order.
pub const PSG_CHANNELS: &[&str] = &[
    "ECG",
    "EMG_Chin",
    "EMG_LLeg",
    "EMG_RLeg",
    "ABD",
    "THX",
    "NP",
    "SN",
    "EOG_E1_A2",
    "EOG_E2_A1",
    "EEG_C3_A2",
    "EEG_C4_A1",
];

/// Number of PSG channels.
pub const NUM_PSG_CHANNELS: usize = 12;

/// Sampling frequency (Hz).
pub const SAMPLE_RATE: usize = 64;

/// Epoch duration (seconds).
pub const EPOCH_SEC: usize = 30;

/// Samples per epoch.
pub const EPOCH_SAMPLES: usize = SAMPLE_RATE * EPOCH_SEC; // 1920