Add some backprop test.

src/tensor.rs

@@ -169,7 +169,7 @@ impl Tensor {
     }
 
     pub fn var<A: crate::device::NdArray>(array: A, device: Device) -> Result<Self> {
-        Self::new_impl(array, device, false)
+        Self::new_impl(array, device, true)
     }
 
     pub(crate) fn same_shape_binary_op(&self, rhs: &Self, op: &'static str) -> Result<&Shape> {
@@ -329,6 +329,7 @@ impl Tensor {
 
     pub fn backward(&self) -> Result<HashMap<TensorId, Tensor>> {
         let sorted_nodes = self.sorted_nodes();
+        println!("{}", sorted_nodes.len());
         let mut grads = HashMap::new();
         grads.insert(self.id, self.ones_like());
         for node in sorted_nodes.iter() {

tests/grad_tests.rs

@@ -0,0 +1,19 @@
+use anyhow::{Context, Result};
+use candle::{Device, Tensor};
+
+#[test]
+fn simple_grad() -> Result<()> {
+    let x = Tensor::var(&[3f32, 1., 4.], Device::Cpu)?;
+    let five = Tensor::new(&[5f32, 5., 5.], Device::Cpu)?;
+    let x_times_five = x.mul(&five)?;
+    let x2 = x.mul(&x)?;
+    let y = x2.add(&x_times_five)?.add(&five)?;
+    let grads = y.backward()?;
+    let grad_x = grads.get(&x.id()).context("no grad for x")?;
+    assert_eq!(x.to_vec1::<f32>()?, [3., 1., 4.]);
+    // y = x^2 + 5.x + 5
+    assert_eq!(y.to_vec1::<f32>()?, [29., 11., 41.]);
+    // dy/dx = 2.x + 5
+    assert_eq!(grad_x.to_vec1::<f32>()?, [11., 7., 13.]);
+    Ok(())
+}