整理 Q1-Q3 实验代码与结果

2026-09-24 16:25:15 +08:00
parent 0261ecdfba
commit 8f5c2c3be6
247 changed files with 69828 additions and 19 deletions
@@ -0,0 +1,3 @@
+.venv/
+__pycache__/
+*.py[cod]
@@ -0,0 +1,39 @@
+# Q2/Q3 algorithm selection built on Q1 alignment
+
+## Decision about reusing Q1
+
+The transferable part of Q1 is its explicit time correspondence and observation mask: features from different modalities share ordered positions, missing values are accompanied by masks, and a position can be traced to source time. That interface is useful for both Q2 local-gap handling and Q3 evidence localization.
+
+The exact Q1 B1 extraction cannot be rerun over the 4,850 Attachment 2 training examples. Attachment 2 supplies precomputed aligned and unaligned feature tensors, but not the source audio/video or CTC word-time posteriors for the full training set. Its `aligned_50.pkl` also has 50 wordpiece positions and no Q1 `time_bounds_s`; those positions must not be described as the 50 equal-duration physical-time bins exported by `final/Q1`.
+
+Accordingly, the Q2 experiment uses the official aligned feature set as its shared wordpiece axis, and compares it with a fixed equal-window pooling control made from the official unaligned audio/vision sequences. This is a downstream alignment-utility check, not a claim that B1 was recomputed on Attachment 2. The official train/validation split is retained; test labels are not used.
+
+## Q2 candidates
+
+All candidates use identical training examples, train-only median/MAD scaling, joint polarity/intensity objectives, and 15 validation corruptions (three contiguous missing rates by five modality patterns).
+
+| Candidate | Fusion rule | What it tests |
+| --- | --- | --- |
+| `concat` | Project each modality, concatenate features and availability flags, then run a bidirectional GRU | Strong, simple early-fusion baseline |
+| `gate` | Learn per-slot modality weights, mask unavailable modalities, then run a bidirectional GRU | Whether explicit reliability-aware fusion handles local gaps |
+| `crossattn` | Apply masked cross-modal attention over the 50 shared slots, then temporal pooling | Whether contextual cross-modal exchange improves robustness |
+
+The report keeps Macro-F1, MAE, and Pearson separate. The default selection is Macro-F1-first across local corruption conditions; MAE and Pearson remain explicit tradeoffs, not terms in a constructed total score. The selected architecture is also trained on fixed-window-resampled features as an alignment control. A separate validation control shifts audio and vision by 1–10 positions to measure sensitivity to cross-modal timing.
+
+## Q3 explanation selection
+
+The selected Q2 model is frozen. Integrated Gradients and five-slot grouped occlusion are compared on held-out Attachment 2 validation clips using deletion comprehensiveness, sufficiency, and local rank stability. Attachment 4 has original videos and transcripts, so B1's CTC hard word-time procedure can be applied to those 20 clips to map high-importance wordpiece positions back to seconds. The saved Attachment 4 pickle files do not include `time_bounds_s`; explanations therefore retain both the model slot and the CTC-derived word interval, with alignment quality recorded.
+
+## Run
+
+The project environment is managed by `uv` and installs the CUDA 13.0 PyTorch build:
+
+```bash
+cd deep_learning/Q2
+uv sync
+uv run python -m q2.train_compare
+cd ../Q3
+uv run --project ../Q2 python -m q3.explain_selection
+```
+
+The main outputs are written to `outputs/algorithm_selection/`; plots, CSV metrics, run metadata, and checkpoints stay under this directory. The source data, `math`, and `final/Q1` are read-only inputs.
@@ -0,0 +1,54 @@
+# Q2 algorithm selection results
+
+## Q1 alignment transfer decision
+
+Q1 B1 aligns BERT word features and audio/vision observations with hard CTC word intervals, projects observed features onto a 0.1-second common grid, exports 50 equal-duration physical-time bins, and keeps observation masks. For Q2, the shared ordered axis and explicit masks transfer directly: a local gap stays a local gap after alignment and can be represented without inventing feature values.
