pymc-extras 0.2.5__py3-none-any.whl → 0.2.7__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- pymc_extras/__init__.py +5 -1
- pymc_extras/deserialize.py +224 -0
- pymc_extras/distributions/continuous.py +3 -2
- pymc_extras/distributions/discrete.py +3 -1
- pymc_extras/inference/find_map.py +62 -17
- pymc_extras/inference/laplace.py +10 -7
- pymc_extras/prior.py +1356 -0
- pymc_extras/statespace/core/statespace.py +191 -52
- pymc_extras/statespace/filters/distributions.py +15 -16
- pymc_extras/statespace/filters/kalman_filter.py +1 -18
- pymc_extras/statespace/filters/kalman_smoother.py +2 -6
- pymc_extras/statespace/models/ETS.py +10 -0
- pymc_extras/statespace/models/SARIMAX.py +26 -5
- pymc_extras/statespace/models/VARMAX.py +12 -2
- pymc_extras/statespace/models/structural.py +18 -5
- pymc_extras-0.2.7.dist-info/METADATA +321 -0
- pymc_extras-0.2.7.dist-info/RECORD +66 -0
- {pymc_extras-0.2.5.dist-info → pymc_extras-0.2.7.dist-info}/WHEEL +1 -2
- pymc_extras/utils/pivoted_cholesky.py +0 -69
- pymc_extras/version.py +0 -11
- pymc_extras/version.txt +0 -1
- pymc_extras-0.2.5.dist-info/METADATA +0 -112
- pymc_extras-0.2.5.dist-info/RECORD +0 -108
- pymc_extras-0.2.5.dist-info/top_level.txt +0 -2
- tests/__init__.py +0 -13
- tests/distributions/__init__.py +0 -19
- tests/distributions/test_continuous.py +0 -185
- tests/distributions/test_discrete.py +0 -210
- tests/distributions/test_discrete_markov_chain.py +0 -258
- tests/distributions/test_multivariate.py +0 -304
- tests/distributions/test_transform.py +0 -77
- tests/model/__init__.py +0 -0
- tests/model/marginal/__init__.py +0 -0
- tests/model/marginal/test_distributions.py +0 -132
- tests/model/marginal/test_graph_analysis.py +0 -182
- tests/model/marginal/test_marginal_model.py +0 -967
- tests/model/test_model_api.py +0 -38
- tests/statespace/__init__.py +0 -0
- tests/statespace/test_ETS.py +0 -411
- tests/statespace/test_SARIMAX.py +0 -405
- tests/statespace/test_VARMAX.py +0 -184
- tests/statespace/test_coord_assignment.py +0 -181
- tests/statespace/test_distributions.py +0 -270
- tests/statespace/test_kalman_filter.py +0 -326
- tests/statespace/test_representation.py +0 -175
- tests/statespace/test_statespace.py +0 -872
- tests/statespace/test_statespace_JAX.py +0 -156
- tests/statespace/test_structural.py +0 -836
- tests/statespace/utilities/__init__.py +0 -0
- tests/statespace/utilities/shared_fixtures.py +0 -9
- tests/statespace/utilities/statsmodel_local_level.py +0 -42
- tests/statespace/utilities/test_helpers.py +0 -310
- tests/test_blackjax_smc.py +0 -222
- tests/test_find_map.py +0 -103
- tests/test_histogram_approximation.py +0 -109
- tests/test_laplace.py +0 -281
- tests/test_linearmodel.py +0 -208
- tests/test_model_builder.py +0 -306
- tests/test_pathfinder.py +0 -297
- tests/test_pivoted_cholesky.py +0 -24
- tests/test_printing.py +0 -98
- tests/test_prior_from_trace.py +0 -172
- tests/test_splines.py +0 -77
- tests/utils.py +0 -0
- {pymc_extras-0.2.5.dist-info → pymc_extras-0.2.7.dist-info}/licenses/LICENSE +0 -0
