Close custody main-review checks for named B0 units and coordinate scaling
This commit is contained in:
1 parent
6e4d991d78
commit
522ee84f6a
7 files changed
+233
-7
No files matched your search
@@ -0,0 +1,181 @@
|
||||
{
|
||||
"version": "S2-custody-v1",
|
||||
"execution_approved": true,
|
||||
"ruling": "1794729482.815749",
|
||||
"main_review_required_before_prediction": true,
|
||||
"roster": "artifacts/s2-main-preparation-2026-11-15/final-roster.json",
|
||||
"roster_sha256": "fb0c984f2e553343a598ce2e3c7012d3017053dc0fef58b5aaf46dd3c67a207e",
|
||||
"exposure_membership": "artifacts/s2-g1-components-2026-11-15/custodian/not-known-exposed-test-subsets.json",
|
||||
"exposure_membership_sha256": "8d037786040c6fc11fa80ceb0f41dfdd58478a2576404505a64aa29a2216a22f",
|
||||
"exposure_audit": "artifacts/s2-custody-planning-2026-11-15/exposure-audit.json",
|
||||
"models": "artifacts/s2-terminal-freeze-2026-11-15/models.json",
|
||||
"models_sha256": "cd544ec7f0e49f0759660a71076566976cacdeaa66dd510aec049b2218f57faf",
|
||||
"archive": "data.zip",
|
||||
"archive_stat": {
|
||||
"status": "official_size_and_md5_pass",
|
||||
"source": "/home/frank14f/optane/MeltMind/data.zip",
|
||||
"source_url": "https://zenodo.org/records/18598121",
|
||||
"bytes": 32757812726,
|
||||
"md5": "75ec15002234c2cdcdf4c751bdefab85",
|
||||
"sha256": "d151f018294832fd80fbd23e865642cb16c860da150f245033d010c388fa0241",
|
||||
"mtime_ns": 1794548962944204621,
|
||||
"inode": 12582980,
|
||||
"members": 1872,
|
||||
"triplets": 624,
|
||||
"source_tokens": 249,
|
||||
"max_group": 34,
|
||||
"total_uncompressed_bytes": 43966308383,
|
||||
"finished_utc": "2026-11-13T08:04:57.180660+00:00",
|
||||
"wall_seconds": 97.27923265600111,
|
||||
"peak_rss_kib": 32056,
|
||||
"crc_scope": "ZIP stored CRC inventory only; selected members checked on read; no full decompression",
|
||||
"grouping_scope": "Filename source tokens, not verified independent CAD families"
|
||||
},
|
||||
"gpu_uuids": [
|
||||
"GPU-e8853ae7-a302-6a17-bcfa-aa767e9ed20c",
|
||||
"GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36",
|
||||
"GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76",
|
||||
"GPU-8b71b032-1b75-2891-935c-ce106a53034a"
|
||||
],
|
||||
"cpu_affinities": [
|
||||
[
|
||||
0,
|
||||
1
|
||||
],
|
||||
[
|
||||
2,
|
||||
3
|
||||
],
|
||||
[
|
||||
4,
|
||||
5
|
||||
],
|
||||
[
|
||||
6,
|
||||
7
|
||||
]
|
||||
],
|
||||
"code_sha256": {
|
||||
"tools/audit_samples.py": "a3daf70ec3bcb77bf46d9dc4887d552cf7c422117880a169969780012278cd70",
|
||||
"tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
|
||||
"tools/prepare_s2_custody.py": "6a64c6d16760cf243b150c6dab67a5e255af4c70bc71b8726edbead2f2cebe16",
|
||||
"tools/prepare_s2_incident_recovery.py": "4f772573dedc2b7921986e25eaedb14f91d98b235421829c284f3b40b85710aa",
