Final examples.

examples/fancy_indexing.rs

@@ -171,4 +171,48 @@ fn main() {
 
     // then sum reduce the result, and remove the remaining original dimension
     let_example!(sum_result, mul_result.sum(&[0]).squeeze(&Axes::Axis(0)));
+
+    // vix
+    // we want: t.vix(i, j)
+    let_example!(t, TrI::new(&[4, 6], &(1..25).collect::<Vec<_>>()));
+    let_example!(i, TrI::new(&[2], &[2, 0]));
+    let_example!(j, TrI::new(&[2], &[1, 0]));
+
+    // first convert the indexing tensor i to a one hot tensor i_one_hot
+    let_example!(
+        i_range,
+        TrI::new(&[4], (0..4).collect::<Vec<_>>().as_slice())
+    );
+    let_example!(i, i.reshape(&[2, 1]));
+    // now i_one_hot has shape [2, 4]
+    let_example!(i_one_hot, i.eq(&i_range).cast::<i32>());
+
+    // multiply the one hot tensor with the original tensor
+    // but first reshape so the right dimensions are broadcasted
+    let_example!(i_one_hot, i_one_hot.reshape(&[2, 4, 1]));
+    let_example!(mul_result, t.mul(&i_one_hot));
+
+    // then sum reduce the result, and remove the remaining original dimension
+    let_example!(t, mul_result.sum(&[1]).squeeze(&Axes::Axis(1)));
+
+    // then do the same for j
+    let_example!(
+        j_range,
+        TrI::new(&[6], (0..6).collect::<Vec<_>>().as_slice())
+    );
+    let_example!(j, j.reshape(&[2, 1]));
+    // j_one_hot has shape [2,6]...
+    let_example!(j_one_hot, j.eq(&j_range).cast::<i32>());
+
+    // the same shape as t! So we can just multiply...
+    let_example!(mul_result, t.mul(&j_one_hot));
+    // and sum:
+    let_example!(sum_result, mul_result.sum(&[1]).squeeze(&Axes::Axis(1)));
+
+    // masks
+    let_example!(
+        b,
+        TrB::new(&[2, 3], &[false, false, true, false, true, false])
+    );
+    let_example!(i_b, TrI::new(&[2], &[2, 4]));
 }