jung-geun/PSO
1
0
Fork 0
mirror of https://github.com/jung-geun/PSO.git synced 2026-09-20 14:11:48 +09:00
Code Issues Packages Projects Releases Wiki Activity
Files
master
PSO/benchmark_results/pso_v8_post_training_ensemble.json
T
jung-geun 813433000a feat: modernize PSO and add convergence research 
Migrate the package and examples to the tensor-native PyTorch implementation, add benchmark evidence, and add the guarded post-training convergence protocol with TensorBoard progress monitoring and hash-verified recovery.

Constraint: Preserve one-shot official-test sealing and auditable research artifacts

Rejected: Commit local .omc runs and downloaded datasets | multi-gigabyte runtime state is machine-local

Confidence: high

Scope-risk: broad

Not-tested: Production CUDA run on pieroot-server
2026-09-07 22:03:25 +09:00

683 lines
20 KiB
JSON
Raw Permalink Blame History

{
"protocol_version": "POST-TRAINING-PSO-ENSEMBLE 1.1.0",
"config": {
"iteration": 1,
"archived_iteration0_reference": {
"epochs": 50,
"queries_per_seed": 1500,
"sample_evaluations_per_seed": 15000000,
"reason": "wall_time_ratio_gate_exceeded"
},
"datasets": [
"mnist",
"fashion_mnist"
],
"split_seed": 20260904,
"search_samples": 50000,
"validation_samples": 10000,
"pool_seeds": [
201,
202,
203,
204,
205
],
"reference_single_seed": 201,
"equal_budget_single_epochs": 50,
"adam_lr": 0.001,
"adam_batch_size": 256,
"pso": {
"method": "constriction",
"evaluation": "full",
"renewal": "loss",
"particles": 30,
"epochs": 30,
"swarm_seeds": [
301,
302,
303
],
"queries_per_seed": 900,
"sample_evaluations_per_seed": 9000000,
"particle_bounds": [
-4.0,
4.0
],
"boundary_strategy": "reflect",
"velocity_limit_ratio": 0.1,
"initial_position_noise": 0.0
},
"device": "mps"
},
"workloads": {
"mnist": {
"provenance": {
"dataset_name": "MNIST",
"split_seed": 20260904,
"search_samples": 50000,
"validation_samples": 10000,
"normalization": {
"mean": 0.1307128667831421,
"std": 0.3081730008125305
},
"data_fingerprint": "861b07884cffefb9",
"split_fingerprint": "258b42dde7da6324"
},
"training": {
"architecture": "CompactCNN",
"parameters": 9098,
"pool_seeds": [
201,
202,
203,
204,
205
],
"pool_epochs_each": 10,
"adam_pool_epochs": 50,
"adam_lr": 0.001,
"adam_batch_size": 256,
"adam_pool_wall_time_seconds": 22.098886500985827,
"single_50e_epochs": 50,
"single_50e_wall_time_seconds": 22.06671220799035,
"model_fingerprints": {
"201": "046605462c4fc294",
"single_50e": "53ff4c61db919c1e",
"202": "cd3150799e0db43e",
"203": "1feb0f4dd7630ec2",
"204": "9dc0606cd8076cb2",
"205": "5bc16dc8f86cc869"
}
},
"validation_cache": {
"valid": true,
"pool_forward_passes": 5,
"long_single_forward_passes": 1,
"base_cnn_forward_passes_during_optimization": 0,
"shape": [
5,
10000,
10
],
"memory_bytes": 2400000,
"wall_time_seconds": 0.09316845799912699
},
"validation": {
"methods": {
"reference_single_10e": {
"accuracy": 98.23,
"nll": 0.060105,
"brier": 0.027982,
"ece": 0.003458,
"margin": 0.969858
},
"best_single_10e": {
"accuracy": 98.29,
"nll": 0.054029,
"brier": 0.025526,
"ece": 0.002795,
"margin": 0.971419,
"selected_seed": 205
},
"single_50e": {
"accuracy": 98.52,
"nll": 0.073988,
"brier": 0.024922,
"ece": 0.009463,
"margin": 0.989411
},
"uniform_ensemble": {
"accuracy": 98.63,
"nll": 0.046385,
"brier": 0.021527,
"ece": 0.006228,
"margin": 0.965339
},
"uniform_temperature": {
"fitted_temperature": 0.854572,
"wall_time_seconds": 0.04119049999280833,
"evaluations": 23,
"metrics": {
"accuracy": 98.63,
"nll": 0.045355,
"brier": 0.021099,
"ece": 0.002518,
"margin": 0.972273
}
},
"slsqp_weights": {
"weights": [
