paramz

0.9.6 verified Mon Apr 27 auth: no python maintenance

paramz is a parameterization framework for Python, providing mutable, constrained parameters with optimization capabilities (e.g., scipy-based L-BFGS-B). Current version is 0.9.6; release cadence is irregular with no recent updates since 2020.

pip install paramz

Common errors

error AttributeError: 'Model' object has no attribute '_gradient_array_' ↓

cause The optimizer expects a _gradient_array_ method; this is missing if you didn't call super().__init__() or if parameter names are inconsistent.

fix

Ensure that Model.__init__() is called via super().__init__() and that all parameters are added before optimization.

error Param object has no attribute 'values' ↓

cause Param was not properly initialized or the object was overwritten.

fix

Check that Param() is called with correct arguments, e.g., Param('name', np.array([1.0])).

Warnings

deprecated paramz is in maintenance mode; consider switching to GPy's paramz (built-in) or Pyro's parameterization for active development. ↓

fix Migrate to GPyTorch or Pyro.

gotcha Param objects have their values stored as a NumPy array; setting values directly via assignment (e.g., param.values = new_array) may not trigger constraints properly. Use param[:] = new_array instead. ↓

fix Always assign using slice syntax: param[:] = new_values.

gotcha When using gradient-based optimization, ensure that the model's log_likelihood() and gradients() methods are implemented; otherwise the optimizer may fail silently. ↓

fix Define log_likelihood() returning the negative log-likelihood and optionally gradients() for custom gradient computation.

Imports

Param
```
from paramz import Param
```
Correct import for the main parameter class; note that paramz is not namespaced.
ParamArray
wrong
```
from paramz.core import ParamArray
```
correct
```
from paramz import ParamArray
```
ParamArray is directly exposed in the top-level package, not in core.
Model
wrong
```
from paramz.model import Model
```
correct
```
from paramz import Model
```
Model is a top-level class, not a submodule.

Quickstart

Create a model with a parameter and display its values.

from paramz import Param, Model
import numpy as np

class MyModel(Model):
    def __init__(self):
        super().__init__()
        self.param = Param('x', np.array([1.0, 2.0]))

m = MyModel()
print(m)
print(m.param.values)