tiny-hinglish-turn-detector / reports /partial_source_holdout_tinytcn_metrics.json
suvradeepp's picture
Publish Tiny Hinglish Turn Detector development preview
35d483e verified
Raw
History Blame Contribute Delete
18.5 kB
{
"checkpoint": "artifacts/partial-source-holdout-tinytcn/best.pt",
"checkpoint_sha256": "4d5286c562fa7d2ccaf3ae3da7e0e09ad370a8b9821d9fd32ed0037972457a1e",
"data_revision": "e564e2ac567f774d1880aa1db6ce97afb8c519b7",
"data_scope": "preview-only source-held-out stress: 3265 rows from train-00010-of-00083; official test remains sealed",
"dataset_revision": null,
"development_only": true,
"evaluation_source": {
"bytes": 2962445,
"dataset_id": null,
"kind": "file",
"path": "data/processed/partial-source-holdout.jsonl",
"revision": "e564e2ac567f774d1880aa1db6ce97afb8c519b7",
"sha256": "ab7e65eba9a8130fb94549a9f4ae106169a4db9c9465c3dd2dd38f7f4375e85d"
},
"freeze_manifest_sha256": null,
"grouped_bootstrap_95ci": {
"false_positive_rate": {
"estimate": 0.018461538461538463,
"lower": 0.0,
"upper": 0.046249495899986545
},
"recall": {
"estimate": 0.04024767801857585,
"lower": 0.02043705778301887,
"upper": 0.0955957896857373
}
},
"metrics": {
"accuracy": 0.5123456790123457,
"average_precision": 0.5499581204674366,
"balanced_accuracy": 0.5108930697785187,
"brier_score": 0.28202905638850445,
"count": 648,
"expected_calibration_error": 0.16490815351084795,
"f1": 0.07602339181286549,
"false_negative_rate": 0.9597523219814241,
"false_positive_rate": 0.018461538461538463,
"fn": 310,
"fp": 6,
"log_loss": 0.7782068838177573,
"macro_f1": 0.37239324726911616,
"negative_count": 325,
"positive_count": 323,
"precision": 0.6842105263157895,
"recall": 0.04024767801857585,
"roc_auc": 0.5522743510359609,
"specificity": 0.9815384615384616,
"threshold": 0.7510702013969421,
"tn": 319,
"tp": 13
},
"official_test": false,
"operating_points": {
"0.01": {
"accuracy": 0.5138888888888888,
"average_precision": 0.5499581204674366,
"balanced_accuracy": 0.5124124791617052,
"brier_score": 0.28202905638850445,
"count": 648,
"expected_calibration_error": 0.16490815351084795,
"f1": 0.06528189910979229,
"false_negative_rate": 0.9659442724458205,
"false_positive_rate": 0.009230769230769232,
"fn": 312,
"fp": 3,
"log_loss": 0.7782068838177573,
"macro_f1": 0.36840737291256037,
"max_false_positive_rate": 0.01,
"negative_count": 325,
"positive_count": 323,
"precision": 0.7857142857142857,
"recall": 0.034055727554179564,
"roc_auc": 0.5522743510359609,
"specificity": 0.9907692307692307,
"threshold": 0.755566418170929,
"tn": 322,
"tp": 11
},
"0.02": {
"accuracy": 0.5123456790123457,
"average_precision": 0.5499581204674366,
"balanced_accuracy": 0.5108930697785187,
"brier_score": 0.28202905638850445,
"count": 648,
"expected_calibration_error": 0.16490815351084795,
"f1": 0.07602339181286549,
"false_negative_rate": 0.9597523219814241,
"false_positive_rate": 0.018461538461538463,
"fn": 310,
"fp": 6,
"log_loss": 0.7782068838177573,
"macro_f1": 0.37239324726911616,
"max_false_positive_rate": 0.02,
"negative_count": 325,
"positive_count": 323,
"precision": 0.6842105263157895,
"recall": 0.04024767801857585,
"roc_auc": 0.5522743510359609,
"specificity": 0.9815384615384616,
"threshold": 0.7510702013969421,
"tn": 319,
"tp": 13
},
"0.05": {
"accuracy": 0.5185185185185185,
"average_precision": 0.5499581204674366,
"balanced_accuracy": 0.5171802810192904,
"brier_score": 0.28202905638850445,
"count": 648,
"expected_calibration_error": 0.16490815351084795,
"f1": 0.14754098360655737,
"false_negative_rate": 0.9164086687306502,
"false_positive_rate": 0.04923076923076923,
"fn": 296,
"fp": 16,
"log_loss": 0.7782068838177573,
"macro_f1": 0.4060285563194077,
"max_false_positive_rate": 0.05,
"negative_count": 325,
"positive_count": 323,
