зеркало из
https://github.com/docxology/cognitive.git
synced 2025-10-29 12:16:04 +02:00
Updates
Этот коммит содержится в:
Daniel Ari Friedman
родитель
ccdeafd068
Коммит
b4279401c2
41
.obsidian/workspace.json
поставляемый
41
.obsidian/workspace.json
поставляемый
@ -163,26 +163,29 @@
|
||||
},
|
||||
"active": "0ba8a1b9b9dd949a",
|
||||
"lastOpenFiles": [
|
||||
"Things/Simple_POMDP/Output/efe_components_detailed.png",
|
||||
"Things/Simple_POMDP/Output/policy_evolution.png",
|
||||
"Things/Simple_POMDP/Output/free_energies.png",
|
||||
"Things/Simple_POMDP/Output/belief_history.png",
|
||||
"Things/Simple_POMDP/Output/action_history.png",
|
||||
"Things/Simple_POMDP/Output/observation_likelihood.png",
|
||||
"Things/Simple_POMDP/Output/state_transitions.png",
|
||||
"Things/Simple_POMDP/Output/belief_evolution.png",
|
||||
"Things/Simple_POMDP/Output/simulation_log.txt",
|
||||
"Things/Simple_POMDP/Output/all_matrices.png",
|
||||
"Things/Simple_POMDP/Output/C_preferences.png",
|
||||
"Things/Simple_POMDP/Output/B_matrices.png",
|
||||
"Things/Simple_POMDP/Output/matrix_validation.txt",
|
||||
"Things/Simple_POMDP/Output",
|
||||
"src/models/active_inference/docs/dispatcher.md",
|
||||
"src/models/active_inference/docs/active_inference_dispatcher.md",
|
||||
"src/models/active_inference/docs/active_inference.md",
|
||||
"src/models/active_inference/docs",
|
||||
"src/models/active_inference/config_examples/continuous_sampling.yaml",
|
||||
"src/models/active_inference/config_examples/discrete_variational.yaml",
|
||||
"src/models/active_inference/config_examples",
|
||||
"src/models/active_inference/dispatcher.py",
|
||||
"Things/Continuous_Generic/__pycache__/test_continuous_generic.cpython-310-pytest-8.3.2.pyc",
|
||||
"Things/Continuous_Generic/__pycache__/conftest.cpython-310-pytest-8.3.2.pyc",
|
||||
"Things/Continuous_Generic/Output/tests/single_step",
|
||||
"Things/Continuous_Generic/Output/tests/multi_step",
|
||||
"Things/Continuous_Generic/Output/tests/complex",
|
||||
"Things/Continuous_Generic/Output/tests/basic",
|
||||
"Things/Continuous_Generic/Output/tests",
|
||||
"Things/Continuous_Generic/Output",
|
||||
"Things/Continuous_Generic/__pycache__/visualization.cpython-310.pyc",
|
||||
"Things/Continuous_Generic/__pycache__/continuous_generic.cpython-310.pyc",
|
||||
"Things/Continuous_Generic/Output/tests/complex/energy_conservation/energy_evolution.gif",
|
||||
"Things/Continuous_Generic/Output/tests/complex/energy_conservation/energy_analysis.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/taylor_prediction/prediction_analysis.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/generalized_coordinates/error_analysis.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/generalized_coordinates/derivative_analysis.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/generalized_coordinates/generalized_coordinates.gif",
|
||||
"Things/Continuous_Generic/Output/tests/complex/driven_oscillator/prediction_analysis.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/driven_oscillator/state_correlations.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/driven_oscillator/time_evolution.png",
|
||||
"Things/Continuous_Generic/Output/tests/complex/driven_oscillator/phase_space.png"
|
||||
"Things/Continuous_Generic/__pycache__/conftest.cpython-310-pytest-8.3.2.pyc"
|
||||
]
|
||||
}
|
||||
Двоичные данные
Things/Simple_POMDP/Output/A_matrix.png
Двоичные данные
Things/Simple_POMDP/Output/A_matrix.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 26 KiB После Ширина: | Высота: | Размер: 26 KiB |
Двоичные данные
Things/Simple_POMDP/Output/B_matrices.png
Двоичные данные
Things/Simple_POMDP/Output/B_matrices.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 53 KiB После Ширина: | Высота: | Размер: 50 KiB |
Двоичные данные
Things/Simple_POMDP/Output/C_matrix.png
Двоичные данные
Things/Simple_POMDP/Output/C_matrix.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 22 KiB |
Двоичные данные
Things/Simple_POMDP/Output/action_history.png
Двоичные данные
Things/Simple_POMDP/Output/action_history.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 21 KiB После Ширина: | Высота: | Размер: 21 KiB |
Двоичные данные
Things/Simple_POMDP/Output/all_matrices.png
Двоичные данные
Things/Simple_POMDP/Output/all_matrices.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 200 KiB После Ширина: | Высота: | Размер: 196 KiB |
Двоичные данные
Things/Simple_POMDP/Output/belief_evolution.png
Двоичные данные
Things/Simple_POMDP/Output/belief_evolution.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 92 KiB После Ширина: | Высота: | Размер: 86 KiB |
Двоичные данные
Things/Simple_POMDP/Output/belief_history.png
Двоичные данные
Things/Simple_POMDP/Output/belief_history.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 24 KiB После Ширина: | Высота: | Размер: 22 KiB |
Двоичные данные
Things/Simple_POMDP/Output/efe_components_detailed.png
