Gen tabular datasets

[oojo12]

Goal:

This PR aims to make writing benchmarking code a bit easier for developers by parametrizing the datasets dimension.

Additions to linfa-datasets:

1. A function to allow generation of fictitious tabular datasets.
2. Two optional dependencies (statrs and rand)

resolves #259

Example

use ndarray::Array;
use linfa::Dataset;
use ndarray_rand::RandomExt;
use statrs::distribution::{Binomial, DiscreteUniform};
use rand::distributions::Distribution as randDistribution;

fn main() {
    let feat_distr = Binomial::new(0.5, 5 ).unwrap();
    let target_distr = DiscreteUniform::new(0, 5).unwrap();
    let data = make_dataset(5, 5, 2, feat_distr, target_distr);
    
    println!("{:#?}", data);
}

fn make_dataset<X, Y>(num_rows: usize, num_feats: usize, num_targets: usize, feat_distr: X, target_distr: Y) -> Dataset<f64, f64>
where
    X: randDistribution<f64>,
    Y: randDistribution<f64>
{
    let features = Array::random((num_rows, num_feats), feat_distr);
    let targets = Array::random((num_rows, num_targets), target_distr);

    println!("{:#?}", features);
    println!("{:#?}", targets);

    Dataset::new(features, targets)
}

Associated output

[[1.0, 2.0, 0.0, 3.0, 2.0],
 [3.0, 1.0, 2.0, 4.0, 2.0],
 [2.0, 3.0, 5.0, 2.0, 2.0],
 [2.0, 1.0, 4.0, 2.0, 2.0],
 [2.0, 4.0, 3.0, 1.0, 3.0]], shape=[5, 5], strides=[5, 1], layout=Cc (0x5), const ndim=2
[[3.0, 4.0],
 [5.0, 3.0],
 [5.0, 5.0],
 [2.0, 1.0],
 [2.0, 5.0]], shape=[5, 2], strides=[2, 1], layout=Cc (0x5), const ndim=2
DatasetBase {
    records: [[1.0, 2.0, 0.0, 3.0, 2.0],
     [3.0, 1.0, 2.0, 4.0, 2.0],
     [2.0, 3.0, 5.0, 2.0, 2.0],
     [2.0, 1.0, 4.0, 2.0, 2.0],
     [2.0, 4.0, 3.0, 1.0, 3.0]], shape=[5, 5], strides=[5, 1], layout=Cc (0x5), const ndim=2,
    targets: [[3.0, 4.0],
     [5.0, 3.0],
     [5.0, 5.0],
     [2.0, 1.0],
     [2.0, 5.0]], shape=[5, 2], strides=[2, 1], layout=Cc (0x5), const ndim=2,
    weights: [], shape=[0], strides=[0], layout=CFcf (0xf), const ndim=1,
    feature_names: [],
}

the feature array can look like this with a Laplace distribution so we are able to handle both continuous and discrete cases

[[-4.360595740165293, 8.155513405816892, -8.478208753936332, 17.23637839788624, 1.7384987675286867],
 [2.6468478875196775, -0.0055614415309673815, 3.6241064859319967, 5.072362710780806, 7.617604811173227],
 [0.79789581354939, -1.4199269716778213, 0.18709110069858814, 1.4115467400849417, -0.8056722910319407],
 [-11.46311409107569, 3.2708729527840843, 4.16040511713887, 1.6879416430935898, 2.451106903259817],
 [-2.7718365458998515, 1.941470709966388, 14.061929511357462, 5.111381148046025, 0.7341567591021234]], shape=[5, 5], strides=[5, 1], layout=Cc (0x5), const ndim=2

Adjustments to replicate with the above example

use statrs::distribution::{Binomial, DiscreteUniform, Laplace};
let feat_distr = Laplace::new(0.5, 5. ).unwrap();

Outstanding Questions

1. Should an example be added to an "examples" directory?
2. How do we go about documenting?

[codecov-commenter]

Codecov Report

Base: 39.17% // Head: 38.68% // Decreases project coverage by -0.48% ⚠️

Coverage data is based on head (2829ed2) compared to base (1f260de).
Patch coverage: 0.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #260      +/-   ##
==========================================
- Coverage   39.17%   38.68%   -0.49%     
==========================================
  Files          92       92              
  Lines        6096     6080      -16     
==========================================
- Hits         2388     2352      -36     
- Misses       3708     3728      +20     

Impacted Files	Coverage Δ
datasets/src/generate.rs	60.00% <0.00%> (-25.72%)	⬇️
algorithms/linfa-linear/src/float.rs	20.00% <0.00%> (-13.34%)	⬇️
algorithms/linfa-nn/src/heap_elem.rs	0.00% <0.00%> (-10.00%)	⬇️
algorithms/linfa-logistic/src/argmin_param.rs	41.17% <0.00%> (-8.83%)	⬇️
...rithms/linfa-clustering/src/k_means/hyperparams.rs	27.08% <0.00%> (-6.25%)	⬇️
...linfa-clustering/src/appx_dbscan/cells_grid/mod.rs	46.42% <0.00%> (-6.21%)	⬇️
...orithms/linfa-clustering/src/optics/hyperparams.rs	20.58% <0.00%> (-5.89%)	⬇️
algorithms/linfa-linear/src/glm/distribution.rs	54.90% <0.00%> (-5.89%)	⬇️
algorithms/linfa-svm/src/classification.rs	42.10% <0.00%> (-5.27%)	⬇️
src/composing/platt_scaling.rs	30.43% <0.00%> (-5.22%)	⬇️
... and 52 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[YuhanLiin]

For documentation just describe what the function generates. Something like: "This function generates a random, multi-target dataset with the specified number of rows, features, and targets. The values of the features and targets are determined using the provided distributions". You can just put the example into the function docs as a doctest.

[oojo12]

cool I'll do that tomorrow.