"precision": 0.627906976744186,
"recall": 0.08359133126934984,
"roc_auc": 0.5522743510359609,
"specificity": 0.9507692307692308,
"threshold": 0.727832555770874,
"tn": 309,
"tp": 27
}
},
"operating_points_note": "Label-dependent threshold sweep for curve analysis only. The deployed result is `metrics` at the checkpoint's frozen validation-selected threshold.",
"operational": {
"accuracy": 0.5123456790123457,
"audio_hours": 1.552732013888889,
"average_precision": 0.5499581204674366,
"balanced_accuracy": 0.5108930697785187,
"brier_score": 0.28202905638850445,
"count": 648,
"expected_calibration_error": 0.16490815351084795,
"f1": 0.07602339181286549,
"false_interruptions": 6,
"false_interruptions_per_hour": 3.8641568192908724,
"false_negative_rate": 0.9597523219814241,
"false_positive_rate": 0.018461538461538463,
"fn": 310,
"fp": 6,
"log_loss": 0.7782068838177573,
"macro_f1": 0.37239324726911616,
"negative_count": 325,
"positive_count": 323,
"precision": 0.6842105263157895,
"recall": 0.04024767801857585,
"roc_auc": 0.5522743510359609,
"scope": "independent labeled clips; per-hour rate is a clip-normalized proxy, not an online conversation measurement",
"sequence_metrics_available": false,
"sequence_metrics_unavailable_reason": "source records contain no genuine turn_id or conversation_id",
"specificity": 0.9815384615384616,
"threshold": 0.7510702013969421,
"tn": 319,
"tp": 13
},
"reliability": [
{
"count": 32,
"lower": 0.0,
"mean_probability": 0.08213151805102825,
"observed_rate": 0.21875,
"upper": 0.1
},
{
"count": 116,
"lower": 0.1,
"mean_probability": 0.15063592823671884,
"observed_rate": 0.5,
"upper": 0.2
},
{
"count": 84,
"lower": 0.2,
"mean_probability": 0.25053229023303303,
"observed_rate": 0.5,
"upper": 0.3
},
{
"count": 73,
"lower": 0.3,
"mean_probability": 0.3514139423631642,
"observed_rate": 0.4931506849315068,
"upper": 0.4
},
{
"count": 102,
"lower": 0.4,
"mean_probability": 0.4513314664363861,
"observed_rate": 0.5196078431372549,
"upper": 0.5
},
{
"count": 108,
"lower": 0.5,
"mean_probability": 0.5456479418056982,
"observed_rate": 0.48148148148148145,
"upper": 0.6
},
{
"count": 73,
"lower": 0.6,
"mean_probability": 0.6492549148324418,
"observed_rate": 0.547945205479452,
"upper": 0.7
},
{
"count": 60,
"lower": 0.7,
"mean_probability": 0.7395102163155873,
"observed_rate": 0.5833333333333334,
"upper": 0.8
},
{
"count": 0,
"lower": 0.8,
"mean_probability": 0.0,
"observed_rate": 0.0,
"upper": 0.9
},
{
"count": 0,
"lower": 0.9,
"mean_probability": 0.0,
"observed_rate": 0.0,
"upper": 1.0
}
],
"slices": {
"dataset": {
"human_5": {
"accuracy": 0.5319148936170213,
"average_precision": 0.7346027845874109,
"balanced_accuracy": 0.56,
"brier_score": 0.2237639915125358,
"count": 47,
"expected_calibration_error": 0.11744399448024466,
"f1": 0.21428571428571427,
"false_negative_rate": 0.88,
"false_positive_rate": 0.0,
"fn": 22,
"fp": 0,
"log_loss": 0.6315959169824472,
"macro_f1": 0.44047619047619047,
"negative_count": 22,
"positive_count": 25,
"precision": 1.0,
"recall": 0.12,
"roc_auc": 0.6727272727272727,
"specificity": 1.0,
"threshold": 0.7510702013969421,
"tn": 22,
"tp": 3
},
"liva_1": {
"accuracy": 0.511049723756906,
"average_precision": 0.534675734736703,
"balanced_accuracy": 0.4976173020527859,
"brier_score": 0.28168741657223345,
"count": 362,
"expected_calibration_error": 0.1480350368836308,
"f1": 0.022099447513812154,
"false_negative_rate": 0.9886363636363636,
"false_positive_rate": 0.016129032258064516,
"fn": 174,
"fp": 3,
"log_loss": 0.7808010075770082,
"macro_f1": 0.34806629834254144,
"negative_count": 186,
"positive_count": 176,
"precision": 0.4,
"recall": 0.011363636363636364,
"roc_auc": 0.5610031769305963,
"specificity": 0.9838709677419355,
"threshold": 0.7510702013969421,
"tn": 183,