Двоичные данные
Things/Simple_POMDP/Output/efe_components_detailed.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 190 KiB После Ширина: | Высота: | Размер: 208 KiB |
Двоичные данные
Things/Simple_POMDP/Output/free_energies.png
Двоичные данные
Things/Simple_POMDP/Output/free_energies.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 102 KiB После Ширина: | Высота: | Размер: 116 KiB |
@ -7,15 +7,15 @@ A Matrix (Observation Model):
|
||||
- Non-negative: True
|
||||
|
||||
Observation probabilities:
|
||||
P(Low|Low) = 0.092
|
||||
P(Low|Medium) = 0.219
|
||||
P(Low|High) = 0.423
|
||||
P(Medium|Low) = 0.270
|
||||
P(Medium|Medium) = 0.375
|
||||
P(Medium|High) = 0.405
|
||||
P(High|Low) = 0.638
|
||||
P(High|Medium) = 0.406
|
||||
P(High|High) = 0.173
|
||||
P(Low|Low) = 0.202
|
||||
P(Low|Medium) = 0.003
|
||||
P(Low|High) = 0.411
|
||||
P(Medium|Low) = 0.582
|
||||
P(Medium|Medium) = 0.714
|
||||
P(Medium|High) = 0.186
|
||||
P(High|Low) = 0.216
|
||||
P(High|Medium) = 0.283
|
||||
P(High|High) = 0.403
|
||||
|
||||
B Matrix (Transition Model):
|
||||
- Shape: (3, 3, 3)
|
||||
@ -25,45 +25,45 @@ Action: Decrease
|
||||
- Non-negative: True
|
||||
|
||||
Transition probabilities:
|
||||
P(Low|Low,Decrease) = 0.172
|
||||
P(Low|Medium,Decrease) = 0.278
|
||||
P(Low|High,Decrease) = 0.042
|
||||
P(Medium|Low,Decrease) = 0.158
|
||||
P(Medium|Medium,Decrease) = 0.296
|
||||
P(Medium|High,Decrease) = 0.157
|
||||
P(High|Low,Decrease) = 0.670
|
||||
P(High|Medium,Decrease) = 0.426
|
||||
P(High|High,Decrease) = 0.801
|
||||
P(Low|Low,Decrease) = 0.324
|
||||
P(Low|Medium,Decrease) = 0.186
|
||||
P(Low|High,Decrease) = 0.012
|
||||
P(Medium|Low,Decrease) = 0.320
|
||||
P(Medium|Medium,Decrease) = 0.162
|
||||
P(Medium|High,Decrease) = 0.722
|
||||
P(High|Low,Decrease) = 0.356
|
||||
P(High|Medium,Decrease) = 0.651
|
||||
P(High|High,Decrease) = 0.266
|
||||
|
||||
Action: Stay
|
||||
- Column stochastic: True
|
||||
- Non-negative: True
|
||||
|
||||
Transition probabilities:
|
||||
P(Low|Low,Stay) = 0.413
|
||||
P(Low|Medium,Stay) = 0.319
|
||||
P(Low|High,Stay) = 0.453
|
||||
P(Medium|Low,Stay) = 0.048
|
||||
P(Medium|Medium,Stay) = 0.660
|
||||
P(Medium|High,Stay) = 0.462
|
||||
P(High|Low,Stay) = 0.539
|
||||
P(High|Medium,Stay) = 0.021
|
||||
P(High|High,Stay) = 0.085
|
||||
P(Low|Low,Stay) = 0.423
|
||||
P(Low|Medium,Stay) = 0.005
|
||||
P(Low|High,Stay) = 0.182
|
||||
P(Medium|Low,Stay) = 0.257
|
||||
P(Medium|Medium,Stay) = 0.738
|
||||
P(Medium|High,Stay) = 0.719
|
||||
P(High|Low,Stay) = 0.320
|
||||
P(High|Medium,Stay) = 0.257
|
||||
P(High|High,Stay) = 0.099
|
||||
|
||||
Action: Increase
|
||||
- Column stochastic: True
|
||||
- Non-negative: True
|
||||
|
||||
Transition probabilities:
|
||||
P(Low|Low,Increase) = 0.272
|
||||
P(Low|Medium,Increase) = 0.423
|
||||
P(Low|High,Increase) = 0.258
|
||||
P(Medium|Low,Increase) = 0.264
|
||||
P(Medium|Medium,Increase) = 0.401
|
||||
P(Medium|High,Increase) = 0.364
|
||||
P(High|Low,Increase) = 0.464
|
||||
P(High|Medium,Increase) = 0.176
|
||||
P(High|High,Increase) = 0.377
|
||||
P(Low|Low,Increase) = 0.320
|
||||
P(Low|Medium,Increase) = 0.367
|
||||
P(Low|High,Increase) = 0.486
|
||||
P(Medium|Low,Increase) = 0.241
|
||||
P(Medium|Medium,Increase) = 0.364
|
||||
P(Medium|High,Increase) = 0.264
|
||||
P(High|Low,Increase) = 0.438
|
||||
P(High|Medium,Increase) = 0.269
|
||||
P(High|High,Increase) = 0.250
|
||||
|
||||
C Matrix (Log Preferences over Observations):
|
||||
- Shape: (3,)
|
||||
|
||||
Двоичные данные
Things/Simple_POMDP/Output/observation_likelihood.png
Двоичные данные
Things/Simple_POMDP/Output/observation_likelihood.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 17 KiB После Ширина: | Высота: | Размер: 17 KiB |
Двоичные данные
Things/Simple_POMDP/Output/policy_distribution.png
Двоичные данные
Things/Simple_POMDP/Output/policy_distribution.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 20 KiB |
Двоичные данные
Things/Simple_POMDP/Output/policy_evolution.png
Двоичные данные
Things/Simple_POMDP/Output/policy_evolution.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 47 KiB После Ширина: | Высота: | Размер: 53 KiB |
@ -3,321 +3,321 @@ Simulation Log
|
||||
|
||||
|
||||
Step 1:
|
||||
State: Low
|
||||
Observation: High
|
||||
Action: Increase
|
||||
Variational FE: 0.914
|
||||
State: High
|
||||
Observation: Low
|
||||
Action: Decrease
|
||||
Variational FE: 1.557
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.095
|
||||
Stay: 0.319
|
||||
Increase: 0.330
|
||||
Decrease: 0.047
|
||||
Stay: -0.200
|
||||
Increase: 0.052
|
||||
|
||||
Beliefs:
|
||||
Low: 0.420
|
||||
Medium: 0.340
|
||||
High: 0.240
|
||||
Low: 0.250
|
||||
Medium: 0.170
|
||||
High: 0.580
|
||||
|
||||
Step 2:
|
||||
State: Low
|
||||
Observation: High
|
||||
Action: Decrease
|
||||
Variational FE: 1.194
|
||||
State: High
|
||||
Observation: Low
|
||||
