Hastic standalone
https://hastic.io
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
151 lines
4.3 KiB
151 lines
4.3 KiB
#[derive(Debug, Clone)] |
|
pub struct LearningResults { |
|
// model: Vec<f64>, |
|
patterns: Vec<Vec<f64>>, |
|
anti_patterns: Vec<Vec<f64>>, |
|
} |
|
|
|
const CORR_THRESHOLD: f64 = 0.95; |
|
|
|
#[derive(Clone)] |
|
pub struct PatternDetector { |
|
learning_results: LearningResults, |
|
} |
|
|
|
fn nan_to_zero(n: f64) -> f64 { |
|
if n.is_nan() { |
|
return 0.; |
|
} |
|
return n; |
|
} |
|
|
|
// TODO: move this to loginc of analytic unit |
|
impl PatternDetector { |
|
pub fn new(learning_results: LearningResults) -> PatternDetector { |
|
PatternDetector { learning_results } |
|
} |
|
|
|
pub async fn learn( |
|
reads: &Vec<Vec<(u64, f64)>>, |
|
anti_reads: &Vec<Vec<(u64, f64)>>, |
|
) -> LearningResults { |
|
// let size_avg = reads.iter().map(|r| r.len()).sum::<usize>() / reads.len(); |
|
|
|
let mut patterns = Vec::<Vec<f64>>::new(); |
|
let mut anti_patterns = Vec::<Vec<f64>>::new(); |
|
|
|
// TODO: implement actual learning |
|
|
|
for r in reads { |
|
let xs: Vec<f64> = r.iter().map(|e| e.1).map(nan_to_zero).collect(); |
|
patterns.push(xs); |
|
} |
|
|
|
for r in anti_reads { |
|
let xs: Vec<f64> = r.iter().map(|e| e.1).map(nan_to_zero).collect(); |
|
anti_patterns.push(xs); |
|
} |
|
|
|
// let model = stat.iter().map(|(c, v)| v / *c as f64).collect(); |
|
|
|
LearningResults { |
|
patterns, |
|
anti_patterns, |
|
} |
|
} |
|
|
|
// TODO: get iterator instead of vector |
|
pub fn detect(&self, ts: &Vec<(u64, f64)>) -> Vec<(u64, u64)> { |
|
let mut results = Vec::new(); |
|
// let mut i = 0; |
|
|
|
|
|
let pt = &self.learning_results.patterns; |
|
let apt = &self.learning_results.anti_patterns; |
|
|
|
|
|
for i in 0..ts.len() { |
|
|
|
let mut pattern_match_score = 0f64; |
|
let mut pattern_match_len = 0usize; |
|
let mut anti_pattern_match_score = 0f64; |
|
|
|
for p in pt { |
|
if i + p.len() < ts.len() { |
|
let mut backet = Vec::<f64>::new(); |
|
for j in 0..p.len() { |
|
backet.push(nan_to_zero(ts[i + j].1)); |
|
} |
|
let score = PatternDetector::corr_aligned(p, &backet); |
|
if score > pattern_match_score { |
|
pattern_match_score = score; |
|
pattern_match_len = p.len(); |
|
} |
|
} |
|
} |
|
|
|
for p in apt { |
|
if i + p.len() < ts.len() { |
|
let mut backet = Vec::<f64>::new(); |
|
for j in 0..p.len() { |
|
backet.push(nan_to_zero(ts[i + j].1)); |
|
} |
|
let score = PatternDetector::corr_aligned(p, &backet); |
|
if score > anti_pattern_match_score { |
|
anti_pattern_match_score = score; |
|
} |
|
} |
|
} |
|
|
|
if pattern_match_score > anti_pattern_match_score && pattern_match_score >= CORR_THRESHOLD { |
|
results.push((ts[i].0, ts[i + pattern_match_len - 1].0)); |
|
} |
|
} |
|
|
|
return results; |
|
} |
|
|
|
fn corr_aligned(xs: &Vec<f64>, ys: &Vec<f64>) -> f64 { |
|
let n = xs.len() as f64; |
|
let mut s_xs: f64 = 0f64; |
|
let mut s_ys: f64 = 0f64; |
|
let mut s_xsys: f64 = 0f64; |
|
let mut s_xs_2: f64 = 0f64; |
|
let mut s_ys_2: f64 = 0f64; |
|
|
|
let min = xs.len().min(ys.len()); |
|
xs.iter() |
|
.take(min) |
|
.zip(ys.iter().take(min)) |
|
.for_each(|(xi, yi)| { |
|
s_xs += xi; |
|
s_ys += yi; |
|
s_xsys += xi * yi; |
|
s_xs_2 += xi * xi; |
|
s_ys_2 += yi * yi; |
|
}); |
|
|
|
let numerator: f64 = n * s_xsys - s_xs * s_ys; |
|
let denominator: f64 = ((n * s_xs_2 - s_xs * s_xs) * (n * s_ys_2 - s_ys * s_ys)).sqrt(); |
|
|
|
// IT"s a hack |
|
if denominator < 0.01 { |
|
return 0.; |
|
} |
|
|
|
// TODO: case when denominator = 0 |
|
|
|
let result: f64 = numerator / denominator; |
|
|
|
// assert!(result.abs() <= 1.01); |
|
|
|
if result.abs() > 1.1 { |
|
println!("{:?}", xs); |
|
println!("------------"); |
|
println!("{:?}", ys); |
|
println!("WARNING: corr result > 1: {}", result); |
|
} |
|
|
|
return result; |
|
} |
|
}
|
|
|