[Feature Request] Slice tensors using "standard" indexing

[ss32]

Review Mojo's priorities

• I have read the roadmap and priorities and I believe this request falls within the priorities.

What is your request?

Slicing tensors to assign new variables is not intuitive right now. Both Golang and Python use the same indexing and another modern language like Mojo should as well, especially if the aim is to lower the barrier to entry for new users.

package main
 
import (
 "fmt"
)
 
func main() {
 v := []int{1,2,3,4}
 x := v[0:2]
 y := v[2:4]
 fmt.Println("Vector v:", v)
 fmt.Println("Slice x:", x)
 fmt.Println("Slice y:", y)
}

import numpy as np
v = np.array([1,2,3,4])
x = v[0:2]
y = v[2:4]
print(f'Vector v: {v}')
print(f'Slice x: {x}')
print(f'Slice y: {y}')

 
## I think this is the easiest way? Even the official docs show a
## similar method to assign a numpy array from a Tensor
## calling numpy_array.itemset((col, row), tensor[col, row])
 
var t = Tensor[DType.float64](
    (3, 1),
    -1,
    0,
    1,
    0
)
var x: Tensor[DType.float64] = Tensor[DType.float64](2, 1)
var y: Tensor[DType.float64] = Tensor[DType.float64](2, 1)
for i in range(4):
    if i < 2:
        x[i] = t[i, 0]
    else:
        y[i] = t[i, 0]
print(x)

$ mojo run test.mojo
Tensor([[-1.0],
[0.0]], dtype=float64, shape=2x1)

What is your motivation for this change?

Following the examples and documentation I see syntax that is very nearly Python and I would expect the Tensor type to behave like tensors in Python as well.

Any other details?

A first cut could be a middle ground between what's possible now and how Golang and Python achieve this where the user still has to preallocate the memory but is able to assign using slice indices

var t = Tensor[DType.float64](
    (3, 1),
    -1,
    0,
    1,
    0
)
var x: Tensor[DType.float64] = Tensor[DType.float64](2, 1)
x = t[Index(0,0:2)]

A logical next iteration might relieve the user of having to call Index

var t = Tensor[DType.float64](
    (3, 1),
    -1,
    0,
    1,
    0
)
var x: Tensor[DType.float64] = Tensor[DType.float64](2, 1)
x = t[0,0:2]

The holy grail would be something akin to Python's default behavior, Golang's :=, and C++'s auto where the memory allocation is handled on the fly.

var t = Tensor[DType.float64](
    (3, 1),
    -1,
    0,
    1,
    0
)

x = t[0,0:2]

[StandinKP]

Hey @JoeLoser any suggestion on how to proceed with this? I would like to start working on this