Action: Increase
|
||||
Variational FE: 1.561
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.116
|
||||
Stay: 0.334
|
||||
Increase: 0.330
|
||||
Decrease: -0.112
|
||||
Stay: -0.280
|
||||
Increase: 0.031
|
||||
|
||||
Beliefs:
|
||||
Low: 0.396
|
||||
Medium: 0.307
|
||||
High: 0.296
|
||||
Low: 0.329
|
||||
Medium: 0.087
|
||||
High: 0.584
|
||||
|
||||
Step 3:
|
||||
State: Low
|
||||
State: Medium
|
||||
Observation: Medium
|
||||
Action: Stay
|
||||
Variational FE: 1.077
|
||||
Action: Increase
|
||||
Variational FE: 0.708
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.094
|
||||
Stay: 0.332
|
||||
Increase: 0.328
|
||||
Decrease: -0.131
|
||||
Stay: -0.230
|
||||
Increase: 0.043
|
||||
|
||||
Beliefs:
|
||||
Low: 0.355
|
||||
Medium: 0.351
|
||||
High: 0.294
|
||||
Low: 0.413
|
||||
Medium: 0.235
|
||||
High: 0.351
|
||||
|
||||
Step 4:
|
||||
State: High
|
||||
Observation: Medium
|
||||
Action: Stay
|
||||
Variational FE: 1.078
|
||||
Action: Decrease
|
||||
Variational FE: 0.729
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.109
|
||||
Stay: 0.323
|
||||
Increase: 0.331
|
||||
Decrease: 0.025
|
||||
Stay: -0.153
|
||||
Increase: 0.064
|
||||
|
||||
Beliefs:
|
||||
Low: 0.333
|
||||
Medium: 0.388
|
||||
High: 0.280
|
||||
Low: 0.317
|
||||
Medium: 0.432
|
||||
High: 0.252
|
||||
|
||||
Step 5:
|
||||
State: Medium
|
||||
Observation: Medium
|
||||
Action: Increase
|
||||
Variational FE: 1.049
|
||||
Variational FE: 0.702
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.123
|
||||
Stay: 0.315
|
||||
Increase: 0.333
|
||||
Decrease: 0.090
|
||||
Stay: -0.206
|
||||
Increase: 0.051
|
||||
|
||||
Beliefs:
|
||||
Low: 0.292
|
||||
Medium: 0.379
|
||||
High: 0.329
|
||||
Low: 0.383
|
||||
Medium: 0.432
|
||||
High: 0.186
|
||||
|
||||
Step 6:
|
||||
State: Medium
|
||||
Observation: Medium
|
||||
Action: Increase
|
||||
Variational FE: 1.048
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.111
|
||||
Stay: 0.304
|
||||
Increase: 0.332
|
||||
|
||||
Beliefs:
|
||||
Low: 0.271
|
||||
Medium: 0.376
|
||||
High: 0.353
|
||||
|
||||
Step 7:
|
||||
State: Medium
|
||||
Observation: Low
|
||||
Action: Stay
|
||||
Variational FE: 1.577
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.105
|
||||
Stay: 0.298
|
||||
Increase: 0.332
|
||||
|
||||
Beliefs:
|
||||
Low: 0.223
|
||||
Medium: 0.412
|
||||
High: 0.365
|
||||
|
||||
Step 8:
|
||||
State: Medium
|
||||
Observation: High
|
||||
Action: Stay
|
||||
Variational FE: 0.778
|
||||
Variational FE: 1.204
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.114
|
||||
Stay: 0.276
|
||||
Increase: 0.333
|
||||
Decrease: 0.110
|
||||
Stay: -0.158
|
||||
Increase: 0.062
|
||||
|
||||
Beliefs:
|
||||
Low: 0.382
|
||||
Medium: 0.406
|
||||
High: 0.212
|
||||
Low: 0.263
|
||||
Medium: 0.475
|
||||
High: 0.262
|
||||
|
||||
Step 7:
|
||||
State: High
|
||||
Observation: Low
|
||||
Action: Decrease
|
||||
Variational FE: 1.465
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.090
|
||||
Stay: -0.247
|
||||
Increase: 0.042
|
||||
|
||||
Beliefs:
|
||||
Low: 0.209
|
||||
Medium: 0.240
|
||||
High: 0.551
|
||||
|
||||
Step 8:
|
||||
State: Low
|
||||
Observation: Low
|
||||
Action: Stay
|
||||
Variational FE: 2.159
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: -0.082
|
||||
Stay: -0.306
|
||||
Increase: 0.025
|
||||
|
||||
Beliefs:
|
||||
Low: 0.270
|
||||
Medium: 0.128
|
||||
High: 0.601
|
||||
|
||||
Step 9:
|
||||
State: High
|
||||
Observation: Low
|
||||
Action: Decrease
|
||||
Variational FE: 1.149
|
||||
Action: Increase
|
||||
Variational FE: 1.554
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.142
|
||||
Stay: 0.325
|
||||
Increase: 0.334
|
||||
Decrease: -0.134
|
||||
Stay: -0.269
|
||||
Increase: 0.034
|
||||
|
||||
Beliefs:
|
||||
Low: 0.218
|
||||
Medium: 0.277
|
||||
High: 0.505
|
||||
Low: 0.338
|
||||
Medium: 0.066
|
||||
High: 0.595
|
||||
|
||||
Step 10:
|
||||
State: Medium
|
||||
Observation: High
|
||||
Action: Stay
|
||||
Variational FE: 0.772
|
||||
Variational FE: 1.244
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.036
|
||||
Stay: 0.287
|
||||
Increase: 0.327
|
||||
Decrease: -0.143
|
||||
Stay: -0.226
|
||||
Increase: 0.044
|
||||
|
||||
Beliefs:
|
||||
Low: 0.390
|
||||
Medium: 0.326
|
||||
High: 0.284
|
||||
Low: 0.264
|
||||
Medium: 0.310
|
||||
High: 0.426
|
||||
|
||||
Step 11:
|
||||
State: Medium
|
||||
State: High
|
||||
Observation: High
|
||||
Action: Stay
|
||||
Variational FE: 0.818
|
||||
Action: Increase
|
||||
Variational FE: 1.229
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.103
|
||||
Stay: 0.331
|
||||
Increase: 0.330
|
||||
Decrease: 0.000
|
||||
Stay: -0.255
|
||||
Increase: 0.039
|
||||
|
||||
Beliefs:
|
||||
Low: 0.480
|
||||
Medium: 0.331
|
||||
High: 0.189
|
||||
Low: 0.282
|
||||
Medium: 0.295
|
||||
High: 0.424
|
||||
|
||||
Step 12:
|
||||
State: Low
|
||||
Observation: Medium
|
||||