+
+The exact B1 extraction was not recomputed over Attachment 2. The official 4,850-row feature package contains precomputed aligned and unaligned tensors, but no full-set source audio/video or word-time posterior. Its `aligned_50.pkl` has 50 wordpiece positions and no per-slot `time_bounds_s`; those positions are not Q1's 50 equal-duration bins. This experiment therefore trains on the official aligned features and compares them with an equal-window audio/vision resampling control. The comparison tests the value of an aligned ordered representation for the downstream Q2 task; it does not claim to reproduce B1 on all 4,850 clips.
+
+## Data and protocol
+
+- Attachment 2 official split: 3,395 training clips and 728 validation clips. Their source-video ID sets do not overlap.
+- Each official aligned example has 50 positions with Text 768-D, Audio 74-D, Vision 35-D features and modality observation masks.
+- Attachment 2 test labels were not used.
+- The three candidates shared train-only median/MAD normalization, the joint polarity/intensity objective, and training-time contiguous block masking.
+- Validation corruption covered 10%, 20%, and 30% of 50 positions for Text, Audio, Vision, Audio+Vision, and all three modalities. This is a wordpiece-position proxy for a continuous time gap; full-set second-level timestamps are not supplied.
+- Each candidate was run with seeds 42, 3407, and 2026. Reported `±` values are seed standard deviations over the fixed official validation set and deterministic corruption draws; they are not confidence intervals over new videos.
+
+## Fusion comparison
+
+| Model | Clean Accuracy | Clean Macro-F1 | Corrupt Accuracy, mean | Corrupt Macro-F1, mean | Worst condition Macro-F1 | Corrupt MAE | Corrupt Pearson |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| Early concatenation + BiGRU | 0.626 ± 0.013 | 0.580 ± 0.018 | 0.623 ± 0.011 | **0.575 ± 0.017** | **0.516** | **0.640 ± 0.004** | 0.607 ± 0.004 |
+| Reliability gate + BiGRU | 0.621 ± 0.011 | 0.570 ± 0.019 | 0.617 ± 0.011 | 0.565 ± 0.018 | 0.498 | 0.643 ± 0.008 | **0.610 ± 0.008** |
+| Masked cross-modal attention | 0.604 ± 0.012 | 0.540 ± 0.049 | 0.601 ± 0.009 | 0.539 ± 0.047 | 0.462 | 0.649 ± 0.024 | 0.592 ± 0.016 |
+
+Early concatenation has the best mean corrupted Macro-F1 and MAE. The gate has slightly higher Pearson, so the metrics do not collapse to one score. Cross-modal attention is lower and more variable at this sample size. It is not selected for the next Q2 stage.
+
+For the selected concatenation model, the hardest tested case is 30% Text masking: Macro-F1 0.547 and MAE 0.665, compared with clean Macro-F1 0.580 and MAE 0.636. Audio-only or Vision-only masking has a smaller effect in these runs. This is evidence about this feature set and these simulated spans; it does not establish a universal modality ranking.
+
+## Alignment utility control
+
+The same concatenation model was trained either on the supplied aligned wordpiece features or on equal-window-resampled audio/vision features from the official unaligned tensors.
+
+| Representation | Clean Macro-F1 | Corrupt Macro-F1 | Corrupt MAE | Corrupt Pearson |
+| --- | ---: | ---: | ---: | ---: |
+| Supplied word-aligned 50 positions | 0.580 ± 0.018 | **0.575 ± 0.017** | **0.640 ± 0.004** | **0.607 ± 0.004** |
+| Equal-window resampled unaligned input | 0.501 ± 0.014 | 0.504 ± 0.016 | 0.665 ± 0.005 | 0.577 ± 0.010 |
+
+On the selected model, shifting Audio and Vision by 1–10 positions changed aligned Macro-F1 from 0.580 to 0.557. That is a modest timing-sensitivity signal; it does not prove the model uses precise physical-time correspondence. Together with the fixed-window comparison, the result supports retaining the supplied aligned sequence for Q2.
+
+## Selected Q2 direction
+
+Continue with mask-aware early concatenation plus a bidirectional GRU, using local block masking during training. Keep the reliability gate as an ablation because its Pearson is slightly higher. Revisit cross-attention only if a later run has stronger evidence and enough data to control overfitting.
