added temp enc

Cargo.lock

@@ -5107,6 +5107,7 @@ name = "spikers"
 version = "0.1.0"
 dependencies = [
  "burn",
+ "burn-autodiff",
  "rstest",
  "rstest_reuse",
 ]

Cargo.toml

@@ -4,7 +4,8 @@ version = "0.1.0"
 edition = "2024"
 
 [dependencies]
-burn = { version = "0.20.1", features = ["wgpu", "std", "fusion", "ndarray"] }
+burn = { version = "0.20.1", features = ["wgpu", "std", "fusion", "ndarray", "cuda"] }
+burn-autodiff = "0.20.1"
 
 [dev-dependencies]
 rstest = "0.26.1"

examples/simple1d.rs

@@ -0,0 +1,41 @@
+#![allow(non_snake_case)]
+use burn::Tensor;
+//
+use burn::prelude::Backend;
+use spikers::encoders;
+
+fn main() {
+    let useGPU = {
+        // use burn::backend::Wgpu;
+        use burn::backend::wgpu::WgpuDevice;
+        match WgpuDevice::DefaultDevice {
+            WgpuDevice::Cpu => {
+                println!("GPU Not Found, using NDArray Backend");
+                false
+            }
+            _ => true,
+        }
+    };
+
+    if useGPU {
+        use burn::backend::Wgpu;
+        use burn::backend::wgpu::WgpuDevice;
+        run::<Wgpu>(WgpuDevice::default());
+    } else {
+        use burn::backend::NdArray;
+        use burn::backend::ndarray::NdArrayDevice;
+        run::<NdArray>(NdArrayDevice::default());
+    }
+    // encode(data, numSteps);
+}
+
+fn run<B: Backend>(device: B::Device) {
+    println!("Using {:?} Device", B::name(&device));
+    // let someTens = Tensor::<B, 1>::random(Shape::new([3]), Distribution::Normal(0., 1.), &device);
+    let someTens = Tensor::<B, 1>::from_floats([0., 0.2, 0.4, 0.6, 0.8, 1.0], &device);
+    println!("{}", encoders::rate::encode::<B, 1, 2>(&someTens, 10));
+    println!(
+        "{}",
+        encoders::temporal::encode::<B, 1, 2>(&someTens, 10, true)
+    );
+}

src/encoders/rate.rs

@@ -1,5 +1,5 @@
-#[allow(non_snake_case)]
+#![allow(non_snake_case)]
-use burn::{Tensor, prelude::*, tensor::Distribution};
+use burn::{Tensor, prelude::Backend, tensor::Distribution};
 
 pub fn encode<B: Backend, const D: usize, const D2: usize>(
     data: &Tensor<B, D>,