pymc_extras/__init__.py
CHANGED
|
@@ -13,6 +13,8 @@
|
|
|
13
13
|
# limitations under the License.
|
|
14
14
|
import logging
|
|
15
15
|
|
|
16
|
+
from importlib.metadata import version
|
|
17
|
+
|
|
16
18
|
from pymc_extras import gp, statespace, utils
|
|
17
19
|
from pymc_extras.distributions import *
|
|
18
20
|
from pymc_extras.inference import find_MAP, fit, fit_laplace, fit_pathfinder
|
|
@@ -22,7 +24,6 @@ from pymc_extras.model.marginal.marginal_model import (
|
|
|
22
24
|
recover_marginals,
|
|
23
25
|
)
|
|
24
26
|
from pymc_extras.model.model_api import as_model
|
|
25
|
-
from pymc_extras.version import __version__
|
|
26
27
|
|
|
27
28
|
_log = logging.getLogger("pmx")
|
|
28
29
|
|
|
@@ -31,3 +32,6 @@ if not logging.root.handlers:
|
|
|
31
32
|
if len(_log.handlers) == 0:
|
|
32
33
|
handler = logging.StreamHandler()
|
|
33
34
|
_log.addHandler(handler)
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
__version__ = version("pymc-extras")
|
|
@@ -0,0 +1,224 @@
|
|
|
1
|
+
"""Deserialize dictionaries into Python objects.
|
|
2
|
+
|
|
3
|
+
This is a two step process:
|
|
4
|
+
|
|
5
|
+
1. Determine if the data is of the correct type.
|
|
6
|
+
2. Deserialize the data into a python object.
|
|
7
|
+
|
|
8
|
+
Examples
|
|
9
|
+
--------
|
|
10
|
+
Make use of the already registered deserializers:
|
|
11
|
+
|
|
12
|
+
.. code-block:: python
|
|
13
|
+
|
|
14
|
+
from pymc_extras.deserialize import deserialize
|
|
15
|
+
|
|
16
|
+
prior_class_data = {
|
|
17
|
+
"dist": "Normal",
|
|
18
|
+
"kwargs": {"mu": 0, "sigma": 1}
|
|
19
|
+
}
|
|
20
|
+
prior = deserialize(prior_class_data)
|
|
21
|
+
# Prior("Normal", mu=0, sigma=1)
|
|
22
|
+
|
|
23
|
+
Register custom class deserialization:
|
|
24
|
+
|
|
25
|
+
.. code-block:: python
|
|
26
|
+
|
|
27
|
+
from pymc_extras.deserialize import register_deserialization
|
|
28
|
+
|
|
29
|
+
class MyClass:
|
|
30
|
+
def __init__(self, value: int):
|
|
31
|
+
self.value = value
|
|
32
|
+
|
|
33
|
+
def to_dict(self) -> dict:
|
|
34
|
+
# Example of what the to_dict method might look like.
|
|
35
|
+
return {"value": self.value}
|
|
36
|
+
|
|
37
|
+
register_deserialization(
|
|
38
|
+
is_type=lambda data: data.keys() == {"value"} and isinstance(data["value"], int),
|
|
39
|
+
deserialize=lambda data: MyClass(value=data["value"]),
|
|
40
|
+
)
|
|
41
|
+
|
|
42
|
+
Deserialize data into that custom class:
|
|
43
|
+
|
|
44
|
+
.. code-block:: python
|
|
45
|
+
|
|
46
|
+
from pymc_extras.deserialize import deserialize
|
|
47
|
+
|
|
48
|
+
data = {"value": 42}
|
|
49
|
+
obj = deserialize(data)
|
|
50
|
+
assert isinstance(obj, MyClass)
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
"""
|
|
54
|
+
|
|
55
|
+
from collections.abc import Callable
|
|
56
|
+
from dataclasses import dataclass
|
|
57
|
+
from typing import Any
|
|
58
|
+
|
|
59
|
+
IsType = Callable[[Any], bool]
|
|
60
|
+
Deserialize = Callable[[Any], Any]
|
|
61
|
+
|
|
62
|
+
|
|
63
|
+
@dataclass
|
|
64
|
+
class Deserializer:
|
|
65
|
+
"""Object to store information required for deserialization.
|
|
66
|
+
|
|
67
|
+
All deserializers should be stored via the :func:`register_deserialization` function
|
|
68
|
+
instead of creating this object directly.
|
|
69
|
+
|
|
70
|
+
Attributes
|
|
71
|
+
----------
|
|
72
|
+
is_type : IsType
|
|
73
|
+
Function to determine if the data is of the correct type.
