ncps 0.1.1

Neural Circuit Policies - Sparse RNNs inspired by C. elegans, implemented in Rust with Burn
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

#[cfg(test)]
mod tests {
    use burn::backend::NdArray;
    use burn::tensor::Tensor;
    use ncps::cells::cfc_cell::{CfCCell, CfcMode};

    type Backend = NdArray<f32>;

    fn create_test_cell(mode: CfcMode) -> CfCCell<Backend> {
        let device = Default::default();
        CfCCell::new(20, 50, &device)
            .with_mode(mode)
            .with_backbone(128, 1, 0.0)
            .with_activation("lecun_tanh")
    }

    #[test]
    fn test_cfc_cell_creation() {
        let device = Default::default();
        let cell = CfCCell::<Backend>::new(20, 50, &device);

        assert_eq!(cell.input_size(), 20);
        assert_eq!(cell.hidden_size(), 50);
        assert_eq!(cell.mode(), CfcMode::Default);
    }

    #[test]
    fn test_cfc_forward_default() {
        let device = Default::default();
        let cell = create_test_cell(CfcMode::Default);

        let batch_size = 4;
        let input = Tensor::<Backend, 2>::zeros([batch_size, 20], &device);
        let hx = Tensor::<Backend, 2>::zeros([batch_size, 50], &device);

        let (output, new_hidden) = cell.forward(input, hx, 1.0);

        // Output is hidden_size, state is hidden_size
        assert_eq!(output.dims(), [batch_size, 50]);
        assert_eq!(new_hidden.dims(), [batch_size, 50]);
    }

    #[test]
    fn test_cfc_forward_pure() {
        let device = Default::default();
        let cell = create_test_cell(CfcMode::Pure);

        let input = Tensor::<Backend, 2>::random(
            [2, 20],
            burn::tensor::Distribution::Uniform(-0.5, 0.5),
            &device,
        );
        let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

        let (output, _) = cell.forward(input, hx, 1.0);

        // Pure mode should return correct shape output
        assert_eq!(output.dims(), [2, 50]);
        assert_eq!(cell.mode(), CfcMode::Pure);
    }

    #[test]
    fn test_cfc_forward_no_gate() {
        let device = Default::default();
        let cell = create_test_cell(CfcMode::NoGate);

        let input = Tensor::<Backend, 2>::ones([2, 20], &device);
        let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

        let (output, new_hidden) = cell.forward(input, hx, 1.0);

        // NoGate mode should return correct shape output
        assert_eq!(output.dims(), [2, 50]);
        assert_eq!(new_hidden.dims(), [2, 50]);
        assert_eq!(cell.mode(), CfcMode::NoGate);
    }

    #[test]
    fn test_cfc_state_change() {
        let device = Default::default();
        let cell = create_test_cell(CfcMode::Default);

        let input = Tensor::<Backend, 2>::ones([2, 20], &device);
        let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

        let (output, new_hidden) = cell.forward(input, hx.clone(), 1.0);

        // State should have changed
        let diff = (new_hidden.clone() - hx).abs().mean().into_scalar();
        assert!(diff > 0.0);

        // Output should equal new_hidden for CfC
        let output_diff = (output - new_hidden).abs().mean().into_scalar();
        assert!(output_diff < 1e-6, "Output should equal new_hidden");
    }

    #[test]
    fn test_cfc_different_modes_produce_different_results() {
        let device = Default::default();

        let cell_default = create_test_cell(CfcMode::Default);
        let cell_no_gate = create_test_cell(CfcMode::NoGate);

        let input = Tensor::<Backend, 2>::random(
            [2, 20],
            burn::tensor::Distribution::Uniform(-1.0, 1.0),
            &device,
        );
        let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

        let (out1, _) = cell_default.forward(input.clone(), hx.clone(), 1.0);
        let (out2, _) = cell_no_gate.forward(input, hx, 1.0);

        let diff = (out1 - out2).abs().mean().into_scalar();
        assert!(
            diff > 0.01,
            "Different modes should produce different outputs"
        );
    }

    #[test]
    fn test_cfc_backbone_configurations() {
        let device = Default::default();

        // Test different backbone configurations - just ensure they're created without panic
        let _cell_no_backbone = CfCCell::<Backend>::new(20, 50, &device).with_backbone(0, 0, 0.0); // No backbone

        let _cell_deep_backbone =
            CfCCell::<Backend>::new(20, 50, &device).with_backbone(64, 3, 0.2); // Deep with dropout

        // Test forward passes work
        let input = Tensor::<Backend, 2>::zeros([2, 20], &device);
        let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

        let _cell = CfCCell::<Backend>::new(20, 50, &device).with_backbone(0, 0, 0.0);
        let (out1, _) = _cell.forward(input.clone(), hx.clone(), 1.0);
        assert_eq!(out1.dims(), [2, 50]);

        let _cell2 = CfCCell::<Backend>::new(20, 50, &device).with_backbone(64, 3, 0.2);
        let (out2, _) = _cell2.forward(input, hx, 1.0);
        assert_eq!(out2.dims(), [2, 50]);
    }

    #[test]
    fn test_cfc_activations() {
        let device = Default::default();

        for activation in ["relu", "tanh", "gelu", "silu", "lecun_tanh"] {
            let cell = CfCCell::<Backend>::new(20, 50, &device)
                .with_backbone(64, 1, 0.0)
                .with_activation(activation);

            let input = Tensor::<Backend, 2>::zeros([2, 20], &device);
            let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

            let (output, _) = cell.forward(input, hx, 1.0);
            assert_eq!(output.dims()[0], 2);
        }
    }

    #[test]
    #[should_panic]
    fn test_cfc_invalid_activation() {
        let device = Default::default();
        CfCCell::<Backend>::new(20, 50, &device).with_activation("invalid_activation");
    }

    #[test]
    fn test_cfc_batch_processing() {
        let device = Default::default();
        let cell = create_test_cell(CfcMode::Default);

        // Test with batch sizes 1, 8, 32
        for batch in [1, 8, 32] {
            let input = Tensor::<Backend, 2>::zeros([batch, 20], &device);
            let hx = Tensor::<Backend, 2>::zeros([batch, 50], &device);

            let (output, _) = cell.forward(input, hx, 1.0);
            assert_eq!(output.dims(), [batch, 50]);
        }
    }

    #[test]
    fn test_cfc_sparsity_mask() {
        let device = Default::default();
        use ndarray::Array2;

        let mask = Array2::from_shape_vec((50, 50), vec![1.0f32; 2500]).unwrap();

        let cell = CfCCell::<Backend>::new(20, 50, &device).with_sparsity_mask(mask, &device);

        // Ensure forward pass works with mask configured
        let input = Tensor::<Backend, 2>::zeros([2, 20], &device);
        let hx = Tensor::<Backend, 2>::zeros([2, 50], &device);

        let (output, _) = cell.forward(input, hx, 1.0);
        assert_eq!(output.dims(), [2, 50]);
    }
}