0.05767137130295378,
0.2585033373652759,
0.10841769748910005,
0.25254869759276993,
0.3228588962499004
],
"evaluations": 23,
"wall_time_seconds": 0.009686834004241973,
"success": true,
"message": "Optimization terminated successfully",
"metrics": {
"accuracy": 98.6,
"nll": 0.045902,
"brier": 0.021307,
"ece": 0.005994,
"margin": 0.965581
}
},
"pso_weights": {
"per_seed_runs": [
{
"seed": 301,
"queries": 900,
"sample_evaluations": 9000000,
"wall_time_seconds": 1.8924723340023775,
"metrics": {
"accuracy": 98.61,
"nll": 0.045902,
"brier": 0.021307,
"ece": 0.005893,
"margin": 0.965582
},
"weights": [
0.057233214378356934,
0.2590548098087311,
0.10862385481595993,
0.25228697061538696,
0.3228012025356293
]
},
{
"seed": 302,
"queries": 900,
"sample_evaluations": 9000000,
"wall_time_seconds": 1.7105932499980554,
"metrics": {
"accuracy": 98.6,
"nll": 0.045902,
"brier": 0.021306,
"ece": 0.006183,
"margin": 0.965578
},
"weights": [
0.058317527174949646,
0.25828817486763,
0.10924994200468063,
0.252756267786026,
0.3213881254196167
]
},
{
"seed": 303,
"queries": 900,
"sample_evaluations": 9000000,
"wall_time_seconds": 1.8959678749961313,
"metrics": {
"accuracy": 98.61,
"nll": 0.045902,
"brier": 0.021307,
"ece": 0.005833,
"margin": 0.965583
},
"weights": [
0.056311722844839096,
0.2563577890396118,
0.11076968908309937,
0.25350210070610046,
0.3230587840080261
]
}
],
"selected_seed": 301,
"selected_weights": [
0.057233214378356934,
0.2590548098087311,
0.10862385481595993,
0.25228697061538696,
0.3228012025356293
],
"metrics": {
"accuracy": 98.61,
"nll": 0.045902,
"brier": 0.021307,
"ece": 0.005893,
"margin": 0.965582
},
"queries_per_seed": 900,
"sample_evaluations_per_seed": 9000000,
"total_queries": 2700,
"total_sample_evaluations": 27000000,
"median_one_seed_wall_time_seconds": 1.8924723340023775,
"total_wall_time_seconds": 5.499033458996564
}
}
},
"official_test_data_loaded_before_freeze": false,
"official_test_evaluations_before_freeze": 0,
"confirmation": {
"official_test_data_loaded": true,
"test_cache_counts": {
"dataset_loads": 1,
"pool_forward_passes": 5,
"long_single_forward_passes": 1,
"base_cnn_forward_passes_during_optimization": 0,
"memory_bytes": 2400000,
"wall_time_seconds": 0.11975424998672679
},
"frozen_methods": {
"selected_pso_seed": 301,
"selected_pso_weights": [
0.057233214378356934,
0.2590548098087311,
0.10862385481595993,
0.25228697061538696,
0.3228012025356293
],
"slsqp_weights": [
0.05767137130295378,
0.2585033373652759,
0.10841769748910005,
0.25254869759276993,
0.3228588962499004
],
"fitted_temperature": 0.854572
},
"methods": {
"reference_single_10e": {
"accuracy": 98.47,
"nll": 0.044991,
"brier": 0.022366,
"ece": 0.003516,
"margin": 0.970575
},
"best_single_10e": {
"accuracy": 98.55,
"nll": 0.044944,
"brier": 0.022052,
"ece": 0.002515,
"margin": 0.97465
},
"single_50e": {
"accuracy": 98.6,
"nll": 0.062102,
"brier": 0.022912,
"ece": 0.008283,
"margin": 0.988829
},
"uniform_ensemble": {
"accuracy": 98.86,
"nll": 0.036184,
"brier": 0.01801,
"ece": 0.006311,
"margin": 0.968415
},
"uniform_temperature": {
"accuracy": 98.86,
"nll": 0.034129,
"brier": 0.017658,
"ece": 0.00334,
"margin": 0.974869
},
"slsqp_weights": {
"accuracy": 98.83,
"nll": 0.036179,
"brier": 0.017989,
"ece": 0.005818,
"margin": 0.969195
},
"pso_weights": {
"accuracy": 98.83,
"nll": 0.036178,
"brier": 0.017989,
"ece": 0.005817,
"margin": 0.969196
}
},
"confirmation_gates": {
"all_values_finite": true,
"official_test_dataset_loads": 1,
"official_test_pool_forward_passes": 5,
"official_test_long_single_forward_passes": 1,
"maximum_pso_accuracy_regression_vs_uniform_pp": true,