+
+## Reproducible artifacts
+
+- [Model and representation summary](outputs/algorithm_selection/summary.csv)
+- [Metrics by missing type and rate](outputs/algorithm_selection/validation_metrics_by_condition.csv)
+- [Aligned versus fixed-window and temporal-shift controls](outputs/algorithm_selection/alignment_transfer_ablation.csv)
+- [Training/data audit and run manifest](outputs/algorithm_selection/data_audit.json), [run manifest](outputs/algorithm_selection/run_manifest.json)
+- [Validation plot](outputs/algorithm_selection/missing_rate_comparison.png)
+- [Selected seed-42 checkpoint](outputs/algorithm_selection/models/aligned/concat/model_best.pt)
+
+The fitted checkpoint is for algorithm selection, not the final Attachment 3 submission model. The final model should be trained on train+validation after the architecture and thresholds are frozen.
@@ -0,0 +1,7 @@
+method,representation,condition,n_valid,n_seeds,accuracy,accuracy_sd,macro_f1,macro_f1_sd,mae,mae_sd,pearson,pearson_sd,missing_rate
+concat,provided_word_aligned_50,clean,728,3,0.6259157509157509,0.012689016905266455,0.5803030257575642,0.01849482950112713,0.6362011035283407,0.0035474183737517766,0.6131094378711319,0.003049322677391937,
+concat,provided_word_aligned_50,audio_vision_shifted_1_to_10_slots,728,3,0.6144688644688645,0.017174907814570563,0.5566710058858901,0.02417454695632849,0.6264231006304423,0.003071737263575265,0.610966440919508,0.00045089311901344093,
+concat,provided_word_aligned_50,all_local_corruption_mean,728,3,0.6226800976800977,0.011084117339613314,0.5751264735191365,0.016846851540436875,0.6402280900213454,0.004374268275822229,0.6068152054284395,0.004485336212715804,0.20000000000000004
+concat,equal_window_resampled_unaligned,clean,728,3,0.586996336996337,0.010491244722884272,0.5005019574320758,0.013925639871727626,0.6615431904792786,0.005397772426619935,0.581966026863303,0.009642037378255953,
+concat,equal_window_resampled_unaligned,audio_vision_shifted_1_to_10_slots,728,3,0.5956959706959707,0.010309826235528995,0.5144238712048543,0.010932166832616294,0.6616438627243042,0.005509720037298075,0.5807893064362651,0.010274359593119802,
+concat,equal_window_resampled_unaligned,all_local_corruption_mean,728,3,0.5905677655677655,0.006480515023060099,0.5042920441199125,0.015580461355158729,0.6651763810051813,0.004962154081458186,0.5766306314815771,0.010310184996660417,0.20000000000000004
@@ -0,0 +1,21 @@
+{
+  "source": "/home/gloamxun/modeling_zhaocui/E题数据/附件2-数据集特征文件/aligned_50.pkl",
+  "train_samples": 3395,
+  "valid_samples": 728,
+  "train_classes": [
+    967,
+    758,
+    1670
+  ],
+  "valid_classes": [
+    206,
+    184,
+    338
+  ],
+  "mean_observed_slots": {
+    "text": 24.645655375552284,
+    "audio": 22.626509572901327,
+    "vision": 21.394108983799704
+  },
+  "train_valid_video_overlap": 0
+}
@@ -0,0 +1,10 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.05146148469713,0.9808105859127674
+2.0,0.8792424190927435,0.8854341854105939
+3.0,0.7658277087741427,0.8630200951963991
+4.0,0.7162007325225406,0.8734401222113725
+5.0,0.666606965440291,0.8838760400866414
+6.0,0.6495787705536242,0.9118019499621548
+7.0,0.5989178496378439,0.9559067448416909
+8.0,0.5539200300419772,0.9710462656649914
+9.0,0.5062780554095904,1.0261535592131563
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0488198929362826,0.9925541471649002