|
|
74
|
+
deserialize : Deserialize
|
|
75
|
+
Function to deserialize the data.
|
|
76
|
+
|
|
77
|
+
Examples
|
|
78
|
+
--------
|
|
79
|
+
.. code-block:: python
|
|
80
|
+
|
|
81
|
+
from typing import Any
|
|
82
|
+
|
|
83
|
+
class MyClass:
|
|
84
|
+
def __init__(self, value: int):
|
|
85
|
+
self.value = value
|
|
86
|
+
|
|
87
|
+
from pymc_extras.deserialize import Deserializer
|
|
88
|
+
|
|
89
|
+
def is_type(data: Any) -> bool:
|
|
90
|
+
return data.keys() == {"value"} and isinstance(data["value"], int)
|
|
91
|
+
|
|
92
|
+
def deserialize(data: dict) -> MyClass:
|
|
93
|
+
return MyClass(value=data["value"])
|
|
94
|
+
|
|
95
|
+
deserialize_logic = Deserializer(is_type=is_type, deserialize=deserialize)
|
|
96
|
+
|
|
97
|
+
"""
|
|
98
|
+
|
|
99
|
+
is_type: IsType
|
|
100
|
+
deserialize: Deserialize
|
|
101
|
+
|
|
102
|
+
|
|
103
|
+
DESERIALIZERS: list[Deserializer] = []
|
|
104
|
+
|
|
105
|
+
|
|
106
|
+
class DeserializableError(Exception):
|
|
107
|
+
"""Error raised when data cannot be deserialized."""
|
|
108
|
+
|
|
109
|
+
def __init__(self, data: Any):
|
|
110
|
+
self.data = data
|
|
111
|
+
super().__init__(
|
|
112
|
+
f"Couldn't deserialize {data}. Use register_deserialization to add a deserialization mapping."
|
|
113
|
+
)
|
|
114
|
+
|
|
115
|
+
|
|
116
|
+
def deserialize(data: Any) -> Any:
|
|
117
|
+
"""Deserialize a dictionary into a Python object.
|
|
118
|
+
|
|
119
|
+
Use the :func:`register_deserialization` function to add custom deserializations.
|
|
120
|
+
|
|
121
|
+
Deserialization is a two step process due to the dynamic nature of the data:
|
|
122
|
+
|
|
123
|
+
1. Determine if the data is of the correct type.
|
|
124
|
+
2. Deserialize the data into a Python object.
|
|
125
|
+
|
|
126
|
+
Each registered deserialization is checked in order until one is found that can
|
|
127
|
+
deserialize the data. If no deserialization is found, a :class:`DeserializableError` is raised.
|
|
128
|
+
|
|
129
|
+
A :class:`DeserializableError` is raised when the data fails to be deserialized
|
|
130
|
+
by any of the registered deserializers.
|
|
131
|
+
|
|
132
|
+
Parameters
|
|
133
|
+
----------
|
|
134
|
+
data : Any
|
|
135
|
+
The data to deserialize.
|
|
136
|
+
|
|
137
|
+
Returns
|
|
138
|
+
-------
|
|
139
|
+
Any
|
|
140
|
+
The deserialized object.
|
|
141
|
+
|
|
142
|
+
Raises
|
|
143
|
+
------
|
|
144
|
+
DeserializableError
|
|
145
|
+
Raised when the data doesn't match any registered deserializations
|
|
146
|
+
or fails to be deserialized.
|
|
147
|
+
|
|
148
|
+
Examples
|
|
149
|
+
--------
|
|
150
|
+
Deserialize a :class:`pymc_extras.prior.Prior` object:
|
|
151
|
+
|
|
152
|
+
.. code-block:: python
|
|
153
|
+
|
|
154
|
+
from pymc_extras.deserialize import deserialize
|
|
155
|
+
|
|
156
|
+
data = {"dist": "Normal", "kwargs": {"mu": 0, "sigma": 1}}
|
|
157
|
+
prior = deserialize(data)
|
|
158
|
+
# Prior("Normal", mu=0, sigma=1)
|
|
159
|
+
|
|
160
|
+
"""
|
|
161
|
+
for mapping in DESERIALIZERS:
|
|
162
|
+
try:
|
|
163
|
+
is_type = mapping.is_type(data)
|
|
164
|
+
except Exception:
|
|
165
|
+
is_type = False
|
|
166
|
+
|
|
167
|
+
if not is_type:
|
|
168
|
+
continue
|
|
169
|
+
|
|
170
|
+
try:
|
|
171
|
+
return mapping.deserialize(data)
|
|
172
|
+
except Exception as e:
|
|
173
|
+
raise DeserializableError(data) from e
|
|
174
|
+
else:
|
|
175
|
+
raise DeserializableError(data)
|
|
176
|
+
|
|
177
|
+
|
|
178
|
+
def register_deserialization(is_type: IsType, deserialize: Deserialize) -> None:
|
|
179
|
+
"""Register an arbitrary deserialization.
