NDArray Creation and Manipulation¶

This guide covers common NDArray creation and transformation workflows in NuMojo.

Imports used in examples¶

```/dev/null/imports.mojo#L1-3 import numojo as nm from numojo.prelude import *

---

## 1) Creating arrays

### Zeros, ones, full, empty

```/dev/null/create_basic.mojo#L1-12
import numojo as nm
from numojo.prelude import *

fn main() raises:
    var z = nm.zeros[f32](Shape(2, 3))
    var o = nm.ones[f32](Shape(2, 3))
    var f = nm.full[f32](Shape(2, 3), fill_value=2.5)
    var e = nm.empty[f32](Shape(2, 3))

    print(z)
    print(o)
    print(f)
    print(e)

Notes: - empty allocates memory without initializing values. - Prefer zeros/ones/full unless you immediately overwrite all values.

Ranges and spaces¶

```/dev/null/create_ranges.mojo#L1-11 import numojo as nm from numojo.prelude import *

fn main() raises: var a = nm.arangef32 var b = nm.linspacef32 var c = nm.logspacef32 var d = nm.geomspacef32

print(a)
print(b)
print(c)
print(d)

```

Construct from text / literals¶

/dev/null/create_from_text.mojo#L1-9 import numojo as nm from numojo.prelude import * fn main() raises: var a = nm.fromstring[f32]("[[1, 2, 3], [4, 5, 6]]") var b = nm.array[f32]("[[10, 20], [30, 40]]") print(a) print(b)

2) Inspecting array metadata¶

```/dev/null/inspect_metadata.mojo#L1-13 import numojo as nm from numojo.prelude import *

fn main() raises: var a = nm.arangef32.reshape(Shape(3, 4))

print("ndim:", nm.ndim(a))
print("shape:", nm.shape(a))
print("size:", nm.size(a))
print("dtype:", a.dtype)
print("is C contiguous:", a.is_c_contiguous())
print("strides:", a.strides)

```

3) Reshaping and layout transforms¶

reshape¶

/dev/null/reshape.mojo#L1-10 import numojo as nm from numojo.prelude import * fn main() raises: var a = nm.arange[f32](12) var b = nm.reshape(a, Shape(3, 4)) var c = nm.reshape(b, Shape(2, 2, 3)) print(a) print(b) print(c)

ravel (flatten view/copy style behavior depends on layout safety)¶

```/dev/null/ravel.mojo#L1-8 import numojo as nm from numojo.prelude import *

fn main() raises: var a = nm.arangef32.reshape(Shape(3, 4)) var flat = nm.ravel(a) print(flat)

### transpose

```/dev/null/transpose.mojo#L1-9
import numojo as nm
from numojo.prelude import *

fn main() raises:
    var a = nm.arange[f32](12).reshape(Shape(3, 4))
    var t = nm.transpose(a)

    print(a)
    print(t)

4) Broadcasting and shape expansion¶

broadcast_to¶

```/dev/null/broadcast_to.mojo#L1-10 import numojo as nm from numojo.prelude import *

fn main() raises: var row = nm.arangef32.reshape(Shape(1, 4)) var expanded = nm.broadcast_to(row, Shape(3, 4))

print("row:")
print(row)
print("expanded:")
print(expanded)

```

Scalar broadcasting in arithmetic¶

/dev/null/scalar_broadcast.mojo#L1-10 import numojo as nm from numojo.prelude import * fn main() raises: var a = nm.arange[f32](6).reshape(Shape(2, 3)) var b = nm.add(a, SIMD[f32, 1](10)) var c = nm.mul(a, SIMD[f32, 1](2)) print(b) print(c)

5) Common manipulation operations¶

flip¶

```/dev/null/flip.mojo#L1-9 import numojo as nm from numojo.prelude import *

fn main() raises: var a = nm.arangef32.reshape(Shape(3, 4)) var flipped = nm.flip(a)

print(a)
print(flipped)

```

Basic slicing and subarray extraction¶

/dev/null/slicing_extract.mojo#L1-12 import numojo as nm from numojo.prelude import * fn main() raises: var a = nm.arange[f32](20).reshape(Shape(4, 5)) var row0 = a[0, :] var col2 = a[:, 2] var block = a[1:3, 1:4] var step = a[::2, ::2] print(row0) print(col2) print(block) print(step)

6) Updating values after creation¶

Set by index¶

```/dev/null/set_by_index.mojo#L1-10 import numojo as nm from numojo.prelude import *

fn main() raises: var a = nm.zerosf32 a[Item(1, 1)] = 9 a[Item(0, 2)] = 5

print(a)

```

Set by slice¶

/dev/null/set_by_slice.mojo#L1-10 import numojo as nm from numojo.prelude import * fn main() raises: var a = nm.zeros[f32](Shape(4, 4)) var patch = nm.ones[f32](Shape(2, 2)) a[1:3, 1:3] = patch print(a)

7) Patterns for real projects¶

Pattern A: initialize -> transform -> reduce¶

```/dev/null/pattern_a.mojo#L1-12 import numojo as nm from numojo.prelude import *

fn main() raises: var x = nm.linspacef32 var y = nm.sin(x) var y2 = nm.mul(y, y)

print("mean(y^2):", nm.mean(y2))
print("max(y):", nm.max(y))
print("min(y):", nm.min(y))

```

Pattern B: batch-friendly shaping¶

/dev/null/pattern_b.mojo#L1-12 import numojo as nm from numojo.prelude import * fn main() raises: var raw = nm.arange[f32](2 * 3 * 4) var batch = nm.reshape(raw, Shape(2, 3, 4)) # (batch, rows, cols) var flattened_batch0 = nm.ravel(batch[0, :, :]) print("batch shape:", batch.shape) print("batch[0] flattened:") print(flattened_batch0)

8) Pitfalls and best practices¶

Prefer explicit Shape(...) for readability.
Use consistent dtype in a workflow (f32 vs f64) to avoid accidental casts.
Validate axis values before reductions/manipulations in library code.
Use zeros/ones/full over empty unless performance profiling proves otherwise.
Keep transformations composable:
create → reshape → compute → reduce.

docs/user-guide/indexing.md
docs/user-guide/linalg.md
docs/user-guide/io.md
docs/getting-started/ndarray-vs-matrix.md