+2.0,0.877586845446516,0.9104853272438049
+3.0,0.7854031710712998,0.8691042694416675
+4.0,0.7275559962899597,0.8561014971890293
+5.0,0.6785599307881461,0.8777891502275572
+6.0,0.6318395165381608,0.9196370329175677
+7.0,0.5972505140083807,0.9164495874237228
+8.0,0.5567877353341492,0.9559657193802216
+9.0,0.5066572507774388,1.0156079124618362
+10.0,0.47001609758094504,1.059160087134812
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0536950241636347,0.9706045127176977
+2.0,0.8524193581607606,0.8777912927197886
+3.0,0.760877827251399,0.8636277507949661
+4.0,0.7150518761740791,0.8624753559028709
+5.0,0.6774439651657034,0.877832626248454
+6.0,0.6516634187212696,0.8901654671836685
+7.0,0.5939983526865641,0.918816069325248
+8.0,0.5618191918841114,0.9479289251369435
+9.0,0.513365975132695,1.013728333043528
+10.0,0.4933904481154901,1.030043561379988
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0536950241636347,0.9706045127176977
+2.0,0.8524193581607606,0.8777912927197886
+3.0,0.760877827251399,0.8636277507949661
+4.0,0.7150518761740791,0.8624753559028709
+5.0,0.6774439651657034,0.877832626248454
+6.0,0.6516634187212696,0.8901654671836685
+7.0,0.5939983526865641,0.918816069325248
+8.0,0.5618191918841114,0.9479289251369435
+9.0,0.513365975132695,1.013728333043528
+10.0,0.4933904481154901,1.030043561379988
@@ -0,0 +1,10 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0184006403993677,0.9758902355864808
+2.0,0.8721449165432541,0.9131481346193251
+3.0,0.7584562389938919,0.9061737309445391
+4.0,0.70247816046079,0.908711409830785
+5.0,0.6474666976266437,0.9444795060943771
+6.0,0.6255715103061111,0.9916293214965652
+7.0,0.5631065650118722,1.0601372142414471
+8.0,0.5009032366452394,1.1017653536010574
+9.0,0.4493949540235378,1.177744794677902
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0344635292335793,0.9769261263229034
+2.0,0.8467512428760529,0.8954217656628116
+3.0,0.7528229709024783,0.8857204809293642
+4.0,0.7020640406343672,0.8774335417118702
+5.0,0.6610019422239728,0.9038714190105815
+6.0,0.6052676102629414,0.9758975283130185
+7.0,0.5659312236088293,0.9750687735421317
+8.0,0.5310967906757638,1.049424912903335
+9.0,0.46808629096658144,1.1149894405197311
+10.0,0.4314451297676122,1.177863896548093
@@ -0,0 +1,9 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.011995311136599,0.9519830608105921
+2.0,0.8360813723670112,0.8962717344472696
+3.0,0.7534678158936677,0.8991011263249995
+4.0,0.7069296527791906,0.9073571071519957
+5.0,0.6639300579274142,0.9432451947704776
+6.0,0.6353109103661997,0.9912105065125686
+7.0,0.5673890513954339,1.0443138106838687
+8.0,0.5254902547156369,1.1027433977022276
@@ -0,0 +1,9 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0119953784677718,0.9519833208440425
+2.0,0.8360813088991024,0.8962707964928596
+3.0,0.7534678666679947,0.8991013843934614
+4.0,0.7069298084135409,0.9073559026141743
+5.0,0.6639298437922089,0.9432453493495564
+6.0,0.6353108998801973,0.9912109388099922
+7.0,0.56738873405589,1.0443127351802783
+8.0,0.5254902160829968,1.1027441640476605
@@ -0,0 +1,10 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0707936783631642,1.0067818826371497
+2.0,0.9140718049473233,0.9000171115110208
+3.0,0.7863181178216581,0.8512143093151051
+4.0,0.7349886541013364,0.851207211122408
+5.0,0.6789810916891804,0.8575038864062383
+6.0,0.6610487986493994,0.8806245772393195
+7.0,0.6122590667671628,0.9165601101550427