Action: Increase
|
||||
Variational FE: 1.045
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.123
|
||||
Stay: 0.339
|
||||
Increase: 0.329
|
||||
|
||||
Beliefs:
|
||||
Low: 0.362
|
||||
Medium: 0.340
|
||||
High: 0.297
|
||||
|
||||
Step 13:
|
||||
State: High
|
||||
Observation: Medium
|
||||
Action: Increase
|
||||
Variational FE: 1.046
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.105
|
||||
Stay: 0.325
|
||||
Increase: 0.330
|
||||
|
||||
Beliefs:
|
||||
Low: 0.304
|
||||
Medium: 0.352
|
||||
High: 0.345
|
||||
|
||||
Step 14:
|
||||
State: High
|
||||
Observation: Low
|
||||
Action: Increase
|
||||
Variational FE: 1.403
|
||||
Variational FE: 1.564
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.099
|
||||
Stay: 0.310
|
||||
Increase: 0.331
|
||||
Decrease: -0.000
|
||||
Stay: -0.242
|
||||
Increase: 0.042
|
||||
|
||||
Beliefs:
|
||||
Low: 0.212
|
||||
Medium: 0.330
|
||||
High: 0.458
|
||||
Low: 0.336
|
||||
Medium: 0.149
|
||||
High: 0.515
|
||||
|
||||
Step 13:
|
||||
State: Medium
|
||||
Observation: Medium
|
||||
Action: Stay
|
||||
Variational FE: 0.547
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: -0.075
|
||||
Stay: -0.217
|
||||
Increase: 0.047
|
||||
|
||||
Beliefs:
|
||||
Low: 0.287
|
||||
Medium: 0.424
|
||||
High: 0.289
|
||||
|
||||
Step 14:
|
||||
State: High
|
||||
Observation: Medium
|
||||
Action: Increase
|
||||
Variational FE: 0.697
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.075
|
||||
Stay: -0.229
|
||||
Increase: 0.046
|
||||
|
||||
Beliefs:
|
||||
Low: 0.370
|
||||
Medium: 0.427
|
||||
High: 0.203
|
||||
|
||||
Step 15:
|
||||
State: Medium
|
||||
Observation: High
|
||||
Action: Increase
|
||||
Variational FE: 0.910
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.066
|
||||
Stay: 0.280
|
||||
Increase: 0.330
|
||||
|
||||
Beliefs:
|
||||
Low: 0.356
|
||||
Medium: 0.344
|
||||
High: 0.300
|
||||
|
||||
Step 16:
|
||||
State: Low
|
||||
Observation: High
|
||||
Action: Increase
|
||||
Variational FE: 0.913
|
||||
Action: Stay
|
||||
Variational FE: 1.205
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.105
|
||||
Stay: 0.324
|
||||
Increase: 0.331
|
||||
Stay: -0.167
|
||||
Increase: 0.060
|
||||
|
||||
Beliefs:
|
||||
Low: 0.433
|
||||
Medium: 0.345
|
||||
High: 0.223
|
||||
Low: 0.256
|
||||
Medium: 0.476
|
||||
High: 0.268
|
||||
|
||||
Step 16:
|
||||
State: Low
|
||||
Observation: Medium
|
||||
Action: Increase
|
||||
Variational FE: 0.692
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.088
|
||||
Stay: -0.252
|
||||
Increase: 0.041
|
||||
|
||||
Beliefs:
|
||||
Low: 0.353
|
||||
Medium: 0.456
|
||||
High: 0.191
|
||||
|
||||
Step 17:
|
||||
State: High
|
||||
Observation: Medium
|
||||
Action: Stay
|
||||
Variational FE: 1.074
|
||||
Observation: High
|
||||
Action: Increase
|
||||
Variational FE: 1.214
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.121
|
||||
Stay: 0.336
|
||||
Increase: 0.330
|
||||
Decrease: 0.111
|
||||
Stay: -0.178
|
||||
Increase: 0.058
|
||||
|
||||
Beliefs:
|
||||
Low: 0.370
|
||||
Medium: 0.365
|
||||
High: 0.265
|
||||
Low: 0.312
|
||||
Medium: 0.371
|
||||
High: 0.316
|
||||
|
||||
Step 18:
|
||||
State: High
|
||||
Observation: Low
|
||||
Action: Decrease
|
||||
Variational FE: 1.134
|
||||
Action: Increase
|
||||
Variational FE: 1.563
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.119
|
||||
Stay: 0.325
|
||||
Increase: 0.332
|
||||
Decrease: 0.059
|
||||
Stay: -0.213
|
||||
Increase: 0.049
|
||||
|
||||
Beliefs:
|
||||
Low: 0.210
|
||||
Medium: 0.253
|
||||
High: 0.537
|
||||
Low: 0.344
|
||||
Medium: 0.188
|
||||
High: 0.468
|
||||
|
||||
Step 19:
|
||||
State: Medium
|
||||
State: Low
|
||||
Observation: Medium
|
||||
Action: Stay
|
||||
Variational FE: 1.089
|
||||
Variational FE: 0.553
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.019
|
||||
Stay: 0.286
|
||||
Increase: 0.325
|
||||
Decrease: -0.041
|
||||
Stay: -0.206
|
||||
Increase: 0.050
|
||||
|
||||
Beliefs:
|
||||
Low: 0.270
|
||||
Medium: 0.363
|
||||
High: 0.367
|
||||
Low: 0.289
|
||||
Medium: 0.443
|
||||
High: 0.267
|
||||
|
||||
Step 20:
|
||||
State: High
|
||||
Observation: High
|
||||
Action: Decrease
|
||||
Variational FE: 1.214
|
||||
State: Low
|
||||
Observation: Medium
|
||||
Action: Stay
|
||||
Variational FE: 0.565
|
||||
|
||||
Expected Free Energies:
|
||||
Decrease: 0.098
|
||||
Stay: 0.298
|
||||
Increase: 0.331
|
||||
Decrease: 0.085
|
||||
Stay: -0.227
|
||||
Increase: 0.047
|
||||
|
||||
Beliefs:
|
||||
Low: 0.310
|
||||
Medium: 0.324
|
||||
High: 0.367
|
||||
Low: 0.233
|
||||
Medium: 0.595
|
||||
High: 0.172
|
||||
|
||||
Двоичные данные
Things/Simple_POMDP/Output/state_transitions.png
Двоичные данные
Things/Simple_POMDP/Output/state_transitions.png
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 36 KiB После Ширина: | Высота: | Размер: 34 KiB |
@ -2,4 +2,21 @@
|
||||
Active Inference model implementations.
