pycontrails 0.54.3__cp311-cp311-win_amd64.whl → 0.54.4__cp311-cp311-win_amd64.whl

pycontrails/models/cocip/cocip_uncertainty.py

@@ -2,8 +2,8 @@
 
 from __future__ import annotations
 
+import dataclasses
 import logging
-from dataclasses import asdict, dataclass
 from typing import Any, ClassVar
 
 import numpy as np
@@ -70,7 +70,7 @@ class habit_dirichlet(rv_frozen):
         return habit_weights
 
 
-@dataclass
+@dataclasses.dataclass
 class CocipUncertaintyParams(CocipParams):
     """Model parameters for CoCiP epistemic uncertainty.
 
@@ -117,26 +117,30 @@ class CocipUncertaintyParams(CocipParams):
     #: Schumann takes ``wind_shear_enhancement_exponent`` = 0.5 and discusses the case of 0 and 2/3
     #: as possibilities.
     #: With a value of 0, wind shear is not enhanced.
-    wind_shear_enhancement_exponent_uncertainty: rv_frozen | None = stats.triang(
-        loc=0.0, c=CocipParams.wind_shear_enhancement_exponent, scale=1.0
+    wind_shear_enhancement_exponent_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=lambda: stats.triang(
+            loc=0.0, c=CocipParams.wind_shear_enhancement_exponent, scale=1.0
+        )
     )
 
     #: Schumann takes ``initial_wake_vortex_depth`` = 0.5 and discusses some
     #: uncertainty in this value. This parameter should be non-negative.
-    initial_wake_vortex_depth_uncertainty: rv_frozen | None = stats.triang(
-        loc=0.3, c=CocipParams.initial_wake_vortex_depth, scale=0.4
+    initial_wake_vortex_depth_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=lambda: stats.triang(
+            loc=0.3, c=CocipParams.initial_wake_vortex_depth, scale=0.4
+        )
     )
 
     #: Schumann takes a default value of 0.1 and describes it as an "important adjustable parameter"
     #: Currently, `CocipParams` uses a default value of 0.5
-    sedimentation_impact_factor_uncertainty: rv_frozen | None = stats.norm(
-        loc=CocipParams.sedimentation_impact_factor, scale=0.1
+    sedimentation_impact_factor_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=lambda: stats.norm(loc=CocipParams.sedimentation_impact_factor, scale=0.1)
     )
 
     #: Teoh 2022 (to appear) takes values between 70% decrease and 100% increase.
     #: This coincides with the log normal distribution defined below.
-    nvpm_ei_n_enhancement_factor_uncertainty: rv_frozen | None = stats.lognorm(
-        s=0.15, scale=1 / stats.lognorm(s=0.15).mean()
+    nvpm_ei_n_enhancement_factor_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=lambda: stats.lognorm(s=0.15, scale=1 / stats.lognorm(s=0.15).mean())
     )
 
     #: Scale shortwave radiative forcing.
@@ -145,8 +149,8 @@ class CocipUncertaintyParams(CocipParams):
     #: by `libRadTran <http://www.libradtran.org/doku.php>`_.
     #: We use the average RMS error across all habit types (pg 1397) as the standard deviation
     #: of a normally distributed scaling factor for SW forcing
-    rf_sw_enhancement_factor_uncertainty: rv_frozen | None = stats.norm(
-        loc=CocipParams.rf_sw_enhancement_factor, scale=0.106
+    rf_sw_enhancement_factor_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=lambda: stats.norm(loc=CocipParams.rf_sw_enhancement_factor, scale=0.106)
     )
 
     #: Scale longwave radiative forcing.
@@ -154,8 +158,8 @@ class CocipUncertaintyParams(CocipParams):
     #: fit to the data generated by `libRadTran <http://www.libradtran.org/doku.php>`_.
     #: We use the average RMS error across all habit types (pg 1397) as the standard deviation
     #: of a normally distributed scaling factor for LW forcing.
-    rf_lw_enhancement_factor_uncertainty: rv_frozen | None = stats.norm(
-        loc=CocipParams.rf_lw_enhancement_factor, scale=0.071
+    rf_lw_enhancement_factor_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=lambda: stats.norm(loc=CocipParams.rf_lw_enhancement_factor, scale=0.071)
     )
 
     #: Scale the habit distributions by a dirichlet distribution
@@ -163,7 +167,9 @@ class CocipUncertaintyParams(CocipParams):
     #: where :math:`\text{G}_{i}` is the approximate habit weight distributions
     #: defined in :attr:`CocipParams().habit_distributions`.
     #: Higher values of :math:`\text{C}` correspond to higher confidence in initial estimates.
-    habit_distributions_uncertainty: rv_frozen | None = habit_dirichlet(C=96.0)
+    habit_distributions_uncertainty: rv_frozen | None = dataclasses.field(
+        default_factory=habit_dirichlet
+    )
 
     def __post_init__(self) -> None:
         """Override values of model parameters according to ranges."""
@@ -192,7 +198,7 @@ class CocipUncertaintyParams(CocipParams):
 
         out = {}
 
-        param_dict = asdict(self)
+        param_dict = dataclasses.asdict(self)
         for uncertainty_param, dist in param_dict.items():
             if uncertainty_param.endswith("_uncertainty") and dist is not None:
                 param = uncertainty_param.split("_uncertainty")[0]
@@ -212,7 +218,7 @@ class CocipUncertaintyParams(CocipParams):
 
         return out
 
-    def rvs(self, size: None | int = None) -> dict[str, np.float64 | npt.NDArray[np.float64]]:
+    def rvs(self, size: None | int = None) -> dict[str, np.float64 | npt.NDArray[np.floating]]:
         """Call each distribution's `rvs` method to generate random parameters.
 
         Seed calls to `rvs` with class variable `rng`.
@@ -224,7 +230,7 @@ class CocipUncertaintyParams(CocipParams):
 
         Returns
         -------
-        dict[str, float | npt.NDArray[np.float64]]
+        dict[str, float | npt.NDArray[np.floating]]
             Dictionary of random parameters. Dictionary keys consists of names of parameters in
             `CocipParams` to be overridden by random value.