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

use smartnoise_validator::errors::*;

use crate::NodeArguments;
use smartnoise_validator::base::{ReleaseNode, IndexKey};
use smartnoise_validator::utilities::take_argument;
use crate::components::Evaluable;
use ndarray::{ArrayD, Array};

use smartnoise_validator::{proto, Float};
use crate::components::mean::mean;
use ndarray::prelude::*;
use std::iter::FromIterator;

impl Evaluable for proto::Covariance {
    fn evaluate(&self, _privacy_definition: &Option<proto::PrivacyDefinition>, mut arguments: NodeArguments) -> Result<ReleaseNode> {
        let delta_degrees_of_freedom = if self.finite_sample_correction {1} else {0} as usize;
        if arguments.contains_key::<IndexKey>(&"data".into()) {
            let data = take_argument(&mut arguments, "data")?.array()?.float()?;
            let covariances = matrix_covariance(&data, delta_degrees_of_freedom)?.into_iter()
                .flatten()
                .collect::<Vec<Float>>();

            // flatten into a row vector, every column is a release
            return Ok(ReleaseNode::new(arr1(&covariances).insert_axis(Axis(0)).into_dyn().into()));
        }
        if arguments.contains_key::<IndexKey>(&"left".into()) && arguments.contains_key::<IndexKey>(&"right".into()) {
            let left = take_argument(&mut arguments, "left")?.array()?.float()?;
            let right = take_argument(&mut arguments, "right")?.array()?.float()?;

            let cross_covariances = matrix_cross_covariance(&left, &right, delta_degrees_of_freedom)?;

            // flatten into a row vector, every column is a release
            return Ok(ReleaseNode::new(Array::from_iter(cross_covariances.iter())
                .insert_axis(Axis(0)).into_dyn().mapv(|v| *v).into()));
        }
        Err("insufficient data supplied to Covariance".into())
    }
}

/// Construct upper triangular of covariance matrix from data matrix.
///
/// # Arguments
/// * `data` - Data for which you want covariance matrix.
/// * `delta_degrees_of_freedom` - 0 for population, 1 for finite sample correction
///
/// # Return
/// Upper triangular of covariance matrix of your data.
///
/// # Example
/// ```
/// use ndarray::{ArrayD, arr2};
/// use smartnoise_runtime::components::covariance::matrix_covariance;
///
/// // covariance matrix is:
/// // [12.5  7.5 -7.5]
/// // [ 7.5  4.5 -4.5]
/// // [-7.5 -4.5  4.5]
///
/// let data = arr2(&[ [0., 2., 9.], [5., 5., 6.] ]).into_dyn();
/// let cov_mat = matrix_covariance(&data, 1).unwrap();
/// assert_eq!(cov_mat, vec![ vec![12.5, 7.5, -7.5], vec![4.5, -4.5], vec![4.5] ]);
/// ```
pub fn matrix_covariance(data: &ArrayD<Float>, delta_degrees_of_freedom: usize) -> Result<Vec<Vec<Float>>> {

    let means: Vec<Float> = mean(&data)?.iter().cloned().collect();

    let mut covariances: Vec<Vec<Float>> = Vec::new();
    data.gencolumns().into_iter().enumerate()
        .for_each(|(left_i, left_col)| {
            let mut col_covariances: Vec<Float> = Vec::new();
            data.gencolumns().into_iter().enumerate()
                .filter(|(right_i, _right_col)| &left_i <= right_i)
                .for_each(|(right_i, right_col)|
                    col_covariances.push(covariance(
                        &left_col, &right_col,
                        means[left_i], means[right_i],
                        delta_degrees_of_freedom)));
            covariances.push(col_covariances);
        });

    Ok(covariances)
}

/// Construct cross-covariance matrix from pair of data matrices.
///
/// Element (i,j) of the cross-covariance matrix will be the covariance of the
/// column i of `left` and column `j` of `right`
///
/// # Arguments
/// * `left` - One of the two matrices for which you want the cross-covariance matrix.
/// * `right` - One of the two matrices for which you want the cross-covariance matrix.
/// * `delta_degrees_of_freedom` - 0 for population, 1 for finite sample correction
///
/// # Return
/// Full cross-covariance matrix.
///
/// # Example
/// ```
/// use ndarray::{ArrayD, arr2};
/// use smartnoise_runtime::components::covariance::matrix_cross_covariance;
///
/// let left = arr2(&[ [1., 3., 5.,], [2., 4., 6.] ]).into_dyn();
/// let right = arr2(&[ [2., 4., 6.], [1., 3., 5.] ]).into_dyn();
///
/// let cross_covar = matrix_cross_covariance(&left, &right, 1).unwrap();
/// let left_covar = matrix_cross_covariance(&left, &left, 1).unwrap();
///
/// // cross-covariance of left and right matrices
/// assert_eq!(cross_covar, arr2(&[ [-0.5, -0.5, -0.5], [-0.5, -0.5, -0.5], [-0.5, -0.5, -0.5] ]).into_dyn());
///
/// // cross-covariance of left with itself is equivalent to the standard covariance matrix
/// assert_eq!(left_covar, arr2(&[ [0.5, 0.5, 0.5], [0.5, 0.5, 0.5], [0.5, 0.5, 0.5] ]).into_dyn());
/// ```
pub fn matrix_cross_covariance(
    left: &ArrayD<Float>, right: &ArrayD<Float>,
    delta_degrees_of_freedom: usize
) -> Result<ArrayD<Float>> {

    let left_means: Vec<Float> = mean(&left)?.iter().cloned().collect();
    let right_means: Vec<Float> = mean(&right)?.iter().cloned().collect();

    let covariances = left.gencolumns().into_iter()
        .zip(left_means.iter())
        .flat_map(|(column_left, mean_left)|
            right.gencolumns().into_iter()
                .zip(right_means.iter())
                .map(|(column_right, mean_right)| covariance(
                    &column_left, &column_right,
                    *mean_left, *mean_right,
                delta_degrees_of_freedom))
                .collect::<Vec<Float>>())
        .collect::<Vec<Float>>();

    match Array::from_shape_vec((left_means.len(), right_means.len()), covariances) {
        Ok(array) => Ok(array.into_dyn()),
        Err(_) => Err("unable to form cross-covariance matrix".into())
    }
}

/// Get covariance between two 1D-arrays.
///
/// # Arguments
/// * `left` - One of the two arrays for which you want the covariance.
/// * `right` - One of the two arrays for which you want the covariance.
/// * `mean_left` - Arithmetic mean of the left array.
/// * `mean_right` - Arithmetic mean of the right array.
/// * `delta_degrees_of_freedom` - 0 for population, 1 for finite sample correction
///
/// # Return
/// Covariance of the two arrays.
///
/// # Example
/// ```
/// use ndarray::{ArrayD, arr1};
/// use smartnoise_runtime::components::covariance::covariance;
///
/// let left = arr1(&[1.,2.,3.]);
/// let right = arr1(&[4.,5.,6.]);
/// let mean_left = 2.;
/// let mean_right = 5.;
/// let cov = covariance(&left.view(), &right.view(), mean_left, mean_right, 1);
/// assert_eq!(cov, 1.);
/// ```
pub fn covariance(
    left: &ArrayView1<Float>, right: &ArrayView1<Float>,
    mean_left: Float, mean_right: Float,
    delta_degrees_of_freedom: usize
) -> Float {
    left.iter()
        .zip(right)
        .fold(0., |sum, (val_left, val_right)|
            sum + ((val_left - mean_left) * (val_right - mean_right))) / ( (left.len() - delta_degrees_of_freedom) as Float)
}