+8.0,0.5692078007592095,0.9351011645662916
+9.0,0.5215919649711361,0.99570418714167
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0577489623317011,0.9879156252840063
+2.0,0.8716377052995894,0.8907375866240197
+3.0,0.7750455615697084,0.8654810419449439
+4.0,0.7179531797214791,0.8613645519529071
+5.0,0.6809909796273267,0.8796235846949148
+6.0,0.6348593281926932,0.9029495820894347
+7.0,0.6029366790144531,0.9171899249265482
+8.0,0.5716203340777645,0.9463915248493572
+9.0,0.5201758698180869,0.9928344789442125
+10.0,0.5027363620422505,1.0327370245378096
@@ -0,0 +1,10 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0575931425447818,0.9988141858970726
+2.0,0.8773076059641661,0.8911895647153749
+3.0,0.7700778461164899,0.8651458110128131
+4.0,0.7248913248380026,0.8694970201659988
+5.0,0.682613401501267,0.884684423823933
+6.0,0.659738369010113,0.9030193935383807
+7.0,0.6011747334290434,0.9310533351950593
+8.0,0.5665941779260282,0.9628150620303311
+9.0,0.5270494206084145,1.0122937671430818
@@ -0,0 +1,10 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0575931425447818,0.9988141858970726
+2.0,0.8773076059641661,0.8911895647153749
+3.0,0.7700778461164899,0.8651458110128131
+4.0,0.7248913248380026,0.8694970201659988
+5.0,0.682613401501267,0.884684423823933
+6.0,0.659738369010113,0.9030193935383807
+7.0,0.6011747334290434,0.9310533351950593
+8.0,0.5665941779260282,0.9628150620303311
+9.0,0.5270494206084145,1.0122937671430818
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0648082616152588,1.0257792996836232
+2.0,0.9508535012050912,0.9455079901349414
+3.0,0.83619585191762,0.9239543119629661
+4.0,0.7835445602734884,0.9075145099189256
+5.0,0.7335039586932571,0.9146714800006741
+6.0,0.7189743309109299,0.9153845460860284
+7.0,0.669045564201143,0.9579015452783186
+8.0,0.6221646765867869,0.9763828352257445
+9.0,0.5898675388760037,1.0184288430999924
+10.0,0.5511867072847154,1.0475065275862976
@@ -0,0 +1,12 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0638797470816859,1.028256350821191
+2.0,0.9404247038894229,0.9602446667440645
+3.0,0.8558459458527742,0.9207163734750433
+4.0,0.7991754125665735,0.9120825791096949
+5.0,0.7538611182460079,0.9113014386250422
+6.0,0.7000082863701714,0.9467641167588287
+7.0,0.6632694422646805,0.9526280204018394
+8.0,0.6257995438796503,0.9969304380836067
+9.0,0.5829168972041872,1.0277221163550576
+10.0,0.556718733575609,1.0591561126184987
+11.0,0.5129955758651098,1.1357925462198781
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0713691667274192,1.017243931581686
+2.0,0.9250873768771136,0.9292757629038213
+3.0,0.827078006333775,0.9095677916820233
+4.0,0.7792822652392917,0.9027277290166079
+5.0,0.7457061266457593,0.9076534837156862
+6.0,0.709590431716707,0.9116529927148924
+7.0,0.670481797169756,0.9217575978446793
+8.0,0.6348734536656627,0.9528291304032881
+9.0,0.5908853731773518,0.982715639439258
+10.0,0.5694722047558537,0.9955548689915583
@@ -0,0 +1,11 @@
+epoch,train_loss,valid_clean_loss
+1.0,1.0713691667274192,1.017243931581686
+2.0,0.9250873768771136,0.9292757629038213
+3.0,0.827078006333775,0.9095677916820233
+4.0,0.7792822652392917,0.9027277290166079
+5.0,0.7457061266457593,0.9076534837156862
+6.0,0.709590431716707,0.9116529927148924
+7.0,0.670481797169756,0.9217575978446793
+8.0,0.6348734536656627,0.9528291304032881
+9.0,0.5908853731773518,0.982715639439258