|
||||
"""
|
||||
|
||||
from .base import ActiveInferenceModel
|
||||
from .base import ActiveInferenceModel, ModelState
|
||||
from .dispatcher import (
|
||||
ActiveInferenceDispatcher,
|
||||
ActiveInferenceFactory,
|
||||
InferenceConfig,
|
||||
InferenceMethod,
|
||||
PolicyType
|
||||
)
|
||||
|
||||
__all__ = [
|
||||
'ActiveInferenceModel',
|
||||
'ModelState',
|
||||
'ActiveInferenceDispatcher',
|
||||
'ActiveInferenceFactory',
|
||||
'InferenceConfig',
|
||||
'InferenceMethod',
|
||||
'PolicyType'
|
||||
]
|
||||
@ -0,0 +1,30 @@
|
||||
# Example configuration for continuous sampling-based active inference
|
||||
|
||||
method: sampling
|
||||
policy_type: continuous
|
||||
temporal_horizon: 10
|
||||
learning_rate: 0.005
|
||||
precision_init: 1.0
|
||||
use_gpu: false
|
||||
num_samples: 2000
|
||||
temperature: 0.8
|
||||
|
||||
# Model-specific parameters
|
||||
custom_params:
|
||||
exploration_weight: 0.4
|
||||
state_dimensions: [6] # Continuous state space dimension
|
||||
action_dimensions: [2] # Continuous action space dimension
|
||||
|
||||
# Sampling parameters
|
||||
proposal_std: 0.1
|
||||
min_acceptance_rate: 0.2
|
||||
adaptation_rate: 0.01
|
||||
|
||||
# Prior parameters
|
||||
state_prior_mean: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
|
||||
state_prior_std: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
|
||||
|
||||
# MCMC parameters
|
||||
burn_in: 500
|
||||
thinning: 2
|
||||
num_chains: 4
|
||||
@ -0,0 +1,23 @@
|
||||
# Example configuration for discrete variational active inference
|
||||
|
||||
method: variational
|
||||
policy_type: discrete
|
||||
temporal_horizon: 5
|
||||
learning_rate: 0.01
|
||||
precision_init: 1.0
|
||||
use_gpu: false
|
||||
|
||||
# Model-specific parameters
|
||||
custom_params:
|
||||
exploration_weight: 0.3 # Balance between exploration and exploitation
|
||||
state_dimensions: [10, 10] # Example state space dimensions
|
||||
action_dimensions: [4] # Example action space dimensions
|
||||
|
||||
# Prior preferences
|
||||
goal_prior_weight: 2.0
|
||||
entropy_weight: 0.5
|
||||
|
||||
# Learning parameters
|
||||
belief_momentum: 0.9
|
||||
policy_temperature: 0.5
|
||||
max_iterations: 100
|
||||
@ -34,6 +34,8 @@ class InferenceConfig:
|
||||
learning_rate: float
|
||||
precision_init: float
|
||||
use_gpu: bool = False
|
||||
num_samples: int = 1000 # For sampling-based methods
|
||||
temperature: float = 1.0 # For policy selection
|
||||
custom_params: Optional[Dict[str, Any]] = None
|
||||
|
||||
class ActiveInferenceDispatcher:
|
||||
@ -47,6 +49,7 @@ class ActiveInferenceDispatcher:
|
||||
self.config = config
|
||||
self._setup_implementations()
|
||||
self._initialize_matrices()
|
||||
self._rng = np.random.default_rng() # For sampling methods
|
||||
|
||||
def _setup_implementations(self):
|
||||
"""Set up mapping of operations to implementations."""
|
||||
@ -121,9 +124,41 @@ class ActiveInferenceDispatcher:
|
||||
observation: np.ndarray,
|
||||
state: ModelState,
|
||||
**kwargs) -> np.ndarray:
|
||||
"""Sampling-based implementation of belief updates."""
|
||||
# Implementation for sampling-based updates
|
||||
raise NotImplementedError("Sampling-based belief updates not yet implemented")
|
||||
"""Sampling-based implementation of belief updates using particle filtering."""
|
||||
num_samples = self.config.num_samples
|
||||
generative_matrix = kwargs.get('generative_matrix', np.eye(len(state.beliefs)))
|
||||
|
||||
# Initialize particles
|
||||
particles = self._rng.dirichlet(
|
||||
state.beliefs * num_samples,
|
||||
size=num_samples
|
||||
)
|
||||
|
||||
# Compute weights based on likelihood
|
||||
likelihoods = np.array([
|
||||
self._compute_likelihood(observation, p, generative_matrix)
|
||||
for p in particles
|
||||
])
|
||||
weights = likelihoods / np.sum(likelihoods)
|
||||
|
||||
# Resample particles
|
||||
resampled_indices = self._rng.choice(
|
||||
num_samples,
|
||||
size=num_samples,
|
||||
p=weights
|
||||
)
|
||||
particles = particles[resampled_indices]
|
||||
|
||||
# Return mean belief state
|
||||
return np.mean(particles, axis=0)
|
||||
|
||||
def _compute_likelihood(self,
|
||||
observation: np.ndarray,
|
||||
particle: np.ndarray,
|
||||
generative_matrix: np.ndarray) -> float:
|
||||
"""Compute likelihood of observation given particle state."""
|
||||
prediction = np.dot(particle, generative_matrix)
|
||||
return np.exp(-0.5 * np.sum(np.square(observation - prediction)))
|
||||
|
||||
def _mean_field_belief_update(self,
|
||||
observation: np.ndarray,
|
||||
@ -150,8 +185,45 @@ class ActiveInferenceDispatcher:
|
||||
state: ModelState,
|
||||
goal_prior: Optional[np.ndarray] = None,
|
||||
**kwargs) -> np.ndarray:
|
||||
"""Sampling-based implementation of policy inference."""
|
||||
raise NotImplementedError("Sampling-based policy inference not yet implemented")
|
||||
"""Sampling-based implementation of policy inference using MCMC."""
|
||||
if goal_prior is None:
|
||||
goal_prior = np.ones(len(state.policies)) / len(state.policies)
|
||||
|
||||
num_samples = self.config.num_samples
|
||||
current_policies = state.policies.copy()
|
||||
accepted_policies = []
|
||||
|
||||
# MCMC sampling
|
||||
for _ in range(num_samples):
|
||||
# Propose new policy distribution
|
||||
proposal = self._propose_policy(current_policies)
|
||||
|
||||
# Compute acceptance ratio
|
||||
current_energy = self._policy_energy(current_policies, state, goal_prior)
|
||||
proposal_energy = self._policy_energy(proposal, state, goal_prior)
|
||||
|
||||
# Accept/reject
|
||||
if np.log(self._rng.random()) < proposal_energy - current_energy:
|
||||
current_policies = proposal
|
||||
|
||||
accepted_policies.append(current_policies.copy())
|
||||
|
||||
# Return mean policy distribution
|
||||
return np.mean(accepted_policies, axis=0)
|
||||
|
||||
def _propose_policy(self, current: np.ndarray) -> np.ndarray:
|
||||
"""Generate policy proposal for MCMC."""