src/encoders/temporal.rs

@@ -0,0 +1,120 @@
+#![allow(non_snake_case)]
+use burn::{Tensor, prelude::Backend};
+
+// Must test this with python implementation, across dimensions.
+pub fn encode<B: Backend, const D: usize, const D2: usize>(
+    data: &Tensor<B, D>,
+    time_steps: usize,
+    linear: bool,
+) -> Tensor<B, D2> {
+    if time_steps < 1 {
+        panic!("Time steps must be greater than 0");
+    }
+    // assumes the data is normalised.
+    if !linear {
+        panic!("Log conversion not implemented.")
+    }
+    let idxs = data
+        .clone()
+        .mul_scalar((time_steps - 1) as i32)
+        .round()
+        .unsqueeze_dim::<D2>(0)
+        .repeat_dim(0, time_steps)
+        .int();
+    let out = idxs.zeros_like();
+    let ones = out.ones_like();
+    out.scatter(0, idxs, ones, burn::tensor::IndexingUpdateOp::Add)
+        .float()
+        .div_scalar(time_steps as u32)
+}
+
+#[cfg(test)]
+mod temporalEncoderTests {
+
+    use burn::{
+        Tensor,
+        backend::{NdArray, ndarray::NdArrayDevice},
+        tensor::{Float, Shape},
+    };
+    type B = NdArray;
+    use super::encode;
+    use rstest::rstest;
+    use rstest_reuse::{self, *};
+
+    #[template]
+    #[rstest]
+    #[case(1, 1)]
+    #[case(10, 1)]
+    #[case(100, 1)]
+    #[case(1, 10)]
+    #[case(10, 10)]
+    #[case(100, 10)]
+    #[case(1, 100)]
+    #[case(10, 100)]
+    #[case(100, 100)]
+
+    fn testShapeStepComb(#[case] t: usize, #[case] s: usize) {}
+
+    #[apply(testShapeStepComb)]
+    fn test1DZeros(#[case] t: usize, #[case] s: usize) {
+        let device = NdArrayDevice::default();
+        let t0 = Tensor::<B, 1>::zeros(Shape::new([s]), &device);
+        let out = encode::<B, 1, 2>(&t0, t, true);
+        assert_eq!(
+            out.shape(),
+            Shape::new([t, s]),
+            "Shape testing failed. Expected Shape: [{}, {}], got: {}",
+            s,
+            t,
+            out.shape()
+        );
+        assert_eq!(out.sum().into_scalar(), s as f32);
+    }
+
+    #[apply(testShapeStepComb)]
+    fn test1DOnes(#[case] t: usize, #[case] s: usize) {
+        let device = NdArrayDevice::default();
+        let t0 = Tensor::<B, 1, Float>::ones(Shape::new([s]), &device);
+        let out = encode::<B, 1, 2>(&t0, t, true);
+        assert_eq!(
+            out.shape(),
+            Shape::new([t, s]),
+            "Shape testing failed. Expected Shape: [{}, {}], got: {}",
+            t,
+            s,
+            out.shape()
+        );
+        assert_eq!(out.clone().int().sum().into_scalar(), (s) as i64);
+    }
+    #[apply(testShapeStepComb)]
+    fn test2DZeros(#[case] t: usize, #[case] s: usize) {
+        let device = NdArrayDevice::default();
+        let t0 = Tensor::<B, 2>::zeros(Shape::new([s; 2]), &device);
+        let out = encode::<B, 2, 3>(&t0, t, true);
+        assert_eq!(
+            out.shape(),
+            Shape::new([t, s, s]),
+            "Shape testing failed. Expected Shape: [{}, {}], got: {}",
+            s,
+            t,
+            out.shape()
+        );
+        assert_eq!(out.sum().into_scalar(), (s * s) as f32);
+    }
+
+    #[apply(testShapeStepComb)]
+    fn test2DOnes(#[case] t: usize, #[case] s: usize) {
+        let device = NdArrayDevice::default();
+        let t0 = Tensor::<B, 2, Float>::ones(Shape::new([s; 2]), &device);
+        let out = encode::<B, 2, 3>(&t0, t, true);
+        assert_eq!(
+            out.shape(),
+            Shape::new([t, s, s]),
+            "Shape testing failed. Expected Shape: [{}, {}], got: {}",
+            t,
+            s,
+            out.shape()
+        );
+        assert_eq!(out.clone().int().sum().into_scalar(), (s * s) as i64);
+    }
+}

src/lib.rs

@@ -1,17 +1,4 @@
 #[allow(non_snake_case)]
-mod encoders;
+pub mod encoders;
-
+pub mod neurons;
-pub fn add(left: u64, right: u64) -> u64 {
+pub mod surrogate;
-    left + right
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn it_works() {
-        let result = add(2, 2);
-        assert_eq!(result, 4);
-    }
-}

src/neurons/leaky.rs

@@ -0,0 +1,14 @@
+use burn::{Tensor, module::Module, prelude::Backend};
+
+#[derive(Debug, Module, Clone, Default)]
+pub struct LIF;
+
+impl LIF {
+    pub fn new() -> Self {
+        Self
+    }
+
+    pub fn forward<B: Backend, const D: usize>(&self, input: Tensor<B, D>) -> Tensor<B, D> {
+        input
+    }
+}

src/neurons/mod.rs

@@ -0,0 +1 @@
+mod leaky;

src/surrogate/fast_sigmoid.rs

@@ -0,0 +1 @@
+

src/surrogate/mod.rs

@@ -0,0 +1 @@
+mod fast_sigmoid;