numrs2 0.2.0

A Rust implementation inspired by NumPy for numerical computing (NumRS2)
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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262

//! Neural network operations for Python bindings

use crate::array::Array;
use crate::python::array::PyArray;
use pyo3::exceptions::PyValueError;
use pyo3::prelude::*;
use scirs2_core::ndarray::{ArrayView1, ArrayView2, IxDyn};

/// Helper function to convert Array<f64> to ArrayView1<f64>
fn to_array_view1(arr: &Array<f64>) -> Result<ArrayView1<f64>, String> {
    if arr.ndim() != 1 {
        return Err(format!("Expected 1D array, got {}D", arr.ndim()));
    }
    arr.data
        .view()
        .into_dimensionality::<scirs2_core::ndarray::Ix1>()
        .map_err(|e| format!("Failed to convert to 1D view: {}", e))
}

/// Helper function to convert Array<f64> to ArrayView2<f64>
fn to_array_view2(arr: &Array<f64>) -> Result<ArrayView2<f64>, String> {
    if arr.ndim() != 2 {
        return Err(format!("Expected 2D array, got {}D", arr.ndim()));
    }
    arr.data
        .view()
        .into_dimensionality::<scirs2_core::ndarray::Ix2>()
        .map_err(|e| format!("Failed to convert to 2D view: {}", e))
}

/// Apply ReLU activation function
#[pyfunction]
fn relu(x: &PyArray) -> PyResult<PyArray> {
    use crate::nn::activation;

    if x.inner.ndim() == 1 {
        let view = to_array_view1(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::relu(&view)
            .map_err(|e| PyValueError::new_err(format!("ReLU failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else if x.inner.ndim() == 2 {
        let view = to_array_view2(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::relu_2d(&view)
            .map_err(|e| PyValueError::new_err(format!("ReLU failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else {
        Err(PyValueError::new_err(format!(
            "ReLU only supports 1D and 2D arrays, got {}D",
            x.inner.ndim()
        )))
    }
}

/// Apply sigmoid activation function
#[pyfunction]
fn sigmoid(x: &PyArray) -> PyResult<PyArray> {
    use crate::nn::activation;

    if x.inner.ndim() == 1 {
        let view = to_array_view1(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::sigmoid(&view)
            .map_err(|e| PyValueError::new_err(format!("Sigmoid failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else if x.inner.ndim() == 2 {
        let view = to_array_view2(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::sigmoid_2d(&view)
            .map_err(|e| PyValueError::new_err(format!("Sigmoid failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else {
        Err(PyValueError::new_err(format!(
            "Sigmoid only supports 1D and 2D arrays, got {}D",
            x.inner.ndim()
        )))
    }
}

/// Apply tanh activation function
#[pyfunction]
fn tanh(x: &PyArray) -> PyResult<PyArray> {
    use crate::nn::activation;

    if x.inner.ndim() == 1 {
        let view = to_array_view1(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::tanh(&view)
            .map_err(|e| PyValueError::new_err(format!("Tanh failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else if x.inner.ndim() == 2 {
        let view = to_array_view2(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::tanh_2d(&view)
            .map_err(|e| PyValueError::new_err(format!("Tanh failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else {
        Err(PyValueError::new_err(format!(
            "Tanh only supports 1D and 2D arrays, got {}D",
            x.inner.ndim()
        )))
    }
}

/// Apply softmax activation function
#[pyfunction]
fn softmax(x: &PyArray, axis: Option<isize>) -> PyResult<PyArray> {
    use crate::nn::activation;
    let _axis = axis.unwrap_or(-1);

    if x.inner.ndim() == 1 {
        let view = to_array_view1(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = activation::softmax(&view)
            .map_err(|e| PyValueError::new_err(format!("Softmax failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else if x.inner.ndim() == 2 {
        let view = to_array_view2(&x.inner).map_err(PyValueError::new_err)?;
        // For 2D, apply softmax along last axis (axis 1)
        let result_nd = activation::softmax_2d(&view, 1)
            .map_err(|e| PyValueError::new_err(format!("Softmax failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else {
        Err(PyValueError::new_err(format!(
            "Softmax only supports 1D and 2D arrays, got {}D",
            x.inner.ndim()
        )))
    }
}

/// Compute mean squared error loss
#[pyfunction]
fn mse_loss(predictions: &PyArray, targets: &PyArray) -> PyResult<f64> {
    use crate::nn::loss;
    use crate::nn::ReductionMode;

    if predictions.inner.ndim() == 1 {
        let pred_view = to_array_view1(&predictions.inner).map_err(PyValueError::new_err)?;
        let targ_view = to_array_view1(&targets.inner).map_err(PyValueError::new_err)?;
        loss::mse_loss(&pred_view, &targ_view, ReductionMode::Mean)
            .map_err(|e| PyValueError::new_err(format!("MSE loss calculation failed: {}", e)))
    } else if predictions.inner.ndim() == 2 {
        let pred_view = to_array_view2(&predictions.inner).map_err(PyValueError::new_err)?;
        let targ_view = to_array_view2(&targets.inner).map_err(PyValueError::new_err)?;
        loss::mse_loss_2d(&pred_view, &targ_view, ReductionMode::Mean)
            .map_err(|e| PyValueError::new_err(format!("MSE loss calculation failed: {}", e)))
    } else {
        Err(PyValueError::new_err(format!(
            "MSE loss only supports 1D and 2D arrays, got {}D",
            predictions.inner.ndim()
        )))
    }
}

