add statistics module
provides some basic statistics functions (sum, mean, etc.) and a function to return the parameters of the linear regression of two variables implemented using num_traits to be more flexible for the types Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
This commit is contained in:
parent
6cad8ce4ce
commit
ba97479848
|
@ -30,6 +30,7 @@ lazy_static = "1.4"
|
|||
libc = "0.2"
|
||||
log = "0.4"
|
||||
nix = "0.16"
|
||||
num-traits = "0.2"
|
||||
once_cell = "1.3.1"
|
||||
openssl = "0.10"
|
||||
pam = "0.7"
|
||||
|
|
|
@ -5,6 +5,7 @@ pub mod config;
|
|||
pub mod node;
|
||||
pub mod reader;
|
||||
mod subscription;
|
||||
pub mod status;
|
||||
pub mod types;
|
||||
pub mod version;
|
||||
pub mod pull;
|
||||
|
|
|
@ -30,6 +30,7 @@ pub mod lru_cache;
|
|||
pub mod runtime;
|
||||
pub mod ticket;
|
||||
pub mod timer;
|
||||
pub mod statistics;
|
||||
pub mod systemd;
|
||||
|
||||
mod wrapped_reader_stream;
|
||||
|
|
|
@ -0,0 +1,123 @@
|
|||
//! Helpers for common statistics tasks
|
||||
use num_traits::NumAssignRef;
|
||||
use num_traits::cast::ToPrimitive;
|
||||
|
||||
/// Calculates the sum of a list of numbers
|
||||
/// ```
|
||||
/// # use proxmox_backup::tools::statistics::sum;
|
||||
/// # use num_traits::cast::ToPrimitive;
|
||||
///
|
||||
/// assert_eq!(sum(&[0,1,2,3,4,5]), 15);
|
||||
/// assert_eq!(sum(&[-1,1,-2,2]), 0);
|
||||
/// assert!((sum(&[0.0, 0.1,0.2]).to_f64().unwrap() - 0.3).abs() < 0.001);
|
||||
/// assert!((sum(&[0.0, -0.1,0.2]).to_f64().unwrap() - 0.1).abs() < 0.001);
|
||||
/// ```
|
||||
pub fn sum<T>(list: &[T]) -> T
|
||||
where
|
||||
T: NumAssignRef + ToPrimitive
|
||||
{
|
||||
let mut sum = T::zero();
|
||||
for num in list {
|
||||
sum += num;
|
||||
}
|
||||
sum
|
||||
}
|
||||
|
||||
/// Calculates the mean of a variable x
|
||||
/// ```
|
||||
/// # use proxmox_backup::tools::statistics::mean;
|
||||
///
|
||||
/// assert!((mean(&[0,1,2,3,4,5]).unwrap() - 2.5).abs() < 0.001);
|
||||
/// assert_eq!(mean::<u64>(&[]), None)
|
||||
/// ```
|
||||
pub fn mean<T>(list: &[T]) -> Option<f64>
|
||||
where
|
||||
T: NumAssignRef + ToPrimitive
|
||||
{
|
||||
let len = list.len();
|
||||
if len == 0 {
|
||||
return None
|
||||
}
|
||||
Some(sum(list).to_f64()?/(list.len() as f64))
|
||||
}
|
||||
|
||||
/// Calculates the variance of a variable x
|
||||
/// ```
|
||||
/// # use proxmox_backup::tools::statistics::variance;
|
||||
///
|
||||
/// assert!((variance(&[1,2,3,4]).unwrap() - 1.25).abs() < 0.001);
|
||||
/// assert_eq!(variance::<u64>(&[]), None)
|
||||
/// ```
|
||||
pub fn variance<T>(list: &[T]) -> Option<f64>
|
||||
where
|
||||
T: NumAssignRef + ToPrimitive
|
||||
{
|
||||
covariance(list, list)
|
||||
}
|
||||
|
||||
/// Calculates the (non-corrected) covariance of two variables x,y
|
||||
pub fn covariance<X, Y> (x: &[X], y: &[Y]) -> Option<f64>
|
||||
where
|
||||
X: NumAssignRef + ToPrimitive,
|
||||
Y: NumAssignRef + ToPrimitive,
|
||||
{
|
||||
let len_x = x.len();
|
||||
let len_y = y.len();
|
||||
if len_x == 0 || len_y == 0 || len_x != len_y {
|
||||
return None
|
||||
}
|
||||
|
||||
let mean_x = mean(x)?;
|
||||
let mean_y = mean(y)?;
|
||||
|
||||
let covariance = sum(&(0..len_x).map(|i| {
|
||||
let x = x[i].to_f64().unwrap_or(0.0);
|
||||
let y = y[i].to_f64().unwrap_or(0.0);
|
||||
(x - mean_x)*(y - mean_y)
|
||||
}).collect::<Vec<f64>>());
|
||||
|
||||
Some(covariance/(len_x as f64))
|
||||
}
|
||||
|
||||
/// Returns the factors (a,b) of a linear regression y = a + bx
|
||||
/// for the variables [x,y] or None if the lists are not the same length
|
||||
/// ```
|
||||
/// # use proxmox_backup::tools::statistics::linear_regression;
|
||||
///
|
||||
/// let x = &[0,1,2,3,4];
|
||||
/// let y = &[-4,-2,0,2,4];
|
||||
/// let (a,b) = linear_regression(x,y).unwrap();
|
||||
/// assert!((a - -4.0).abs() < 0.001);
|
||||
/// assert!((b - 2.0).abs() < 0.001);
|
||||
/// ```
|
||||
pub fn linear_regression<X, Y> (x: &[X], y: &[Y]) -> Option<(f64, f64)>
|
||||
where
|
||||
X: NumAssignRef + ToPrimitive,
|
||||
Y: NumAssignRef + ToPrimitive
|
||||
{
|
||||
let len_x = x.len();
|
||||
let len_y = y.len();
|
||||
if len_x == 0 || len_y == 0 || len_x != len_y {
|
||||
return None
|
||||
}
|
||||
|
||||
let mean_x = mean(x)?;
|
||||
let mean_y = mean(y)?;
|
||||
|
||||
let mut covariance = 0.0;
|
||||
let mut variance = 0.0;
|
||||
|
||||
for i in 0..len_x {
|
||||
let x = x[i].to_f64()?;
|
||||
let y = y[i].to_f64()?;
|
||||
|
||||
let x_mean_x = x - mean_x;
|
||||
|
||||
covariance += x_mean_x*(y - mean_y);
|
||||
variance += x_mean_x * x_mean_x;
|
||||
}
|
||||
|
||||
let beta = covariance/variance;
|
||||
let alpha = mean_y - beta*mean_x;
|
||||
Some((alpha,beta))
|
||||
}
|
Loading…
Reference in New Issue