|
||||
"tools/prepare_s2_main_matrix.py": "43b48d82a813175be22be41cb63aaa5d04499000e80197c7405538eff3052ede",
|
||||
"tools/prepare_s2_precision_pilot.py": "c3a71b0ff58bcea4433ff93e18ef14d3a00521438a76baaa69830a5ee8ac098a",
|
||||
"tools/prepare_s2_r2.py": "ec6dd1c73bc2678aa4376c878837a957395c21b82e2da2314c4a9c4e96260caf",
|
||||
"tools/prepare_s2_recovered_pilot_training.py": "ab5a4274363598ff933ff873df2b00497f069b1fb45756a1d1dba8d723294f68",
|
||||
"tools/run_s2_custody_predict.py": "e5c882b3073cf6470d065757742e45ce5097fc91e03e4639bab557449805d09e",
|
||||
"tools/run_s2_custody_predictions.py": "3589c64b09db4f83da7df99faf8c1aa86db17b08df86e6e1189fdffd71dae7cf",
|
||||
"tools/run_s2_custody_score.py": "4e9192edcefaa5db207f52400d0ac79345095077b74947ddb86010f32a64bd88",
|
||||
"tools/run_s2_encoder_diagnostic.py": "8f31d312fb53bfdb9b1804af064e463d3989f1b33c4aa7279a967ca82dd2c0a4",
|
||||
"tools/run_s2_main_matrix.py": "cebd3ccdea80c2914f6bbc1c3de293870535689f224c547c656a860b64af14da",
|
||||
"tools/run_s2_main_training.py": "7fb18f015ab2eaa4968ff50c8b140285f9db6eb822398b2789d6de40f1aea2e7",
|
||||
"tools/run_s2_pilot_probe.py": "2360ab7b44b179144a5ecd2d691e33cd36a6ec4bfeac71452bf062e7f42f81eb",
|
||||
"tools/run_s2_pilot_probe_v2.py": "61c08499346d50405d0c254de54deea24b3522256bf1c8b0aec3e9f4ddab6390",
|
||||
"tools/run_s2_pilot_training.py": "d5e543478543243fe1db3475a4d4cd135fbdcf631d3bce48c4b5af8b2edafb22",
|
||||
"tools/run_s2_pilot_training_v2.py": "f461cb43bc602a32677d7dcfb8a322e206531ad152fa33bb4e36ea8546a1c0f2",
|
||||
"tools/run_s2_precision_continuation.py": "9fb5a22007788f0c18c77a82b6a8b76fbb53a98b49cc9251bc49a7a5b8312f2f",
|
||||
"tools/run_s2_precision_pilot.py": "ac8347be1ca11f48cb67abbf5e5c78b39af46ed3a7c6f89b49098fd5e77a6904",
|
||||
"tools/run_s2_r2.py": "7a93e61cef18f12f5e834bcfdfac81d2092bb74a7216221dd542d5904e71905e",
|
||||
"tools/run_s2_r2_fit.py": "dd235ef8b015a58deaa89d36208c563bea425b9c4d0a0937ed6e0248f23618f8",
|
||||
"tools/run_s2_recovered_pilot.py": "56962009bf5eaaee599ffea6a3d454372e0aea2842fa79eac609fd4a24ecbb39",
|
||||
"tools/run_s2_recovery_chain.py": "76d0808bdb2d624cfbaadf5a3c5e7e9a0c2be5d5ac41d4ac3fd9ccd48743933c",
|
||||
"tools/s2_custody_common.py": "bfa651ce5f7f2b0b74e7099ca90b6a53340c88d0a6d3ac3ddf028055d023e4ec",
|
||||
"tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"tools/s2_e1_supervisor.py": "6753c9ceb0b932047ba505bd15681f109e9407bdb58fd96c55e1431a3f1b4a02",
|
||||
"tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
"tools/s2_incident_recovery_manifest.py": "8cbb2e1067c2cc22762d8ae3cf192aa2c5fa7c7a199a22e7a39a25570ee1c3b1",
|
||||
"tools/s2_main_data.py": "bda31f5e9335b927315292f57e6f689787eacc5b32f7fae3fa23eb13b00343f0",
|
||||