/// Compute cross-entropy loss
#[pyfunction]
fn cross_entropy_loss(predictions: &PyArray, targets: &PyArray) -> PyResult<f64> {
    use crate::nn::loss;
    use crate::nn::ReductionMode;

    if predictions.inner.ndim() == 1 {
        let pred_view = to_array_view1(&predictions.inner).map_err(PyValueError::new_err)?;
        let targ_view = to_array_view1(&targets.inner).map_err(PyValueError::new_err)?;
        loss::binary_cross_entropy(&pred_view, &targ_view, ReductionMode::Mean).map_err(|e| {
            PyValueError::new_err(format!("Cross-entropy loss calculation failed: {}", e))
        })
    } else if predictions.inner.ndim() == 2 {
        let pred_view = to_array_view2(&predictions.inner).map_err(PyValueError::new_err)?;
        let targ_view = to_array_view2(&targets.inner).map_err(PyValueError::new_err)?;
        loss::categorical_cross_entropy(&pred_view, &targ_view, ReductionMode::Mean).map_err(|e| {
            PyValueError::new_err(format!("Cross-entropy loss calculation failed: {}", e))
        })
    } else {
        Err(PyValueError::new_err(format!(
            "Cross-entropy loss only supports 1D and 2D arrays, got {}D",
            predictions.inner.ndim()
        )))
    }
}

/// Apply dropout (training mode simulation)
#[pyfunction]
fn dropout(x: &PyArray, p: f64) -> PyResult<PyArray> {
    if !(0.0..1.0).contains(&p) {
        return Err(PyValueError::new_err(
            "Dropout probability must be in [0, 1)",
        ));
    }

    use crate::nn::normalization;

    if x.inner.ndim() == 1 {
        let view = to_array_view1(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = normalization::dropout(&view, p, true)
            .map_err(|e| PyValueError::new_err(format!("Dropout failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else if x.inner.ndim() == 2 {
        let view = to_array_view2(&x.inner).map_err(PyValueError::new_err)?;
        let result_nd = normalization::dropout_2d(&view, p, true)
            .map_err(|e| PyValueError::new_err(format!("Dropout failed: {}", e)))?;
        let result = Array::from_ndarray(result_nd.into_dyn());
        Ok(PyArray { inner: result })
    } else {
        Err(PyValueError::new_err(format!(
            "Dropout only supports 1D and 2D arrays, got {}D",
            x.inner.ndim()
        )))
    }
}

/// Apply batch normalization
#[pyfunction]
fn batch_norm(x: &PyArray, eps: Option<f64>) -> PyResult<PyArray> {
    let eps = eps.unwrap_or(1e-5);
    use crate::nn::normalization;
    use scirs2_core::ndarray::Array1;

    if x.inner.ndim() != 2 {
        return Err(PyValueError::new_err(
            "Batch normalization requires a 2D array",
        ));
    }

    let view = to_array_view2(&x.inner).map_err(PyValueError::new_err)?;

    // Create learnable parameters (gamma and beta) initialized to 1 and 0
    let n_features = x.inner.shape()[1];
    let gamma = Array1::from_elem(n_features, 1.0);
    let beta = Array1::from_elem(n_features, 0.0);

    let result_nd = normalization::batch_norm_1d(&view, &gamma.view(), &beta.view(), eps)
        .map_err(|e| PyValueError::new_err(format!("Batch normalization failed: {}", e)))?;
    let result = Array::from_ndarray(result_nd.into_dyn());
    Ok(PyArray { inner: result })
}

/// Register neural network functions
pub fn register(m: &Bound<'_, PyModule>) -> PyResult<()> {
    // Create nn submodule
    let nn_module = PyModule::new(m.py(), "nn")?;

    // Add activation functions
    nn_module.add_function(wrap_pyfunction!(relu, m)?)?;
    nn_module.add_function(wrap_pyfunction!(sigmoid, m)?)?;
    nn_module.add_function(wrap_pyfunction!(tanh, m)?)?;
    nn_module.add_function(wrap_pyfunction!(softmax, m)?)?;

    // Add loss functions
    nn_module.add_function(wrap_pyfunction!(mse_loss, m)?)?;
    nn_module.add_function(wrap_pyfunction!(cross_entropy_loss, m)?)?;

    // Add normalization functions
    nn_module.add_function(wrap_pyfunction!(dropout, m)?)?;
    nn_module.add_function(wrap_pyfunction!(batch_norm, m)?)?;

    m.add_submodule(&nn_module)?;

    Ok(())
}