parameter.NormalizedParameter

parameter.NormalizedParameter(value)

Parameter that stores normalized values (ones) internally but presents scaled values (scale * ones) to the outside world.

This enables optimization with normalized coordinates where gradients have consistent magnitudes across different parameter scales, while still returning properly scaled values for computation.

Parameters

Name	Type	Description	Default
value	Union[float, int, jnp.ndarray]	The original parameter value used to compute the static scale.	required

Examples

>>> param = NormalizedParameter(jnp.array([2.0, 4.0, 6.0]))
>>> param.value  # Internal normalized storage (ones)
Array([1., 1., 1.], dtype=float32)
>>> param.__jax_array__()  # External scaled values (scale * ones)
Array([2., 4., 6.], dtype=float32)
>>> param.scale  # Static scale factor
Array([2., 4., 6.], dtype=float32)

Methods

Name	Description
tree_flatten	Flatten for JAX pytree registration.
tree_unflatten	Unflatten for JAX pytree registration.

tree_flatten

parameter.NormalizedParameter.tree_flatten()

Flatten for JAX pytree registration.

tree_unflatten

parameter.NormalizedParameter.tree_unflatten(aux_data, children)

Unflatten for JAX pytree registration.

# parameter.NormalizedParameter { #tvboptim.parameter.NormalizedParameter }

```python
parameter.NormalizedParameter(value)
```

Parameter that stores normalized values (ones) internally but presents
scaled values (scale * ones) to the outside world.

This enables optimization with normalized coordinates where gradients
have consistent magnitudes across different parameter scales, while
still returning properly scaled values for computation.

## Parameters {.doc-section .doc-section-parameters}

| Name   | Type                                                                                                         | Description                                                    | Default    |
|--------|--------------------------------------------------------------------------------------------------------------|----------------------------------------------------------------|------------|
| value  | [Union](`typing.Union`)\[[float](`float`), [int](`int`), [jnp](`jax.numpy`).[ndarray](`jax.numpy.ndarray`)\] | The original parameter value used to compute the static scale. | _required_ |

## Examples {.doc-section .doc-section-examples}

```python
>>> param = NormalizedParameter(jnp.array([2.0, 4.0, 6.0]))
>>> param.value  # Internal normalized storage (ones)
Array([1., 1., 1.], dtype=float32)
>>> param.__jax_array__()  # External scaled values (scale * ones)
Array([2., 4., 6.], dtype=float32)
>>> param.scale  # Static scale factor
Array([2., 4., 6.], dtype=float32)
```

## Methods

| Name | Description |
| --- | --- |
| [tree_flatten](#tvboptim.parameter.NormalizedParameter.tree_flatten) | Flatten for JAX pytree registration. |
| [tree_unflatten](#tvboptim.parameter.NormalizedParameter.tree_unflatten) | Unflatten for JAX pytree registration. |

### tree_flatten { #tvboptim.parameter.NormalizedParameter.tree_flatten }

```python
parameter.NormalizedParameter.tree_flatten()
```

Flatten for JAX pytree registration.

### tree_unflatten { #tvboptim.parameter.NormalizedParameter.tree_unflatten }

```python
parameter.NormalizedParameter.tree_unflatten(aux_data, children)
```

Unflatten for JAX pytree registration.