"tools/s2_main_memory_guard.py": "e094f5443e11518efe9018215c11718a8e18aa462a977bf43e974b2d0b2aebb1",
|
||||
"tools/s2_models/__init__.py": "a0fa1323d3553172da66c5d442c5fc4fcb0d5055c886c776928604550ea97609",
|
||||
"tools/s2_models/attention.py": "7714adfccf6a0fc6b52a44aee3a50321a88898bf3d743680a1108c01e1ff8a61",
|
||||
"tools/s2_models/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
|
||||
"tools/s2_models/network.py": "5137e845d81d1223afc13bf87e03ccba1132a3baa75d5189a8f320b2752be071",
|
||||
"tools/s2_models/normalization.py": "7180696d3845799e28d8c31ddef957d075de6a2e46d96a3a0f0df16fc05234a0",
|
||||
"tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
|
||||
"tools/s2_models_v2/__init__.py": "0302d333bcd8b33e85aaa6a4e7822e7e1ccfc66343136107fea6f4ca283ba1ec",
|
||||
"tools/s2_models_v2/attention.py": "3829e34ca30942c863699c4d43fb6fa104452c6c04c4d606c11019d3dde16ffa",
|
||||
"tools/s2_models_v2/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"tools/s2_models_v2/network.py": "a443470caeecdf529504bffab3ca8cc0b70ce44b505bad60f9c63db9709dbb83",
|
||||
"tools/s2_pilot_data.py": "82c1fd18ec51757479c8fd7dda2db4d5f38c366d77fedcca3216a08998214c97",
|
||||
"tools/s2_pilot_train.py": "e9cceead51a5cc092aad23409ad98909c84a62e7e10571e750d1e9be1ad3b948",
|
||||
"tools/s2_precision_continuation_guard.py": "01235d897c7c03e5ffa109c58b14f5d24d128e8f1ba873e1f76cfb14e02f7a7f",
|
||||
"tools/s2_precision_pilot_gpu_guard.py": "5cf0f41de34b28394dbee4672b5bd34eb9c4fe2344822aa3d648f20561d96328",
|
||||
"tools/s2_r2_data.py": "a32f8987ff0848c0e052e71500f84bd4f0364a60170121f54770ceb275a98566",
|
||||
"tools/s2_recovered_pilot_gpu_guard.py": "9a27f47918c865ce37e46b0be54144dc3f9ab1058e4eddc9f6fbc771f4959eb0",
|
||||
"tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"tools/s2_recovery_supervisor.py": "f563ca3c2665f5baf0c04f296616e4640423a59b815fbb00479af0f29e57a919",
|
||||
"tools/test_s2_custody.py": "362d63e12ccd88f168e8d58aabc71f6b26c3a6992e7f88d6b6ecdfc071f75158",
|
||||
"tools/test_s2_main_matrix.py": "d9b05506dd6e5a8a94cccaee28c6d8239bb3c6cdbe31567be1fdd04559154feb",
|
||||
"tools/test_s2_r2.py": "01d214e6e3ed04b099c30a4567521c37c4cafb0412faa83c973110c8dd600418"
|
||||
},
|
||||
"source_sha256": {
|
||||
"artifacts/s2-terminal-freeze-2026-11-15/terminal-audit.json": "ddd32cd6cb12bc1c893d8aad4928b570f9e2162801ba5e219cb0708f82cb9465",
|
||||
"artifacts/s2-terminal-freeze-2026-11-15/models.json": "cd544ec7f0e49f0759660a71076566976cacdeaa66dd510aec049b2218f57faf",
|
||||
"artifacts/s2-terminal-freeze-2026-11-15/evidence-manifest.json": "34b0aa29f8e8ced0ef11ea2d8889f243f1864e7eab8ebb1015082f3b787542b2",
|
||||
"artifacts/s2-main-matrix-2026-11-15/training-contract.json": "198e5711ed2e7485167068769652c866867229e09ecf0d09b21612debc3b7a56",
|
||||
"artifacts/s2-r2-main-2026-11-15/contract.json": "edec86197769f44bf98f234795950f36c9fa42084ccb3c68448b63b440aa126e",