|
||||
proposal = current + self._rng.normal(0, 0.1, size=current.shape)
|
||||
return self.matrix_ops.normalize_rows(np.maximum(proposal, 0))
|
||||
|
||||
def _policy_energy(self,
|
||||
policies: np.ndarray,
|
||||
state: ModelState,
|
||||
goal_prior: np.ndarray) -> float:
|
||||
"""Compute energy (negative log probability) for policy distribution."""
|
||||
expected_free_energy = self._calculate_expected_free_energy(
|
||||
state, goal_prior)
|
||||
return np.sum(policies * expected_free_energy)
|
||||
|
||||
def _mean_field_policy_inference(self,
|
||||
state: ModelState,
|
||||
@ -165,15 +237,43 @@ class ActiveInferenceDispatcher:
|
||||
goal_prior: np.ndarray,
|
||||
**kwargs) -> np.ndarray:
|
||||
"""Calculate expected free energy for policy evaluation."""
|
||||
# Basic implementation - can be extended based on specific needs
|
||||
pragmatic_value = -np.log(goal_prior + 1e-8) # Avoid log(0)
|
||||
# Enhanced implementation with both pragmatic and epistemic value
|
||||
pragmatic_value = self._calculate_pragmatic_value(state, goal_prior)
|
||||
epistemic_value = self._calculate_epistemic_value(state)
|
||||
return pragmatic_value + epistemic_value
|
||||
|
||||
# Weight between exploration and exploitation
|
||||
exploration_weight = kwargs.get('exploration_weight', 0.5)
|
||||
return (1 - exploration_weight) * pragmatic_value + exploration_weight * epistemic_value
|
||||
|
||||
def _calculate_pragmatic_value(self,
|
||||
state: ModelState,
|
||||
goal_prior: np.ndarray) -> np.ndarray:
|
||||
"""Calculate pragmatic value component of expected free energy."""
|
||||
# KL divergence from current state to goal state
|
||||
return -np.log(goal_prior + 1e-8)
|
||||
|
||||
def _calculate_epistemic_value(self, state: ModelState) -> np.ndarray:
|
||||
"""Calculate epistemic value component of expected free energy."""
|
||||
# Simple implementation - can be extended
|
||||
return -state.prediction_error * np.ones(len(state.policies))
|
||||
# Information gain approximation
|
||||
uncertainty = -np.sum(state.beliefs * np.log(state.beliefs + 1e-8))
|
||||
return -uncertainty * np.ones(len(state.policies))
|
||||
|
||||
def update_precision(self, prediction_error: float) -> float:
|
||||
"""Update precision parameter based on prediction errors."""
|
||||
if self.config.method == InferenceMethod.VARIATIONAL:
|
||||
# Precision updates for variational method
|
||||
self.config.precision_init = (
|
||||
0.9 * self.config.precision_init +
|
||||
0.1 / (prediction_error + 1e-8)
|
||||
)
|
||||
elif self.config.method == InferenceMethod.SAMPLING:
|
||||
# Adaptive step size for sampling method
|
||||
self.config.precision_init = np.clip(
|
||||
1.0 / (prediction_error + 1e-8),
|
||||
0.1,
|
||||
10.0
|
||||
)
|
||||
return self.config.precision_init
|
||||
|
||||
class ActiveInferenceFactory:
|
||||
"""Factory for creating Active Inference instances with specific configurations."""
|
||||
@ -197,6 +297,8 @@ class ActiveInferenceFactory:
|
||||
learning_rate=config_dict['learning_rate'],
|
||||
precision_init=config_dict['precision_init'],
|
||||
use_gpu=config_dict.get('use_gpu', False),
|
||||
num_samples=config_dict.get('num_samples', 1000),
|
||||
temperature=config_dict.get('temperature', 1.0),
|
||||
custom_params=config_dict.get('custom_params', None)
|
||||
)
|
||||
return ActiveInferenceFactory.create(config)
|
||||
212
src/models/active_inference/docs/active_inference_dispatcher.md
Обычный файл
212
src/models/active_inference/docs/active_inference_dispatcher.md
Обычный файл
@ -0,0 +1,212 @@
|
||||
# Active Inference Implementation Documentation
|
||||
|
||||
## Overview
|
||||
|
||||
This document describes the implementation of [[Active Inference]] methods in the cognitive modeling framework. The implementation provides a flexible and extensible architecture for different inference approaches and policy types.
|
||||
|
||||
## Core Components
|
||||
|
||||
### Dispatcher Pattern
|
||||
The [[ActiveInferenceDispatcher]] serves as the main interface for routing operations to specific implementations. It handles:
|
||||
- Belief updates
|
||||
- Policy inference
|
||||
- Free energy calculations
|
||||
- Precision updates
|
||||
|
||||
### Configuration
|
||||
[[InferenceConfig]] provides a structured way to configure the active inference process:
|
||||
- Inference method selection
|
||||
- Policy type specification
|
||||
- Temporal horizon
|
||||
- Learning parameters
|
||||
- GPU acceleration options
|
||||
|
||||
### Inference Methods
|
||||
Three main inference methods are supported:
|
||||
|
||||
1. [[Variational Inference]]
|
||||
- Gradient-based belief updates
|
||||
- Deterministic policy optimization
|
||||
- Precision-weighted prediction errors
|
||||
|
||||
2. [[Sampling Based Inference]]
|
||||
- Particle filtering for belief updates
|
||||
- MCMC for policy inference
|
||||
- Adaptive precision updates
|
||||
|
||||
3. [[Mean Field Inference]]
|
||||
- Factorized approximations
|
||||
- Independent treatment of variables
|
||||
- (Implementation pending)
|
||||
|
||||
### Policy Types
|
||||
|
||||
The framework supports multiple [[Policy Types]]:
|
||||
- `DISCRETE`: Finite action spaces
|
||||
- `CONTINUOUS`: Continuous action spaces
|
||||
- `HIERARCHICAL`: Nested policy structures
|
||||
|
||||
## Usage Examples
|
||||
|
||||
### Basic Configuration
|
||||
```yaml
|
||||
method: variational
|
||||
policy_type: discrete
|
||||
temporal_horizon: 5
|
||||
learning_rate: 0.01
|
||||
precision_init: 1.0
|
||||
```
|
||||
|
||||
See [[Configuration Examples]] for more detailed examples.