"tp": 2
},
"midcentury_1": {
"accuracy": 0.5204081632653061,
"average_precision": 0.49467764090184185,
"balanced_accuracy": 0.53,
"brier_score": 0.33650612559098597,
"count": 98,
"expected_calibration_error": 0.33408278599381447,
"f1": 0.11320754716981132,
"false_negative_rate": 0.94,
"false_positive_rate": 0.0,
"fn": 47,
"fp": 0,
"log_loss": 0.8977698354532542,
"macro_f1": 0.3922681092492413,
"negative_count": 48,
"positive_count": 50,
"precision": 1.0,
"recall": 0.06,
"roc_auc": 0.36125,
"specificity": 1.0,
"threshold": 0.7510702013969421,
"tn": 48,
"tp": 3
},
"mundo_1": {
"accuracy": 0.5581395348837209,
"average_precision": 0.43688015249152684,
"balanced_accuracy": 0.5054824561403509,
"brier_score": 0.31313111102791236,
"count": 43,
"expected_calibration_error": 0.2613871014395425,
"f1": 0.09523809523809523,
"false_negative_rate": 0.9473684210526315,
"false_positive_rate": 0.041666666666666664,
"fn": 18,
"fp": 1,
"log_loss": 0.835991683286482,
"macro_f1": 0.4014652014652015,
"negative_count": 24,
"positive_count": 19,
"precision": 0.5,
"recall": 0.05263157894736842,
"roc_auc": 0.43859649122807015,
"specificity": 0.9583333333333334,
"threshold": 0.7510702013969421,
"tn": 23,
"tp": 1
},
"rime_2": {
"accuracy": 0.46511627906976744,
"average_precision": 0.5809101598141188,
"balanced_accuracy": 0.4751082251082251,
"brier_score": 0.3177046643212644,
"count": 43,
"expected_calibration_error": 0.25279066001260003,
"f1": 0.08,
"false_negative_rate": 0.9545454545454546,
"false_positive_rate": 0.09523809523809523,
"fn": 21,
"fp": 2,
"log_loss": 0.8849156531870208,
"macro_f1": 0.35147540983606557,
"negative_count": 21,
"positive_count": 22,
"precision": 0.3333333333333333,
"recall": 0.045454545454545456,
"roc_auc": 0.5584415584415584,
"specificity": 0.9047619047619048,
"threshold": 0.7510702013969421,
"tn": 19,
"tp": 1
}
},
"duration_bin": {
"2-4s": {
"accuracy": 0.5210084033613446,
"average_precision": 0.6060288083109266,
"balanced_accuracy": 0.5382003395585738,
"brier_score": 0.27422464350587894,
"count": 119,
"expected_calibration_error": 0.17818899408859362,
"f1": 0.2191780821917808,
"false_negative_rate": 0.8709677419354839,
"false_positive_rate": 0.05263157894736842,
"fn": 54,
"fp": 3,
"log_loss": 0.7509756278070072,
"macro_f1": 0.4368617683686177,
"negative_count": 57,
"positive_count": 62,
"precision": 0.7272727272727273,
"recall": 0.12903225806451613,
"roc_auc": 0.5464063384267119,
"specificity": 0.9473684210526315,
"threshold": 0.7510702013969421,
"tn": 54,
"tp": 8
},
"4-8s": {
"accuracy": 0.5058823529411764,
"average_precision": 0.5215388633267756,
"balanced_accuracy": 0.5058823529411764,
"brier_score": 0.29757245936805765,
"count": 170,
"expected_calibration_error": 0.22203312945716525,
"f1": 0.045454545454545456,
"false_negative_rate": 0.9764705882352941,
"false_positive_rate": 0.011764705882352941,
"fn": 83,
"fp": 1,
"log_loss": 0.8124832159457299,
"macro_f1": 0.356060606060606,
"negative_count": 85,
"positive_count": 85,
"precision": 0.6666666666666666,
"recall": 0.023529411764705882,
"roc_auc": 0.5038062283737024,
"specificity": 0.9882352941176471,
"threshold": 0.7510702013969421,
"tn": 84,
"tp": 2
},
">=8s": {
"accuracy": 0.5115606936416185,
"average_precision": 0.5595854046358658,
"balanced_accuracy": 0.5002674422491893,
"brier_score": 0.27696240306064007,
"count": 346,
"expected_calibration_error": 0.14117037789153225,
"f1": 0.023121387283236993,
"false_negative_rate": 0.9881656804733728,
"false_positive_rate": 0.011299435028248588,
"fn": 167,
"fp": 2,
"log_loss": 0.7709613916460462,
"macro_f1": 0.348747591522158,
"negative_count": 177,
"positive_count": 169,
"precision": 0.5,
"recall": 0.011834319526627219,
"roc_auc": 0.5790124694948684,
"specificity": 0.9887005649717514,