|
||||
"artifacts/s2-main-preparation-2026-11-15/final-roster.json": "fb0c984f2e553343a598ce2e3c7012d3017053dc0fef58b5aaf46dd3c67a207e",
|
||||
"artifacts/s2-g1-components-2026-11-15/custodian/not-known-exposed-test-subsets.json": "8d037786040c6fc11fa80ceb0f41dfdd58478a2576404505a64aa29a2216a22f",
|
||||
"artifacts/s2-custody-planning-2026-11-15/exposure-audit.json": "28255a3cf04cf1f651aa7aef12e613cbd11531482ad0bc5730096af79f8f0fc0"
|
||||
},
|
||||
"limits": {
|
||||
"gpu_seconds": 14400,
|
||||
"fold_wall_seconds": 2700,
|
||||
"vram_bytes": 16106127360,
|
||||
"host_bytes": 17179869184,
|
||||
"fold_disk_bytes": 21474836480,
|
||||
"disk_bytes": 107374182400,
|
||||
"custody_cpu_seconds": 36000,
|
||||
"custody_wall_seconds": 10800,
|
||||
"retries": 0
|
||||
},
|
||||
"statistics": {
|
||||
"replicates": 2000,
|
||||
"seed": 2026111408,
|
||||
"margin": 0.1,
|
||||
"percentile_method": "linear",
|
||||
"aggregation": "parts equal within component, fixed seeds equal, pool components across folds",
|
||||
"complete_pair": "all parts and all three fixed seeds in both arms; no backfill",
|
||||
"exposure": "frozen522 membership inclUNKNOWN intersect roster intersect main validity; bool only cross-check",
|
||||
"bootstrap_rng": "PCG64(SeedSequence(2026111408)); sorted component IDs, pooled not fold-stratified",
|
||||
"undefined_replicates": "count and record, no replacement; interval left undefined if present",
|
||||
"secondary": [
|
||||
"near_wall",
|
||||
"kendall_tau_b",
|
||||
"last_location",
|
||||
"worst_component_per_arm"
|
||||
],
|
||||
"primary_pool": 100,
|
||||
"primary_contrast": "1-E/N",
|
||||
"kendall_seed": 2026111410,
|
||||
"kendall_pairs": 5000
|
||||
},
|
||||
"planned_neural_cells": 90,
|
||||
"valid_neural_models": 89,
|
||||
"NA_neural": [
|
||||
{
|
||||
"index": 24,
|
||||
"name": "fold1-pool50-N128-seed0",
|
||||
"arm": "N",
|
||||
"status": "NA",
|
||||
"reason": "WORKER_EXIT_1",
|
||||
"guard_receipt": "artifacts/s2-main-matrix-2026-11-15/cells/fold1-pool50-N128-seed0/fold1-pool50-N128-seed0-guard-receipt.json",
|
||||
"guard_sha256": "ff550868e361c18fc4d04ea5ce1f205c030b22f521b79892c64bfa09ac6c50b4",
|
||||
"worker_status": "STOPPED_OR_FAILED",
|
||||
"worker_returncode": 1,
|
||||
"worker_cpu_seconds": 188.163453,
|
||||
"checkpoint_sha256": null,
|
||||
"standardizer_fingerprint": null,
|
||||
"cell_wall_seconds": 211.4663818500121
|
||||
}
|
||||
],
|
||||
"R2_models": 15,
|
||||
"no_clipping": true,
|
||||
"inference_precision": "FP32 same as validation",
|
||||
"target": "T/t_set",
|
||||
"test_label_gate": "All predictions hash-frozen, scoring code/config immutable Git commit and explicit main review receipt",
|
||||
"labels_decoded": false
|
||||
}
|
||||