|
|
180
|
+
|
|
181
|
+
Use the :func:`deserialize` function to then deserialize data using all registered
|
|
182
|
+
deserialize functions.
|
|
183
|
+
|
|
184
|
+
Parameters
|
|
185
|
+
----------
|
|
186
|
+
is_type : Callable[[Any], bool]
|
|
187
|
+
Function to determine if the data is of the correct type.
|
|
188
|
+
deserialize : Callable[[dict], Any]
|
|
189
|
+
Function to deserialize the data of that type.
|
|
190
|
+
|
|
191
|
+
Examples
|
|
192
|
+
--------
|
|
193
|
+
Register a custom class deserialization:
|
|
194
|
+
|
|
195
|
+
.. code-block:: python
|
|
196
|
+
|
|
197
|
+
from pymc_extras.deserialize import register_deserialization
|
|
198
|
+
|
|
199
|
+
class MyClass:
|
|
200
|
+
def __init__(self, value: int):
|
|
201
|
+
self.value = value
|
|
202
|
+
|
|
203
|
+
def to_dict(self) -> dict:
|
|
204
|
+
# Example of what the to_dict method might look like.
|
|
205
|
+
return {"value": self.value}
|
|
206
|
+
|
|
207
|
+
register_deserialization(
|
|
208
|
+
is_type=lambda data: data.keys() == {"value"} and isinstance(data["value"], int),
|
|
209
|
+
deserialize=lambda data: MyClass(value=data["value"]),
|
|
210
|
+
)
|
|
211
|
+
|
|
212
|
+
Use that custom class deserialization:
|
|
213
|
+
|
|
214
|
+
.. code-block:: python
|
|
215
|
+
|
|
216
|
+
from pymc_extras.deserialize import deserialize
|
|
217
|
+
|
|
218
|
+
data = {"value": 42}
|
|
219
|
+
obj = deserialize(data)
|
|
220
|
+
assert isinstance(obj, MyClass)
|
|
221
|
+
|
|
222
|
+
"""
|
|
223
|
+
mapping = Deserializer(is_type=is_type, deserialize=deserialize)
|
|
224
|
+
DESERIALIZERS.append(mapping)
|
|
@@ -81,7 +81,7 @@ class GenExtreme(Continuous):
|
|
|
81
81
|
|
|
82
82
|
\left\{x: 1 + \xi\left(\frac{x-\mu}{\sigma}\right) > 0 \right\}.
|
|
83
83
|
|
|
84
|
-
Note that this parametrization is per Coles (2001), and differs from that of
|
|
84
|
+
Note that this parametrization is per Coles (2001) [1]_, and differs from that of
|
|
85
85
|
Scipy in the sign of the shape parameter, :math:`\xi`.
|
|
86
86
|
|
|
87
87
|
.. plot::
|
|
@@ -132,7 +132,7 @@ class GenExtreme(Continuous):
|
|
|
132
132
|
|
|
133
133
|
References
|
|
134
134
|
----------
|
|
135
|
-
.. [
|
|
135
|
+
.. [1] Coles, S.G. (2001).
