parameter.NormalizedParameter

parameter.NormalizedParameter(value)

Parameter rescaled by its own initial value, so its leaf starts at one.

The common case of RescaledParameter: the starting value is also the scale, which suits parameters that should explore a range comparable to where they begin. .value holds ones and .constrained_value returns scale * .value.

Prefer RescaledParameter with an explicit scale when the parameter starts near zero or has to change sign, since its own start is then far smaller than the range it needs to cover.

Parameters

Name	Type	Description	Default
value	Union[float, int, jnp.ndarray]	The original parameter value, used as 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)

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

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

Parameter rescaled by its own initial value, so its leaf starts at one.

The common case of `RescaledParameter`: the starting value is also the
scale, which suits parameters that should explore a range comparable to
where they begin. `.value` holds ones and `.constrained_value` returns
`scale * .value`.

Prefer `RescaledParameter` with an explicit scale when the parameter starts
near zero or has to change sign, since its own start is then far smaller
than the range it needs to cover.

## 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 as 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)
```