"pso_nll_below_reference_single": true,
"maximum_pso_nll_regression_vs_equal_budget_single": true,
"pass": true
}
}
},
"fashion_mnist": {
"provenance": {
"dataset_name": "FashionMNIST",
"split_seed": 20260904,
"search_samples": 50000,
"validation_samples": 10000,
"normalization": {
"mean": 0.28573352098464966,
"std": 0.35278451442718506
},
"data_fingerprint": "b56a39e2abf7d9ad",
"split_fingerprint": "489c1a04fb44b59a"
},
"training": {
"architecture": "CompactCNN",
"parameters": 9098,
"pool_seeds": [
201,
202,
203,
204,
205
],
"pool_epochs_each": 10,
"adam_pool_epochs": 50,
"adam_lr": 0.001,
"adam_batch_size": 256,
"adam_pool_wall_time_seconds": 22.363562541009742,
"single_50e_epochs": 50,
"single_50e_wall_time_seconds": 22.406685083013144,
"model_fingerprints": {
"201": "8189d6d2cce8bde8",
"single_50e": "5a5b8444219bf301",
"202": "a5492b64bfda73cd",
"203": "664e02f11e62031f",
"204": "be2a23fca4153703",
"205": "d3a37cfa6ee482bf"
}
},
"validation_cache": {
"valid": true,
"pool_forward_passes": 5,
"long_single_forward_passes": 1,
"base_cnn_forward_passes_during_optimization": 0,
"shape": [
5,
10000,
10
],
"memory_bytes": 2400000,
"wall_time_seconds": 0.08030625000537839
},
"validation": {
"methods": {
"reference_single_10e": {
"accuracy": 89.47,
"nll": 0.303799,
"brier": 0.150671,
"ece": 0.012197,
"margin": 0.808017
},
"best_single_10e": {
"accuracy": 89.47,
"nll": 0.303799,
"brier": 0.150671,
"ece": 0.012197,
"margin": 0.808017,
"selected_seed": 201
},
"single_50e": {
"accuracy": 90.32,
"nll": 0.28966,
"brier": 0.142115,
"ece": 0.024556,
"margin": 0.869484
},
"uniform_ensemble": {
"accuracy": 90.28,
"nll": 0.286751,
"brier": 0.142881,
"ece": 0.023702,
"margin": 0.795588
},
"uniform_temperature": {
"fitted_temperature": 0.906309,
"wall_time_seconds": 0.04135037500236649,
"evaluations": 21,
"metrics": {
"accuracy": 90.28,
"nll": 0.285048,
"brier": 0.141557,
"ece": 0.011886,
"margin": 0.814383
}
},
"slsqp_weights": {
"weights": [
0.23910131729416598,
0.2493341982620314,
0.18054455195387528,
0.021147090032828317,
0.30987284245709906
],
"evaluations": 23,
"wall_time_seconds": 0.010389625007519498,
"success": true,
"message": "Optimization terminated successfully",
"metrics": {
"accuracy": 90.42,
"nll": 0.285338,
"brier": 0.142289,
"ece": 0.023705,
"margin": 0.798413
}
},
"pso_weights": {
"per_seed_runs": [
{
"seed": 301,
"queries": 900,
"sample_evaluations": 9000000,
"wall_time_seconds": 1.8963403329980792,
"metrics": {
"accuracy": 90.42,
"nll": 0.285338,
"brier": 0.142289,
"ece": 0.023661,
"margin": 0.798397
},
"weights": [
0.2385721057653427,
0.24956248700618744,
0.1805897355079651,
0.021864112466573715,
0.30941155552864075
]
},
{
"seed": 302,
"queries": 900,
"sample_evaluations": 9000000,
"wall_time_seconds": 1.7477522080007475,
"metrics": {
"accuracy": 90.43,
"nll": 0.285338,
"brier": 0.142286,
"ece": 0.023912,
"margin": 0.79841
},
"weights": [
0.2415684461593628,
0.2483692765235901,
0.18002673983573914,
0.02121216244995594,
0.30882343649864197
]
},
{
"seed": 303,
"queries": 900,
"sample_evaluations": 9000000,
"wall_time_seconds": 2.0081020840007113,
"metrics": {
"accuracy": 90.41,
"nll": 0.285338,
"brier": 0.14229,
"ece": 0.023601,
"margin": 0.798444
},
"weights": [
0.23893120884895325,
0.24851834774017334,
0.18254171311855316,
0.019977180287241936,
0.31003159284591675
]
}
],
"selected_seed": 301,
"selected_weights": [
0.2385721057653427,
0.24956248700618744,
0.1805897355079651,
0.021864112466573715,
0.30941155552864075
],
"metrics": {
"accuracy": 90.42,
"nll": 0.285338,
"brier": 0.142289,
"ece": 0.023661,
"margin": 0.798397