|
||||
|
||||
### Code Usage
|
||||
```python
|
||||
from models.active_inference import (
|
||||
ActiveInferenceFactory,
|
||||
InferenceConfig,
|
||||
InferenceMethod,
|
||||
PolicyType
|
||||
)
|
||||
|
||||
# Create configuration
|
||||
config = InferenceConfig(...)
|
||||
dispatcher = ActiveInferenceFactory.create(config)
|
||||
|
||||
# Update beliefs
|
||||
updated_beliefs = dispatcher.dispatch_belief_update(
|
||||
observation=current_observation,
|
||||
current_state=model_state
|
||||
)
|
||||
```
|
||||
|
||||
## Implementation Details
|
||||
|
||||
### Free Energy Calculation
|
||||
The [[Expected Free Energy]] calculation combines:
|
||||
1. Pragmatic value (goal-directed behavior)
|
||||
2. Epistemic value (information seeking)
|
||||
|
||||
```python
|
||||
def _calculate_expected_free_energy(self, state, goal_prior, **kwargs):
|
||||
pragmatic_value = self._calculate_pragmatic_value(state, goal_prior)
|
||||
epistemic_value = self._calculate_epistemic_value(state)
|
||||
exploration_weight = kwargs.get('exploration_weight', 0.5)
|
||||
return (1 - exploration_weight) * pragmatic_value + exploration_weight * epistemic_value
|
||||
```
|
||||
|
||||
### Belief Updates
|
||||
Different methods for [[Belief Updates]]:
|
||||
|
||||
1. Variational:
|
||||
- Gradient-based updates
|
||||
- Precision-weighted errors
|
||||
|
||||
2. Sampling:
|
||||
- Particle filtering
|
||||
- Importance resampling
|
||||
|
||||
3. Mean Field:
|
||||
- Factorized updates
|
||||
- Independent parameter optimization
|
||||
|
||||
### Policy Inference
|
||||
[[Policy Inference]] implementations:
|
||||
|
||||
1. Variational:
|
||||
```python
|
||||
expected_free_energy = self._calculate_expected_free_energy(...)
|
||||
return self.matrix_ops.softmax(-expected_free_energy)
|
||||
```
|
||||
|
||||
2. MCMC:
|
||||
- Metropolis-Hastings sampling
|
||||
- Proposal distribution
|
||||
- Energy-based acceptance
|
||||
|
||||
## Advanced Features
|
||||
|
||||
### Precision Updates
|
||||
[[Precision Updates]] adapt based on prediction errors:
|
||||
- Variational: Running average updates
|
||||
- Sampling: Adaptive step sizes
|
||||
- Bounded optimization
|
||||
|
||||
### Matrix Operations
|
||||
Utility functions in [[Matrix Operations]]:
|
||||
- Normalization
|
||||
- Softmax
|
||||
- Information metrics
|
||||
|
||||
### GPU Acceleration
|
||||
[[GPU Support]] preparation:
|
||||
- Flag in configuration
|
||||
- Matrix operation optimization
|
||||
- Batch processing support
|
||||
|
||||
## Configuration Examples
|
||||
|
||||
### [[Discrete Variational Config]]
|
||||
```yaml
|
||||
method: variational
|
||||
policy_type: discrete
|
||||
temporal_horizon: 5
|
||||
custom_params:
|
||||
exploration_weight: 0.3
|
||||
state_dimensions: [10, 10]
|
||||
```
|
||||
|
||||
### [[Continuous Sampling Config]]
|
||||
```yaml
|
||||
method: sampling
|
||||
policy_type: continuous
|
||||
num_samples: 2000
|
||||
custom_params:
|
||||
proposal_std: 0.1
|
||||
burn_in: 500
|
||||
```
|
||||
|
||||
## Best Practices
|
||||
|
||||
### [[Performance Optimization]]
|
||||
1. Use appropriate number of samples
|
||||
2. Enable GPU for large state spaces
|
||||
3. Tune precision updates
|
||||
|
||||
### [[Numerical Stability]]
|
||||
1. Add small constants to denominators
|
||||
2. Use log probabilities where appropriate
|
||||
3. Implement bounds checking
|
||||
|
||||
### [[Debugging Tips]]
|
||||
1. Monitor acceptance rates in MCMC
|
||||
2. Track prediction errors
|
||||
3. Validate belief normalization
|
||||
|
||||
## Related Topics
|
||||
|
||||
- [[Free Energy Principle]]
|
||||
- [[Active Inference Theory]]
|
||||
- [[Variational Bayes]]
|
||||
- [[MCMC Methods]]
|
||||
- [[Particle Filtering]]
|
||||
|
||||
## Future Extensions
|
||||
|
||||
### Planned Features
|
||||
1. [[Hierarchical Policies]]
|
||||
- Nested action spaces
|
||||
- Multi-scale temporal horizons
|
||||
|
||||
2. [[Advanced Sampling]]
|
||||
- Hamiltonian Monte Carlo
|
||||
- Sequential Monte Carlo
|
||||
|
||||
3. [[Neural Implementation]]
|
||||
- Deep active inference
|
||||
- Learned generative models
|
||||
|
||||
## References
|
||||
|
||||
1. Friston, K. (2010). [[The Free-Energy Principle]]
|
||||
2. Da Costa, L., et al. (2020). [[Active Inference Algorithms]]
|
||||
3. Parr, T., & Friston, K. (2019). [[Discrete Active Inference]]
|
||||
214
src/models/active_inference/docs/dispatcher.md
Обычный файл
214
src/models/active_inference/docs/dispatcher.md
Обычный файл
@ -0,0 +1,214 @@
|
||||
# Active Inference Dispatcher
|
||||
|
||||
## Overview
|
||||
|
||||
The [[ActiveInferenceDispatcher]] is the core component that manages different implementations of active inference methods. It provides a clean interface for routing operations while handling the complexity of different inference approaches.
|
||||
|
||||
## Class Structure
|
||||
|
||||
```python
|
||||
class ActiveInferenceDispatcher:
|
||||
def __init__(self, config: InferenceConfig):
|
||||
self.config = config
|
||||
self._setup_implementations()
|
||||
self._initialize_matrices()
|
||||
```
|
||||
|
||||
## Key Components
|
||||
|
||||
### 1. Method Dispatch
|
||||
The dispatcher uses a dictionary-based routing system to map methods to their implementations:
|
||||
|
||||
```python
|
||||
self._implementations = {
|
||||
InferenceMethod.VARIATIONAL: {
|
||||
'belief_update': self._variational_belief_update,
|
||||
'policy_inference': self._variational_policy_inference
|
||||
},
|
||||
# ... other methods
|
||||
}
|
||||
```
|
||||
|
||||
See [[Implementation Mapping]] for details.