|
|
136
136
|
An Introduction to the Statistical Modeling of Extreme Values
|
|
137
137
|
Springer-Verlag, London
|
|
138
138
|
|
|
@@ -260,6 +260,7 @@ class Chi:
|
|
|
260
260
|
Examples
|
|
261
261
|
--------
|
|
262
262
|
.. code-block:: python
|
|
263
|
+
|
|
263
264
|
import pymc as pm
|
|
264
265
|
from pymc_extras.distributions import Chi
|
|
265
266
|
|
|
@@ -116,6 +116,7 @@ class GeneralizedPoisson(pm.distributions.Discrete):
|
|
|
116
116
|
|
|
117
117
|
.. math:: f(x \mid \mu, \lambda) =
|
|
118
118
|
\frac{\mu (\mu + \lambda x)^{x-1} e^{-\mu - \lambda x}}{x!}
|
|
119
|
+
|
|
119
120
|
======== ======================================
|
|
120
121
|
Support :math:`x \in \mathbb{N}_0`
|
|
121
122
|
Mean :math:`\frac{\mu}{1 - \lambda}`
|
|
@@ -135,9 +136,10 @@ class GeneralizedPoisson(pm.distributions.Discrete):
|
|
|
135
136
|
When lam < 0, the mean is greater than the variance (underdispersion).
|
|
136
137
|
When lam > 0, the mean is less than the variance (overdispersion).
|
|
137
138
|
|
|
139
|
+
The PMF is taken from [1]_ and the random generator function is adapted from [2]_.
|
|
140
|
+
|
|
138
141
|
References
|
|
139
142
|
----------
|
|
140
|
-
The PMF is taken from [1] and the random generator function is adapted from [2].
|
|
141
143
|
.. [1] Consul, PoC, and Felix Famoye. "Generalized Poisson regression model."
|
|
142
144
|
Communications in Statistics-Theory and Methods 21.1 (1992): 89-109.
|
|
143
145
|
.. [2] Famoye, Felix. "Generalized Poisson random variate generation." American
|
|
@@ -9,7 +9,7 @@ import pymc as pm
|
|
|
9
9
|
import pytensor
|
|
10
10
|
import pytensor.tensor as pt
|
|
11
11
|
|
|
12
|
-
from better_optimize import minimize
|
|
12
|
+
from better_optimize import basinhopping, minimize
|
|
13
13
|
from better_optimize.constants import MINIMIZE_MODE_KWARGS, minimize_method
|
|
14
14
|
from pymc.blocking import DictToArrayBijection, RaveledVars
|
|
15
15
|
from pymc.initial_point import make_initial_point_fn
|
|
@@ -146,7 +146,7 @@ def _compile_grad_and_hess_to_jax(
|
|
|
146
146
|
orig_loss_fn = f_loss.vm.jit_fn
|
|
147
147
|
|
|
148
148
|
@jax.jit
|
|
149
|
-
def loss_fn_jax_grad(x
|
|
149
|
+
def loss_fn_jax_grad(x):
|
|
150
150
|
return jax.value_and_grad(lambda x: orig_loss_fn(x)[0])(x)
|
|
151
151
|
|
|
152
152
|
f_loss_and_grad = loss_fn_jax_grad
|
|
@@ -301,6 +301,14 @@ def scipy_optimize_funcs_from_loss(
|
|
|
301
301
|
point=initial_point_dict, outputs=[loss], inputs=inputs
|
|
302
302
|
)
|
|
303
303
|
|
|
304
|
+
# If we use pytensor gradients, we will use the pytensor function wrapper that handles shared variables. When
|
|
305
|
+
# computing jax gradients, we discard the function wrapper, so we can't handle shared variables --> rewrite them
|
|
306
|
+
# away.
|
|
307
|
+
if use_jax_gradients:
|
|
308
|
+
from pymc.sampling.jax import _replace_shared_variables
|
|
309
|
+
|
|
310
|
+
[loss] = _replace_shared_variables([loss])
|
|
311
|
+
|
|
304
312
|
compute_grad = use_grad and not use_jax_gradients
|
|
305
313
|
compute_hess = use_hess and not use_jax_gradients
|
|
306
314
|
compute_hessp = use_hessp and not use_jax_gradients
|
|
@@ -327,7 +335,7 @@ def scipy_optimize_funcs_from_loss(
|
|
|
327
335
|
|
|
328
336
|
|
|
329
337
|
def find_MAP(
|
|
330
|
-
method: minimize_method,
|
|
338
|
+
method: minimize_method | Literal["basinhopping"],
|
|
331
339
|
*,
|
|
332
340
|
model: pm.Model | None = None,
|
|
333
341
|
use_grad: bool | None = None,
|
|
@@ -344,14 +352,17 @@ def find_MAP(
|
|
|
344
352
|
**optimizer_kwargs,
|
|
345
353
|
) -> dict[str, np.ndarray] | tuple[dict[str, np.ndarray], OptimizeResult]:
|
|
346
354
|
"""
|
|
347
|
-
Fit a PyMC model via maximum a posteriori (MAP) estimation using JAX and scipy.
