File size: 6,317 Bytes
6374a33 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 | {
"acceptance": {
"artifact_at_most_2_mb": true,
"empty_skops_untrusted_type_set": true,
"exact_reproduction": true,
"lockbox_error_recall_at_least_0.50": true,
"validation_error_recall_margin_delta_at_least_0.02": true,
"validation_routed_accuracy_gain_at_least_0.03": true
},
"acceptance_decision": "release_work_pending",
"artifacts": {
"candidate": {
"bytes": 1104573,
"path": "model.skops",
"sha256": "1d5a785c69b01135981c9052aa8d124cf3d423f7d15e558820502d37a18686b0",
"skops_untrusted_types": []
},
"primary_baseline": {
"bytes": 28966090,
"path": "primary_baseline.skops",
"sha256": "f735dfa1e498fef3c6a0d0a87664cd2e77acaadc892af7875c7ab30148a0e020",
"skops_untrusted_types": []
}
},
"base_model_revision": null,
"campaign_id": "banking77-intent-error-predictor-v1",
"candidate": "histogram_gradient_boosting_score_error_predictor",
"code_sha256": "852699c3ee6b8e545f9f0fd6c4cf8ccce0bccdea3d67ae4ceceefcdda1404bd7",
"configuration": {
"architecture": "histogram gradient boosting over probabilities, predicted-intent one-hot, confidence geometry, and five text-shape features",
"campaign_id": "banking77-intent-error-predictor-v1",
"candidate": "histogram_gradient_boosting_score_error_predictor",
"code_sha256": "852699c3ee6b8e545f9f0fd6c4cf8ccce0bccdea3d67ae4ceceefcdda1404bd7",
"consumer": "banking-support intent router with a human review queue",
"dataset_hashes": {
"categories.json": "53261da888122daf2d120d925458631d9619e15d82e56052e7a42e535ce32b63",
"test.csv": "d12d6e3bc4c3103966ae786dc435913c0c563dfa328f5a3646d0e62cfeeb474d",
"train.csv": "b06e26ac675513959a63135f11b94ea7786ed02da65db93a5650d8838cbc664b"
},
"dataset_hub_revision": "90d4e2ee5521c04fc1488f065b8b083658768c57",
"dataset_source_revision": "57ec275d8078af65b7731c2a98be812d844a6d6b",
"hyperparameters": {
"l2_regularization": 1.0,
"learning_rate": 0.06,
"max_iter": 140,
"max_leaf_nodes": 15,
"min_samples_leaf": 25
},
"input_contract": "77 probabilities in sorted Banking77 label order plus bounded text-shape features",
"objective": "balanced binary log loss for primary-router error prediction",
"output_contract": "error probability and advisory review decision",
"preprocessing": "NFKC casefold whitespace normalization for group hashing; TF-IDF primary; score and text-shape candidate features",
"primary": "24k word plus 36k character TF-IDF with multinomial logistic regression C=4",
"seed": 20260811,
"split": "normalized-text SHA-256 grouped 60/20/20 development partitions; official test untouched lockbox with overlap removal",
"task": "predict whether a fixed BANKING77 primary router is wrong"
},
"dataset_revision": "57ec275d8078af65b7731c2a98be812d844a6d6b",
"experiment_spec_hash": "8a4c7e16a4a0ef3615735c7214e429df5268dc1610612ce6e833aa9e3829ccdc",
"hypothesis": "learned score-shape and intent features improve error capture over margin-only triage at equal review rate",
"lockbox": {
"candidate": {
"confusion": {
"fn": 72,
"fp": 326,
"tn": 2412,
"tp": 263
},
"coverage": 0.8083306215424666,
"error_f1": 0.5692640692640693,
"error_precision": 0.4465195246179966,
"error_prevalence": 0.10901399284087211,
"error_recall": 0.7850746268656716,
"pr_auc": 0.5199652272918066,
"review_rate": 0.19166937845753335,
"roc_auc": 0.9035912475605902,
"routed_accuracy": 0.9710144927536232,
"threshold": 0.27443790545050933
},
"margin_baseline": {
"confusion": {
"fn": 78,
"fp": 366,
"tn": 2372,
"tp": 257
},
"coverage": 0.7972665148063781,
"error_f1": 0.5365344467640919,
"error_precision": 0.41252006420545745,
"error_prevalence": 0.10901399284087211,
"error_recall": 0.7671641791044777,
"pr_auc": 0.5863979952872228,
"review_rate": 0.20273348519362186,
"roc_auc": 0.9092310543702234,
"routed_accuracy": 0.9681632653061224,
"threshold": 0.6936486958535751
}
},
"partitions": {
"complement_train": 1951,
"lockbox": 3073,
"lockbox_overlap_removed": 7,
"primary_train": 6050,
"validation": 2002
},
"primary": {
"lockbox_accuracy": 0.8909860071591279,
"lockbox_macro_f1": 0.8910051474654282,
"validation_accuracy": 0.8951048951048951,
"validation_macro_f1": 0.8919223884704242
},
"record_type": "fit",
"release_validation": {
"bounded_inference_api": true,
"lockbox_scores_persisted": true,
"reference_consumer": true,
"representative_example_passed": true
},
"reproduction": {
"clean_refit_predictions_exact": true,
"clean_refit_scores_exact": true,
"serialization_scores_exact": true
},
"telemetry": {
"cache_growth_bytes": 1081070,
"cpu_seconds": 14.953067,
"minor_page_fault_delta": 37952,
"peak_rss_bytes": 743817216,
"single_training_process": true,
"temporary_disk_growth_bytes": 30076757,
"thread_limit": 8,
"wall_seconds": 6.962122584000099
},
"validation": {
"candidate": {
"confusion": {
"fn": 42,
"fp": 233,
"tn": 1559,
"tp": 168
},
"coverage": 0.7997002997002997,
"error_f1": 0.5499181669394435,
"error_precision": 0.41895261845386533,
"error_prevalence": 0.1048951048951049,
"error_recall": 0.8,
"pr_auc": 0.48498408831658685,
"review_rate": 0.2002997002997003,
"roc_auc": 0.8968856292517007,
"routed_accuracy": 0.9737663960024985,
"threshold": 0.27443790545050933
},
"margin_baseline": {
"confusion": {
"fn": 52,
"fp": 243,
"tn": 1549,
"tp": 158
},
"coverage": 0.7997002997002997,
"error_f1": 0.5171849427168577,
"error_precision": 0.3940149625935162,
"error_prevalence": 0.1048951048951049,
"error_recall": 0.7523809523809524,
"pr_auc": 0.5163906717576305,
"review_rate": 0.2002997002997003,
"roc_auc": 0.8986208545918366,
"routed_accuracy": 0.9675202998126171,
"threshold": 0.6936486958535751
}
}
}
|