},
"queries_per_seed": 900,
"sample_evaluations_per_seed": 9000000,
"total_queries": 2700,
"total_sample_evaluations": 27000000,
"median_one_seed_wall_time_seconds": 1.8963403329980792,
"total_wall_time_seconds": 5.652194624999538
}
}
},
"official_test_data_loaded_before_freeze": false,
"official_test_evaluations_before_freeze": 0,
"confirmation": {
"official_test_data_loaded": true,
"test_cache_counts": {
"dataset_loads": 1,
"pool_forward_passes": 5,
"long_single_forward_passes": 1,
"base_cnn_forward_passes_during_optimization": 0,
"memory_bytes": 2400000,
"wall_time_seconds": 0.11122124998655636
},
"frozen_methods": {
"selected_pso_seed": 301,
"selected_pso_weights": [
0.2385721057653427,
0.24956248700618744,
0.1805897355079651,
0.021864112466573715,
0.30941155552864075
],
"slsqp_weights": [
0.23910131729416598,
0.2493341982620314,
0.18054455195387528,
0.021147090032828317,
0.30987284245709906
],
"fitted_temperature": 0.906309
},
"methods": {
"reference_single_10e": {
"accuracy": 88.9,
"nll": 0.314516,
"brier": 0.161465,
"ece": 0.00532,
"margin": 0.805307
},
"best_single_10e": {
"accuracy": 88.9,
"nll": 0.314516,
"brier": 0.161465,
"ece": 0.00532,
"margin": 0.805307
},
"single_50e": {
"accuracy": 89.93,
"nll": 0.302348,
"brier": 0.148289,
"ece": 0.026401,
"margin": 0.865728
},
"uniform_ensemble": {
"accuracy": 89.65,
"nll": 0.293522,
"brier": 0.1514,
"ece": 0.019111,
"margin": 0.791698
},
"uniform_temperature": {
"accuracy": 89.65,
"nll": 0.291996,
"brier": 0.150581,
"ece": 0.007597,
"margin": 0.810594
},
"slsqp_weights": {
"accuracy": 89.54,
"nll": 0.291696,
"brier": 0.150558,
"ece": 0.017186,
"margin": 0.794293
},
"pso_weights": {
"accuracy": 89.54,
"nll": 0.2917,
"brier": 0.150559,
"ece": 0.017378,
"margin": 0.794274
}
},
"confirmation_gates": {
"all_values_finite": true,
"official_test_dataset_loads": 1,
"official_test_pool_forward_passes": 5,
"official_test_long_single_forward_passes": 1,
"maximum_pso_accuracy_regression_vs_uniform_pp": true,
"pso_nll_below_reference_single": true,
"maximum_pso_nll_regression_vs_equal_budget_single": true,
"pass": true
}
}
}
},
"development_pass": true,
"development_gates": {
"pass": true,
"failed_hard_gate_count": 0,
"gate_results": {
"all_values_finite": true,
"validation_pool_forward_passes_exact": true,
"optimization_base_model_forward_passes": true,
"official_test_data_loaded_before_freeze": true,
"slsqp_solver_success": true,
"query_and_sample_accounting_exact": true,
"maximum_pso_nll_regression_vs_uniform": true,
"maximum_pso_accuracy_regression_vs_uniform_pp": true,
"pso_nll_below_reference_single": true,
"maximum_pso_nll_regression_vs_equal_budget_single": true,
"maximum_relative_pso_nll_gap_vs_slsqp": true,
"cross_dataset_mean_relative_pso_nll_reduction_vs_uniform_minimum": true,
"maximum_median_one_seed_pso_to_pool_training_wall_ratio": true
},
"issues": []
},
"policy_frozen": true,
"official_test_data_loaded": true,
"official_test_evaluations_before_freeze": 0,
"post_test_tuning_or_reruns": 0,
"resource_totals": {
"adam_pool_epochs": 100,
"adam_pool_wall_time_seconds": 44.46244904199557,
"single_50e_wall_time_seconds": 44.473397291003494,
"validation_cache_forward_passes": 12,
"pso_total_queries": 5400,
"pso_total_sample_evaluations": 54000000,
"pso_research_wall_time_seconds": 11.151228083996102,
"pso_production_wall_time_seconds": 3.7888126670004567,
"pso_to_pool_wall_ratio": 0.0852162664640409,
"slsqp_total_evaluations": 46,
"slsqp_total_wall_time_seconds": 0.02007645901176147,
"official_test_cache_forward_passes": 12
}
}
Reference in New Issue View Git Blame Copy Permalink