@@ -64,7 +64,7 @@
|
||||
"tools/prepare_s2_precision_pilot.py": "c3a71b0ff58bcea4433ff93e18ef14d3a00521438a76baaa69830a5ee8ac098a",
|
||||
"tools/prepare_s2_r2.py": "ec6dd1c73bc2678aa4376c878837a957395c21b82e2da2314c4a9c4e96260caf",
|
||||
"tools/prepare_s2_recovered_pilot_training.py": "ab5a4274363598ff933ff873df2b00497f069b1fb45756a1d1dba8d723294f68",
|
||||
"tools/run_s2_custody_predict.py": "e5c882b3073cf6470d065757742e45ce5097fc91e03e4639bab557449805d09e",
|
||||
"tools/run_s2_custody_predict.py": "2dd0c654778ea2db646a6fe7e75b00d5bcd5adf59c3c8d107cbe6a6870448117",
|
||||
"tools/run_s2_custody_predictions.py": "3589c64b09db4f83da7df99faf8c1aa86db17b08df86e6e1189fdffd71dae7cf",
|
||||
"tools/run_s2_custody_score.py": "4e9192edcefaa5db207f52400d0ac79345095077b74947ddb86010f32a64bd88",
|
||||
"tools/run_s2_encoder_diagnostic.py": "8f31d312fb53bfdb9b1804af064e463d3989f1b33c4aa7279a967ca82dd2c0a4",
|
||||
@@ -80,7 +80,7 @@
|
||||
"tools/run_s2_r2_fit.py": "dd235ef8b015a58deaa89d36208c563bea425b9c4d0a0937ed6e0248f23618f8",
|
||||
"tools/run_s2_recovered_pilot.py": "56962009bf5eaaee599ffea6a3d454372e0aea2842fa79eac609fd4a24ecbb39",
|
||||
"tools/run_s2_recovery_chain.py": "76d0808bdb2d624cfbaadf5a3c5e7e9a0c2be5d5ac41d4ac3fd9ccd48743933c",
|
||||
"tools/s2_custody_common.py": "bfa651ce5f7f2b0b74e7099ca90b6a53340c88d0a6d3ac3ddf028055d023e4ec",
|
||||
"tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
|
||||
"tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"tools/s2_e1_supervisor.py": "6753c9ceb0b932047ba505bd15681f109e9407bdb58fd96c55e1431a3f1b4a02",
|
||||
"tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
@@ -107,7 +107,7 @@
|
||||
"tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"tools/s2_recovery_supervisor.py": "f563ca3c2665f5baf0c04f296616e4640423a59b815fbb00479af0f29e57a919",
|
||||
"tools/test_s2_custody.py": "362d63e12ccd88f168e8d58aabc71f6b26c3a6992e7f88d6b6ecdfc071f75158",
|
||||
"tools/test_s2_custody.py": "c1c6fcfac7fd50df4dcd47888127b65d6af5e7c6a9415d8b6530f101733a2130",
|
||||
"tools/test_s2_main_matrix.py": "d9b05506dd6e5a8a94cccaee28c6d8239bb3c6cdbe31567be1fdd04559154feb",
|
||||
"tools/test_s2_r2.py": "01d214e6e3ed04b099c30a4567521c37c4cafb0412faa83c973110c8dd600418"
|
||||
},
|
||||
@@ -177,5 +177,7 @@
|
||||
"inference_precision": "FP32 same as validation",
|
||||
"target": "T/t_set",
|
||||
"test_label_gate": "All predictions hash-frozen, scoring code/config immutable Git commit and explicit main review receipt",
|
||||
"labels_decoded": false
|
||||
"labels_decoded": false,
|
||||
"main_review_ruling": "1794730198.146459",
|
||||
"review_conditions": "named Vf range/units and centered xyz/Rg synthetic checks passed before predictions"
|
||||
}
|
||||
@@ -0,0 +1,14 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"tests": 14,