"threshold": 0.7510702013969421,
"tn": 175,
"tp": 2
}
},
"filler_type": {
"unknown": {
"accuracy": 0.5117370892018779,
"average_precision": 0.5459168650210282,
"balanced_accuracy": 0.5095221301357785,
"brier_score": 0.28353119603052307,
"count": 639,
"expected_calibration_error": 0.17034158953618558,
"f1": 0.07142857142857142,
"false_negative_rate": 0.9622641509433962,
"false_positive_rate": 0.018691588785046728,
"fn": 306,
"fp": 6,
"log_loss": 0.7816831013535286,
"macro_f1": 0.3701091901728844,
"negative_count": 321,
"positive_count": 318,
"precision": 0.6666666666666666,
"recall": 0.03773584905660377,
"roc_auc": 0.5477869864221477,
"specificity": 0.9813084112149533,
"threshold": 0.7510702013969421,
"tn": 315,
"tp": 12
}
},
"language": {
"eng": {
"accuracy": 0.509090909090909,
"average_precision": 0.5689565954839015,
"balanced_accuracy": 0.5114311942647316,
"brier_score": 0.27981849713314155,
"count": 605,
"expected_calibration_error": 0.1584163401860836,
"f1": 0.07476635514018691,
"false_negative_rate": 0.9605263157894737,
"false_positive_rate": 0.016611295681063124,
"fn": 292,
"fp": 5,
"log_loss": 0.774099865012542,
"macro_f1": 0.37034155777256816,
"negative_count": 301,
"positive_count": 304,
"precision": 0.7058823529411765,
"recall": 0.039473684210526314,
"roc_auc": 0.5654288337121874,
"specificity": 0.9833887043189369,
"threshold": 0.7510702013969421,
"tn": 296,
"tp": 12
},
"spa": {
"accuracy": 0.5581395348837209,
"average_precision": 0.43688015249152684,
"balanced_accuracy": 0.5054824561403509,
"brier_score": 0.31313111102791236,
"count": 43,
"expected_calibration_error": 0.2613871014395425,
"f1": 0.09523809523809523,
"false_negative_rate": 0.9473684210526315,
"false_positive_rate": 0.041666666666666664,
"fn": 18,
"fp": 1,
"log_loss": 0.835991683286482,
"macro_f1": 0.4014652014652015,
"negative_count": 24,
"positive_count": 19,
"precision": 0.5,
"recall": 0.05263157894736842,
"roc_auc": 0.43859649122807015,
"specificity": 0.9583333333333334,
"threshold": 0.7510702013969421,
"tn": 23,
"tp": 1
}
},
"synthetic": {
"False": {
"accuracy": 0.5181818181818182,
"average_precision": 0.5245109263654216,
"balanced_accuracy": 0.5095238095238096,
"brier_score": 0.28896361905882645,
"count": 550,
"expected_calibration_error": 0.18397142849185252,
"f1": 0.0636042402826855,
"false_negative_rate": 0.9666666666666667,
"false_positive_rate": 0.014285714285714285,
"fn": 261,
"fp": 4,
"log_loss": 0.7932073801759811,
"macro_f1": 0.36962341757096334,
"negative_count": 280,
"positive_count": 270,
"precision": 0.6923076923076923,
"recall": 0.03333333333333333,
"roc_auc": 0.5231084656084656,
"specificity": 0.9857142857142858,
"threshold": 0.7510702013969421,
"tn": 276,
"tp": 9
},
"True": {
"accuracy": 0.47959183673469385,
"average_precision": 0.6790413484001439,
"balanced_accuracy": 0.5155136268343816,
"brier_score": 0.24311059242241156,
"count": 98,
"expected_calibration_error": 0.15580201186999984,
"f1": 0.13559322033898305,
"false_negative_rate": 0.9245283018867925,
"false_positive_rate": 0.044444444444444446,
"fn": 49,
"fp": 2,
"log_loss": 0.6940204246644598,
"macro_f1": 0.38166522330817765,
"negative_count": 45,
"positive_count": 53,
"precision": 0.6666666666666666,
"recall": 0.07547169811320754,
"roc_auc": 0.680083857442348,
"specificity": 0.9555555555555556,
"threshold": 0.7510702013969421,
"tn": 43,
"tp": 4
}
}
},
"source_split": "validation",
"split": "validation",
"threshold": 0.7510702013969421,
"training_status": "preview-only",
"worst_groups": {
"highest_false_positive_rate": {
"count": 43,
"dimension": "dataset",
"false_positive_rate": 0.09523809523809523,
"value": "rime_2"
},
"lowest_recall": {
"count": 362,
"dimension": "dataset",
"recall": 0.011363636363636364,
"value": "liva_1"
}
}
}