diffai-core 0.5.1

Core library for AI/ML diff analysis - PyTorch, Safetensors, tensor statistics
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

use serde_json::Value;

use crate::types::DiffResult;

pub fn analyze_batch_normalization_analysis(
    old_model: &Value,
    new_model: &Value,
    results: &mut Vec<DiffResult>,
) {
    if let (Value::Object(old_obj), Value::Object(new_obj)) = (old_model, new_model) {
        // Batch normalization layer detection and analysis
        if let Some((old_bn, new_bn)) = analyze_batch_norm_layers(old_obj, new_obj) {
            results.push(DiffResult::ModelArchitectureChanged(
                "batch_normalization_layers".to_string(),
                old_bn,
                new_bn,
            ));
        }

        // Batch normalization parameter analysis
        if let Some((old_params, new_params)) = analyze_batch_norm_parameters(old_obj, new_obj) {
            results.push(DiffResult::ModelArchitectureChanged(
                "batch_normalization_parameters".to_string(),
                old_params,
                new_params,
            ));
        }

        // Moving statistics analysis for batch norm
        if let Some((old_stats, new_stats)) = analyze_batch_norm_statistics(old_obj, new_obj) {
            results.push(DiffResult::ModelArchitectureChanged(
                "batch_normalization_statistics".to_string(),
                old_stats,
                new_stats,
            ));
        }
    }
}

// Regularization Impact Analysis - measure regularization technique effectiveness
// Helper functions for batch normalization analysis
fn analyze_batch_norm_layers(
    old_obj: &serde_json::Map<String, Value>,
    new_obj: &serde_json::Map<String, Value>,
) -> Option<(String, String)> {
    let old_bn_count = count_batch_norm_layers(old_obj);
    let new_bn_count = count_batch_norm_layers(new_obj);

    if old_bn_count != new_bn_count {
        Some((
            format!("batch_norm_layers: {old_bn_count}"),
            format!("batch_norm_layers: {new_bn_count}"),
        ))
    } else {
        None
    }
}

fn count_batch_norm_layers(obj: &serde_json::Map<String, Value>) -> usize {
    let mut count = 0;
    for (key, _) in obj {
        if key.contains("batch_norm") || key.contains("bn") || key.contains("BatchNorm") {
            count += 1;
        }
    }
    count
}

fn analyze_batch_norm_parameters(
    old_obj: &serde_json::Map<String, Value>,
    new_obj: &serde_json::Map<String, Value>,
) -> Option<(String, String)> {
    let old_params = extract_batch_norm_params(old_obj);
    let new_params = extract_batch_norm_params(new_obj);

    if old_params != new_params {
        Some((
            format!(
                "bn_params: momentum={:.3}, eps={:.6}",
                old_params.0, old_params.1
            ),
            format!(
                "bn_params: momentum={:.3}, eps={:.6}",
                new_params.0, new_params.1
            ),
        ))
    } else {
        None
    }
}

fn extract_batch_norm_params(obj: &serde_json::Map<String, Value>) -> (f64, f64) {
    let momentum = obj.get("momentum").and_then(|v| v.as_f64()).unwrap_or(0.1);
    let eps = obj.get("eps").and_then(|v| v.as_f64()).unwrap_or(1e-5);
    (momentum, eps)
}

fn analyze_batch_norm_statistics(
    old_obj: &serde_json::Map<String, Value>,
    new_obj: &serde_json::Map<String, Value>,
) -> Option<(String, String)> {
    let old_stats = extract_batch_norm_stats(old_obj);
    let new_stats = extract_batch_norm_stats(new_obj);

    if (old_stats.0 - new_stats.0).abs() > 0.01 || (old_stats.1 - new_stats.1).abs() > 0.01 {
        Some((
            format!(
                "bn_stats: running_mean={:.3}, running_var={:.3}",
                old_stats.0, old_stats.1
            ),
            format!(
                "bn_stats: running_mean={:.3}, running_var={:.3}",
                new_stats.0, new_stats.1
            ),
        ))
    } else {
        None
    }
}

fn extract_batch_norm_stats(obj: &serde_json::Map<String, Value>) -> (f64, f64) {
    let running_mean = obj
        .get("running_mean")
        .and_then(|v| v.as_f64())
        .unwrap_or(0.0);
    let running_var = obj
        .get("running_var")
        .and_then(|v| v.as_f64())
        .unwrap_or(1.0);
    (running_mean, running_var)
}