rustyml 0.4.0

A high-performance machine learning library in pure Rust, offering statistical utilities, ML algorithms, and future support for neural networks and transformer architectures.
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
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

# Changelog

All notable changes to this project will be documented in this file.
This change log records updates after 2025-3-24.

Please view [SomeB1oody/RustyML](https://github.com/SomeB1oody/RustyML) for more info.

## [v0.4.0] - 2025-4-9
### Added
- Add RMSprop optimizer support to the neural network
- Add MAE(Mean Absolute Error) support in loss function computation
- Add doc comments for neural_network module

## [v0.4.0] - 2025-4-8
### Added
- Add `Adam` optimizer for neural network
- Added new loss functions: `CategoricalCrossEntropy` and `SparseCategoricalCrossEntropy`

### Changed
- Replaced slices with `ndarray`'s `ArrayView` to improve consistency and compatibility with numerical operations in metric module
- Refactor polynomial feature generation to exclude constant term

## [v0.4.0] - 2025-4-7
### Added
- Add neural network support(initial implementation)

## [v0.3.0] - 2025-4-6
### Added
- Add dataset module and put iris and diabetes datasets in it
- Enhance SVM/LDA documentation

## [v0.3.0] - 2025-4-5
### Changed
- Refactor to use `ArrayView` for memory-efficient data handling
- Refactor outputs to use `Array` instead of `Vec` in DBSCAN and KNN

## [v0.2.1] - 2025-4-4
### Added
- Add variance calculation in math module
- Add MSE calculation in metric module
- Print info after training completes for `fit` functions of struct `LDA`, `SVC`, `LinearSVC` and `PCA`
- Integrate Rayon for parallel computation across modules (**7s faster** in `cargo test`!!!)

### Changed
- Refactor functions in math module and metric module to use `ArrayView1` for improved efficiency

### Removed
- Remove MSE calculation(named mean_squared_error, but actually calculate variance) in math module

## [v0.2.1] - 2025-4-3
### Added
- Add t-Distributed Stochastic Neighbor Embedding (t-SNE) implementation

### Changed
- Remove gaussian kernel calculation function, put gaussian kernel calculation directly in function `fit` of struct `MeanShift`

## [v0.2.1] - 2025-4-2
### Added
- Add LinearSVC support
- Add KernelPCA support
- Add `ProcessingError(String)` to `crate::ModelError` 
- Add LDA(Linear Discriminant Analysis) support

### Changed
- Change the location of function `standardize` from `crate::utility::principal_component_analysis` to `crate::utility`

## [v0.2.1] - 2025-4-1
### Added
- Add SVC(Support Vector Classification) support

## [v0.2.0] - 2025-4-1
### Added
- Add `train_test_split` function in utility module to split dataset for training and dataset for test
- Add function `normalized_mutual_info` and `adjusted_mutual_info` to metric module to calculate NMI and AMI info
- Add AUC-ROC value calculation in metric module

## [v0.2.0] - 2025-3-31
### Changed
- Change `principal_component_analysis` module to `utility` module, change `principal_component_analysis_test` module to `utility_test` module
- Keep the algorithm functions in the math module, and move the functions that evaluate the model's performance (such as R-square values) and structures (confusion matrices) to the metric module. Some of them are used in both ways, then keep them in both modules.
- Change the output of some of the functions in math module and metric module from `T` to `Result<T, crate::ModelError>`

## [v0.1.1] - 2025-3-31
### Added
- Add function `preliminary_check` in machine_learning module to performs validation checks on the input data matrices
- Add confusion matrix in math module

### Changed
- Change type of field `coefficients` of struct `LinearRegression` from `Option<Vec<f64>>` to `Option<Array1<f64>>`
- Change the output of some methods of struct `LinearRegression` from `Vec<f64>` to `Array1<f64>`
- Change variant `InputValidationError` of enum type `ModelError` from `InputValidationError(&str)` to `InputValidationError(String)`

## [v0.1.0] - 2025-3-30
### Added
- Add function `fit_predict` for some models
- Add examples for functions in math.rs
- Add input validation
- Add doc comments for machine learning modules
- Add prelude module(all re-exports are there)

### Changed
- Change input types of function `fit`, `predict` and `fit_predict` to `Array1` and `Array2`
- Rename the crate from `rust_ai` to `rustyml`
- Change the output of function `fit` from `&mut Self` to `Result<&mut Self, ModelError>` or `Result<&mut Self, Box<dyn std::error::Error>>`

## [v0.1.0] - 2025-3-29
### Added
- Add function `generate_tree_structure` for `DecisionTree` to generate tree structure as string
- Add isolation forest implementation
- Add PCA(Principal Component Analysis) implementation
- Add function `standard_deviation` in math module to calculates the standard deviation of a set of values

## [v0.1.0] - 2025-3-28
### Added
- Add Decision Tree model
- Add following functions to math.rs:
    - `entropy`: Calculates the entropy of a label set
    - `gini`: Calculates the Gini impurity of a label set
    - `information_gain`: Calculates the information gain when splitting a dataset
    - `gain_ratio`: Calculates the gain ratio for a dataset split
    - `mean_squared_error`: Calculates the Mean Squared Error (MSE) of a set of values

### Changed
- Replaced string-based distance calculation method options with an enum `crate::machine_learning::DistanceCalculation`
- For KNN model: replaced string-based weight function options with an enum `crate::machine_learning::knn::WeightingStrategy`
- For decision tree: replaced string-based algorithm options with an enum `crate::machine_learning::decision_tree::Algorithm`

## [v0.1.0] - 2025-3-27
### Added
- Add changelog.md to record updates
- Add DBSCAN model
- Add function `fit_predict` to fit and predict in one step
- Add doc comments to tell user `p` value of function `minkowski_distance` in model is always 3

## [v0.1.0] - 2025-3-26
### Added
- Add "xx model converged at iteration x, cost: x" when finishing `fit`
- Add description for `n_iter` field
- Add getter functions for `KMeans`
- implement `Default` trait for `KMeans`

### Changed
- Rename `max_iteration` and `tolerance` to `max_iter` and `tol`
- Change doc comments to enhanced consistency

### Removed
- Remove examples in math.rs(add them back later)

## [v0.1.0] - 2025-3-25
### Added
- Add MeanShift model
- Add `InputValidationError` in `ModelError`, indicating the input data provided  does not meet the expected format, type, or validation rules
- Add `gaussian_kernel` in math module, calculate the Gaussian kernel (RBF kernel)

### Changed
- Change the output of all `predict` functions(except KNN) from `T` to `Result<T, crate::ModelError>`
- Correct doc comments