|
||||
"real_labels_read": false,
|
||||
"real_models_evaluated": false,
|
||||
"gpu_used": false,
|
||||
"review_ruling": "1794730198.146459",
|
||||
"checks": [
|
||||
"R2 column0 named Vf in prepared metadata and source",
|
||||
"Vf in[0,1],seconds=Vf*t_set",
|
||||
"base columns named x_star/y_star/z_star,synthetic proves(x-center)/Rg",
|
||||
"wrong column rejected before predictions"
|
||||
]
|
||||
}
|
||||
@@ -35,3 +35,7 @@ Bootstrap单一全有效组件抽样框、不按折分层;component ID排序
|
||||
主100%池Delta=1-E/N,m=0.1、六区间;N/R2、E/R2及B0同表。25/50%学习曲线明确是组件比例并列每折实际件数/节点数;NA按pool/arm;617 included、7 excluded及节点份额,覆盖表逐件,query-only声明。宽128可行性、未测试宽度效应、AMP跳步case未记、R2第二进程身份未记、全库历史筛查、固定fold/seed条件化bootstrap均保留。
|
||||
|
||||
合成验收应包含:预测读入无标签、失败权重拒绝、模型/合同/预测哈希变化拒绝、预测缺件不解码、计数权重与组件权重、三seed完整配对、六区间边界、ties与最小ID、近壁空域、主arm排序和零范数。通过后固定代码再进入真实预测;解码更须全部预测冻结及最终统计裁量。
|
||||
|
||||
## 主审通过及两个条件关闭
|
||||
|
||||
Claude1794730198.146459实读固定6e4d991d主审通过。预测前补上prepare元数据及源码R2_COLUMNS[0]==Vf的按名断言、范围[0,1]与B0秒值Vf*t_set;base前三列按名确认,synthetic明确(x-center)/Rg,末到距离正好除Rg。14项验收PASS,无真实预测/标签。原待审合同保存contract-before-main-review.json;执行合同仅按审查改动更新source pins,尚无started标记。
|
||||
@@ -8,6 +8,8 @@ import numpy as np
|
||||
from s2_custody_common import contract,checked,test_rows,prediction_inputs,expected_models,prediction_path
|
||||
from s2_recovery_supervisor import save
|
||||
from freeze_g1_components import sha
|
||||
from s2_models.features import R2_COLUMNS
|
||||
from s2_models.objectives import b0_seconds
|
||||
|
||||
|
||||
def load_neural(m,device):
|
||||
@@ -50,8 +52,10 @@ def run(root,fold):
|
||||
with torch.inference_mode(),torch.autocast('cuda',enabled=False):
|
||||
y=loaded[m['name']](tensors['base'],tensors['ids'],tensors['gaps'],tensors['edge'] if m['arm']=='E' else None).cpu().numpy().astype(np.float64)
|
||||
elif m['kind']=='R2':y=loaded[m['name']].predict(d['x'])
|
||||
else:y=d['x'][:,0].copy()
|
||||
seconds=y*d['t_set']
|
||||
else:
|
||||
assert R2_COLUMNS[0]=='Vf'
|
||||
y=d['x'][:,R2_COLUMNS.index('Vf')].copy()
|
||||
seconds=b0_seconds(y,d['t_set']) if m['kind']=='B0' else y*d['t_set']
|
||||
if seconds.shape!=d['ids'].shape or not np.isfinite(seconds).all():raise ValueError('Invalid prediction '+m['name'])
|
||||
p=prediction_path(root,m,row['case']);p.parent.mkdir(parents=True,exist_ok=True)
|
||||
if p.exists():raise FileExistsError('No prediction overwrite/retry')
|
||||
|
||||
@@ -3,6 +3,7 @@ import json
|
||||
from pathlib import Path
|
||||
import numpy as np
|
||||
from freeze_g1_components import sha
|
||||
from s2_models.features import BASE_COLUMNS,R2_COLUMNS
|
||||
|
||||
|
||||
def read_json(path):
|
||||
@@ -29,6 +30,11 @@ def test_rows(c, fold):
|
||||
def prediction_inputs(row):
|
||||
"""Only base/queries/R2 features: no training-cache or archive access."""