|
|
355
|
+
Fit a PyMC model via maximum a posteriori (MAP) estimation using JAX and scipy.optimize.
|
|
348
356
|
|
|
349
357
|
Parameters
|
|
350
358
|
----------
|
|
351
359
|
model : pm.Model
|
|
352
360
|
The PyMC model to be fit. If None, the current model context is used.
|
|
353
361
|
method : str
|
|
354
|
-
The optimization method to use.
|
|
362
|
+
The optimization method to use. Valid choices are: Nelder-Mead, Powell, CG, BFGS, L-BFGS-B, TNC, SLSQP,
|
|
363
|
+
trust-constr, dogleg, trust-ncg, trust-exact, trust-krylov, and basinhopping.
|
|
364
|
+
|
|
365
|
+
See scipy.optimize.minimize documentation for details.
|
|
355
366
|
use_grad : bool | None, optional
|
|
356
367
|
Whether to use gradients in the optimization. Defaults to None, which determines this automatically based on
|
|
357
368
|
the ``method``.
|
|
@@ -379,7 +390,9 @@ def find_MAP(
|
|
|
379
390
|
compile_kwargs: dict, optional
|
|
380
391
|
Additional options to pass to the ``pytensor.function`` function when compiling loss functions.
|
|
381
392
|
**optimizer_kwargs
|
|
382
|
-
Additional keyword arguments to pass to the ``scipy.optimize
|
|
393
|
+
Additional keyword arguments to pass to the ``scipy.optimize`` function being used. Unless
|
|
394
|
+
``method = "basinhopping"``, ``scipy.optimize.minimize`` will be used. For ``basinhopping``,
|
|
395
|
+
``scipy.optimize.basinhopping`` will be used. See the documentation of these functions for details.
|
|
383
396
|
|
|
384
397
|
Returns
|
|
385
398
|
-------
|
|
@@ -405,6 +418,18 @@ def find_MAP(
|
|
|
405
418
|
initial_params = DictToArrayBijection.map(
|
|
406
419
|
{var_name: value for var_name, value in start_dict.items() if var_name in vars_dict}
|
|
407
420
|
)
|
|
421
|
+
|
|
422
|
+
do_basinhopping = method == "basinhopping"
|
|
423
|
+
minimizer_kwargs = optimizer_kwargs.pop("minimizer_kwargs", {})
|
|
424
|
+
|
|
425
|
+
if do_basinhopping:
|
|
426
|
+
# For a nice API, we let the user set method="basinhopping", but if we're doing basinhopping we still need
|
|
427
|
+
# another method for the inner optimizer. This will be set in the minimizer_kwargs, but also needs a default
|
|
428
|
+
# if one isn't provided.
|
|
429
|
+
|
|
430
|
+
method = minimizer_kwargs.pop("method", "L-BFGS-B")
|
|
431
|
+
minimizer_kwargs["method"] = method
|
|
432
|
+
|
|
408
433
|
use_grad, use_hess, use_hessp = set_optimizer_function_defaults(
|
|
409
434
|
method, use_grad, use_hess, use_hessp
|
|
410
435
|
)
|
|
@@ -423,17 +448,37 @@ def find_MAP(
|
|
|
423
448
|
args = optimizer_kwargs.pop("args", None)