|
||||
|
||||
### 2. Configuration
|
||||
Uses [[InferenceConfig]] for structured configuration:
|
||||
- Method selection
|
||||
- Policy type
|
||||
- Learning parameters
|
||||
- Custom parameters
|
||||
|
||||
### 3. Matrix Operations
|
||||
Integrates with [[Matrix Operations]] for efficient computations:
|
||||
- Normalization
|
||||
- Softmax
|
||||
- Information metrics
|
||||
|
||||
## Core Operations
|
||||
|
||||
### Belief Updates
|
||||
|
||||
The [[Belief Update Methods]] include:
|
||||
|
||||
1. **Variational**
|
||||
```python
|
||||
def _variational_belief_update(self, observation, state, **kwargs):
|
||||
prediction = np.dot(state.beliefs, generative_matrix)
|
||||
prediction_error = observation - prediction
|
||||
return state.beliefs + state.precision * prediction_error
|
||||
```
|
||||
|
||||
2. **Sampling**
|
||||
```python
|
||||
def _sampling_belief_update(self, observation, state, **kwargs):
|
||||
particles = self._initialize_particles(state)
|
||||
weights = self._compute_weights(particles, observation)
|
||||
return self._resample_particles(particles, weights)
|
||||
```
|
||||
|
||||
See [[Belief Update Implementation]] for more details.
|
||||
|
||||
### Policy Inference
|
||||
|
||||
[[Policy Inference Methods]]:
|
||||
|
||||
1. **Variational**
|
||||
- Expected free energy minimization
|
||||
- Softmax policy selection
|
||||
|
||||
2. **MCMC**
|
||||
- Metropolis-Hastings algorithm
|
||||
- Proposal distribution
|
||||
- Energy-based acceptance
|
||||
|
||||
## Advanced Features
|
||||
|
||||
### 1. Precision Updates
|
||||
[[Precision Adaptation]] mechanisms:
|
||||
```python
|
||||
def update_precision(self, prediction_error: float) -> float:
|
||||
if self.config.method == InferenceMethod.VARIATIONAL:
|
||||
return self._update_variational_precision(prediction_error)
|
||||
elif self.config.method == InferenceMethod.SAMPLING:
|
||||
return self._update_sampling_precision(prediction_error)
|
||||
```
|
||||
|
||||
### 2. Free Energy Calculation
|
||||
[[Free Energy Components]]:
|
||||
1. Pragmatic value
|
||||
2. Epistemic value
|
||||
3. Exploration-exploitation balance
|
||||
|
||||
### 3. GPU Support
|
||||
[[GPU Acceleration]] features:
|
||||
- Configuration flag
|
||||
- Matrix operation optimization
|
||||
- Batch processing
|
||||
|
||||
## Usage Examples
|
||||
|
||||
### Basic Usage
|
||||
```python
|
||||
# Create dispatcher
|
||||
config = InferenceConfig(method=InferenceMethod.VARIATIONAL, ...)
|
||||
dispatcher = ActiveInferenceFactory.create(config)
|
||||
|
||||
# Update beliefs
|
||||
new_beliefs = dispatcher.dispatch_belief_update(
|
||||
observation=current_observation,
|
||||
current_state=model_state
|
||||
)
|
||||
|
||||
# Infer policies
|
||||
policies = dispatcher.dispatch_policy_inference(
|
||||
state=model_state,
|
||||
goal_prior=goal_distribution
|
||||
)
|
||||
```
|
||||
|
||||
See [[Usage Examples]] for more scenarios.
|
||||
|
||||
## Configuration
|
||||
|
||||
### Variational Configuration
|
||||
```yaml
|
||||
method: variational
|
||||
policy_type: discrete
|
||||
temporal_horizon: 5
|
||||
learning_rate: 0.01
|
||||
precision_init: 1.0
|
||||
```
|
||||
|
||||
### Sampling Configuration
|
||||
```yaml
|
||||
method: sampling
|
||||
policy_type: continuous
|
||||
num_samples: 2000
|
||||
custom_params:
|
||||
proposal_std: 0.1
|
||||
```
|
||||
|
||||
See [[Configuration Guide]] for details.
|
||||
|
||||
## Best Practices
|
||||
|
||||
### Performance
|
||||
[[Performance Guidelines]]:
|
||||
1. Use appropriate batch sizes
|
||||
2. Enable GPU for large models
|
||||
3. Optimize matrix operations
|
||||
|
||||
### Numerical Stability
|
||||
[[Stability Guidelines]]:
|
||||
1. Use log probabilities
|
||||
2. Add small constants
|
||||
3. Check bounds
|
||||
|
||||
### Debugging
|
||||
[[Debugging Strategies]]:
|
||||
1. Monitor convergence
|
||||
2. Track prediction errors
|
||||
3. Validate distributions
|
||||
|
||||
## Extension Points
|
||||
|
||||
### Adding New Methods
|
||||
To add a new inference method:
|
||||
1. Add to [[InferenceMethod]] enum
|
||||
2. Implement belief update method
|
||||
3. Implement policy inference method
|
||||
4. Add to implementation mapping
|
||||
|
||||
### Custom Policies
|
||||
[[Custom Policy Implementation]]:
|
||||
1. Extend PolicyType enum
|
||||
2. Implement policy-specific methods
|
||||
3. Add configuration support
|
||||
|
||||
## Related Components
|
||||
|
||||
- [[Active Inference Model]]
|
||||
- [[Matrix Operations]]
|
||||
- [[Configuration System]]
|
||||
- [[Visualization Tools]]
|
||||
|
||||
## Future Development
|
||||
|
||||
### Planned Features
|
||||
1. [[Hierarchical Implementation]]
|
||||
- Nested policies
|
||||
- Multi-scale inference
|
||||
|
||||
2. [[Advanced Methods]]
|
||||
- Hamiltonian Monte Carlo
|
||||
- Variational message passing
|
||||
|
||||
3. [[Optimization]]
|
||||
- Parallel processing
|
||||
- Memory efficiency
|
||||
|
||||
## References
|
||||
|
||||
1. [[Implementation Architecture]]
|
||||
2. [[Method Specifications]]
|
||||
3. [[API Documentation]]
|
||||
Загрузка…
x
Ссылка в новой задаче
Block a user