|
||||
prep=row['preparation'];root=Path(prep['preparation_root'])
|
||||
metadata=read_json(checked(root/'result.json',prep['result_sha256']))
|
||||
if metadata['base_columns']!=list(BASE_COLUMNS) or tuple(BASE_COLUMNS[:3])!=('x_star','y_star','z_star'):
|
||||
raise ValueError('Base coordinate column names differ from centred (x-center)/Rg contract')
|
||||
if metadata['r2']['columns']!=list(R2_COLUMNS) or R2_COLUMNS[0]!='Vf':
|
||||
raise ValueError('B0 requires named dimensionless relative flow volume Vf at column0')
|
||||
for name in ['base.npy','queries.npz','r2.npz']:
|
||||
checked(root/name,prep['output_sha256'][name])
|
||||
base=np.load(root/'base.npy',allow_pickle=False)
|
||||
@@ -40,6 +46,7 @@ def prediction_inputs(row):
|
||||
if base.shape!=(row['nodes'],10) or x.shape!=(len(ids),20) or not len(ids): raise ValueError('Feature/query shape')
|
||||
if not all(np.isfinite(a).all() for a in [base,x,d['gaps'],d['edge']]):raise ValueError('Nonfinite feature')
|
||||
if not np.all(x[:,-1]==prep['t_set']):raise ValueError('Process setting mismatch')
|
||||
if np.any((x[:,0]<0)|(x[:,0]>1)):raise ValueError('Dimensionless relative flow volume outside [0,1]')
|
||||
if np.any(np.diff(d['raw_ids'])<=0):raise ValueError('Ascending unique original IDs required')
|
||||
return dict(**d,base=base,x=x,coordinates_over_Rg=base[ids,:3],t_set=prep['t_set'])
|
||||
|
||||
|
||||
@@ -9,6 +9,8 @@ from freeze_g1_components import sha
|
||||
from s2_custody_common import checked,prediction_inputs,verify_prediction_index
|
||||
from s2_custody_statistics import six_region,kendall_pairs,last_location,part_metrics,component_values,contrast
|
||||
from run_s2_custody_score import decode_gate
|
||||
from s2_models.features import BASE_COLUMNS,R2_COLUMNS,base_features
|
||||
from s2_models.objectives import b0_seconds
|
||||
|
||||
|
||||
def roster():
|
||||
@@ -91,12 +93,24 @@ class Acceptance(unittest.TestCase):
|
||||
np.save(root/'base.npy',base);np.savez(root/'r2.npz',features=x)
|
||||
np.savez(root/'queries.npz',eligible_s2=[True,False,True],anchor_ids=[0,1,2],raw_node_ids=[10,11,12],
|
||||
multiplicity=[2,1,3],g=np.ones((3,2)),edge_summary=np.ones((3,4)),near_wall=[True,False,False])
|
||||
(root/'result.json').write_text(json.dumps(dict(base_columns=list(BASE_COLUMNS),r2=dict(columns=list(R2_COLUMNS)))))
|
||||
row=dict(nodes=3,preparation=dict(preparation_root=str(root),t_set=.4,
|
||||
output_sha256={p.name:sha(p) for p in root.iterdir()}))
|
||||
result_sha256=sha(root/'result.json'),output_sha256={p.name:sha(p) for p in root.iterdir()}))
|
||||
original=np.load;paths=[]
|
||||
def audit(path,**kwargs):paths.append(Path(path).name);return original(path,**kwargs)
|
||||
with patch('s2_custody_common.np.load',side_effect=audit):d=prediction_inputs(row)
|
||||
self.assertEqual(set(paths),{'base.npy','r2.npz','queries.npz'});np.testing.assert_array_equal(d['raw_ids'],[10,12])
|
||||
metadata=json.loads((root/'result.json').read_text());metadata['r2']['columns'][0]='wrong'
|
||||
(root/'result.json').write_text(json.dumps(metadata));row['preparation']['result_sha256']=sha(root/'result.json')
|
||||
with self.assertRaises(ValueError):prediction_inputs(row)
|
||||
|
||||
def test_named_B0_units_and_centred_coordinates_over_Rg(self):
|
||||
self.assertEqual(R2_COLUMNS[0],'Vf');np.testing.assert_array_equal(b0_seconds([0,.25,1],4),[0,1,4])
|
||||
x=np.array([[2.,4.,6.],[6.,10.,14.]])
|
||||
base=base_features(x,[1,2],[0,1],[2,4,6],[1,2,3],2,10,1,4)
|
||||
self.assertEqual(BASE_COLUMNS[:3],('x_star','y_star','z_star'))
|
||||
np.testing.assert_array_equal(base[:,:3],(x-[1,2,3])/2)
|
||||
self.assertEqual(last_location([1,2],[2,1],[0,1],base[:,:3]),np.linalg.norm(x[1]-x[0])/2)
|
||||
|
||||
def test_frozen_real_metadata_denominators_no_label_read(self):
|
||||
# Only immutable JSON membership/roster metadata, never geometry or targets.
|
||||
|
||||
Reference in new issue
Block a user