|
|
424
449
|
|
|
425
450
|
# better_optimize.minimize will check if f_logp is a fused loss+grad Op, and automatically assign the jac argument
|
|
426
|
-
# if so. That is why
|
|
427
|
-
|
|
428
|
-
|
|
429
|
-
|
|
430
|
-
|
|
431
|
-
hess
|
|
432
|
-
|
|
433
|
-
|
|
434
|
-
|
|
435
|
-
|
|
436
|
-
|
|
451
|
+
# if so. That is why the jac argument is not passed here in either branch.
|
|
452
|
+
|
|
453
|
+
if do_basinhopping:
|
|
454
|
+
if "args" not in minimizer_kwargs:
|
|
455
|
+
minimizer_kwargs["args"] = args
|
|
456
|
+
if "hess" not in minimizer_kwargs:
|
|
457
|
+
minimizer_kwargs["hess"] = f_hess
|
|
458
|
+
if "hessp" not in minimizer_kwargs:
|
|
459
|
+
minimizer_kwargs["hessp"] = f_hessp
|
|
460
|
+
if "method" not in minimizer_kwargs:
|
|
461
|
+
minimizer_kwargs["method"] = method
|
|
462
|
+
|
|
463
|
+
optimizer_result = basinhopping(
|
|
464
|
+
func=f_logp,
|
|
465
|
+
x0=cast(np.ndarray[float], initial_params.data),
|
|
466
|
+
progressbar=progressbar,
|
|
467
|
+
minimizer_kwargs=minimizer_kwargs,
|
|
468
|
+
**optimizer_kwargs,
|
|
469
|
+
)
|
|
470
|
+
|
|
471
|
+
else:
|
|
472
|
+
optimizer_result = minimize(
|
|
473
|
+
f=f_logp,
|
|
474
|
+
x0=cast(np.ndarray[float], initial_params.data),
|
|
475
|
+
args=args,
|
|
476
|
+
hess=f_hess,
|
|
477
|
+
hessp=f_hessp,
|
|
478
|
+
progressbar=progressbar,
|
|
479
|
+
method=method,
|
|
480
|
+
**optimizer_kwargs,
|
|
481
|
+
)
|
|
437
482
|
|
|
438
483
|
raveled_optimized = RaveledVars(optimizer_result.x, initial_params.point_map_info)
|
|
439
484
|
unobserved_vars = get_default_varnames(model.unobserved_value_vars, include_transformed)
|
pymc_extras/inference/laplace.py
CHANGED
|
@@ -416,7 +416,7 @@ def sample_laplace_posterior(
|
|
|
416
416
|
|
|
417
417
|
|
|
418
418
|
def fit_laplace(
|
|
419
|
-
optimize_method: minimize_method = "BFGS",
|
|
419
|
+
optimize_method: minimize_method | Literal["basinhopping"] = "BFGS",
|
|
420
420
|
*,
|
|
421
421
|
model: pm.Model | None = None,
|
|
422
422
|
use_grad: bool | None = None,
|
|
@@ -449,8 +449,11 @@ def fit_laplace(
|
|
|
449
449
|
----------
|
|
450
450
|
model : pm.Model
|
|
451
451
|
The PyMC model to be fit. If None, the current model context is used.
|
|
452
|
-
|
|
453
|
-
The optimization method to use.
|
|
452
|
+
method : str
|
|
453
|
+
The optimization method to use. Valid choices are: Nelder-Mead, Powell, CG, BFGS, L-BFGS-B, TNC, SLSQP,
|
|
454
|
+
trust-constr, dogleg, trust-ncg, trust-exact, trust-krylov, and basinhopping.
|
|
455
|
+
|
|
456
|
+
See scipy.optimize.minimize documentation for details.
|
|
454
457
|
use_grad : bool | None, optional
|
|
455
458
|
Whether to use gradients in the optimization. Defaults to None, which determines this automatically based on
|
|
456
459
|
the ``method``.
|
|
@@ -500,10 +503,10 @@ def fit_laplace(
|
|
|
500
503
|
diag_jitter: float | None
|
|
501
504
|
A small value added to the diagonal of the inverse Hessian matrix to ensure it is positive semi-definite.
|
|
502
505
|
If None, no jitter is added. Default is 1e-8.
|
|
503
|
-
optimizer_kwargs
|
|
504
|
-
Additional keyword arguments to pass to scipy.
|
|
505
|
-
|
|
506
|
-
|
|
506
|
+
optimizer_kwargs
|
|
507
|
+
Additional keyword arguments to pass to the ``scipy.optimize`` function being used. Unless
|
|
508
|
+
``method = "basinhopping"``, ``scipy.optimize.minimize`` will be used. For ``basinhopping``,
|
|
509
|
+
``scipy.optimize.basinhopping`` will be used. See the documentation of these functions for details.
|
|
507
510
|
compile_kwargs: dict, optional
|
|
508
511
|
Additional keyword arguments to pass to pytensor.function.
|
|
509
512
|
|