+10.0,0.5694722047558537,0.9955548689915583
@@ -0,0 +1,54 @@
+{
+  "source_feature": "/home/gloamxun/modeling_zhaocui/E题数据/附件2-数据集特征文件/aligned_50.pkl",
+  "source_sha256": "66e867aa74bc70a844e806e5571e371c9abb4a35f9e2887ce9b4d97ff2cb8fcd",
+  "device": "cuda",
+  "cuda_name": "NVIDIA GeForce RTX 5070 Ti",
+  "seeds": [
+    42,
+    3407,
+    2026
+  ],
+  "epochs_max": 32,
+  "patience": 6,
+  "batch_size": 64,
+  "best_epochs": {
+    "concat_seed_42": 4,
+    "concat_seed_3407": 4,
+    "concat_seed_2026": 3,
+    "gate_seed_42": 3,
+    "gate_seed_3407": 4,
+    "gate_seed_2026": 3,
+    "crossattn_seed_42": 2,
+    "crossattn_seed_3407": 4,
+    "crossattn_seed_2026": 3,
+    "fixed_window_concat_seed_42": 4,
+    "fixed_window_concat_seed_3407": 5,
+    "fixed_window_concat_seed_2026": 4
+  },
+  "selected_macro_f1_first": "concat",
+  "selection_policy": "report Macro-F1, MAE, and Pearson separately; selected model maximizes mean validation Macro-F1 across 15 contiguous corruption conditions, then uses MAE and lexical model name only as tie-breaks",
+  "models": [
+    "concat",
+    "gate",
+    "crossattn"
+  ],
+  "corruption_rates": [
+    0.1,
+    0.2,
+    0.3
+  ],
+  "corruption_patterns": [
+    "text",
+    "audio",
+    "vision",
+    "audio_vision",
+    "all_modalities"
+  ],
+  "feature_scaling": "training split median/MAD; fallback to standard deviation for zero-MAD dimensions",
+  "test_labels_used": false,
+  "alignment_transfer_limit": "The official aligned_50 data use a 50-slot wordpiece sequence with no per-slot seconds or stored Q1 B1 time_bounds. The fixed-window comparison is a downstream alignment control, not a re-run of Q1 B1 on the full dataset.",
+  "python": "3.14.7 (main, Aug 10 2026, 00:00:00) [GCC 16.1.1 20260515 (Red Hat 16.1.1-2)]",
+  "torch": "2.14.0+cu130",
+  "numpy": "2.5.3",
+  "created_unix": 1790237371.5417986
+}
@@ -0,0 +1 @@
+Macro-F1-first validation selection: concat. See summary.csv for the full multi-metric tradeoff.
@@ -0,0 +1,5 @@
+method,representation,n_seeds,clean_accuracy,clean_accuracy_sd,clean_macro_f1,clean_macro_f1_sd,clean_mae,clean_mae_sd,clean_pearson,clean_pearson_sd,corrupt_accuracy_mean,corrupt_accuracy_sd,corrupt_macro_f1_mean,corrupt_macro_f1_sd,corrupt_macro_f1_worst,corrupt_mae_mean,corrupt_mae_sd,corrupt_pearson_mean,corrupt_pearson_sd,f1_rate_10,accuracy_rate_10,mae_rate_10,f1_rate_20,accuracy_rate_20,mae_rate_20,f1_rate_30,accuracy_rate_30,mae_rate_30,pareto_nondominated
+concat,provided_word_aligned_50,3,0.6259157509157509,0.012689016905266455,0.5803030257575643,0.01849482950112713,0.6362011035283407,0.0035474183737517766,0.6131094378711319,0.003049322677391937,0.6226800976800976,0.011084117339613314,0.5751264735191365,0.016846851540436875,0.5160231153138954,0.6402280900213454,0.004374268275822229,0.6068152054284395,0.004485336212715804,0.5774116129988989,0.6244505494505495,0.6361218094825745,0.5732483562656322,0.6214285714285714,0.6398131450017294,0.5747194512928783,0.6221611721611722,0.6447493155797323,True
+gate,provided_word_aligned_50,3,0.6213369963369964,0.0106695789356511,0.570042406396691,0.0192417246427628,0.6392609675725301,0.010298306434114075,0.61843647657748,0.008153726338115605,0.6170940170940171,0.011429412792673272,0.5654484786987964,0.018209275544794713,0.4977788775985414,0.6432029167811076,0.00818900735475074,0.6097108251803484,0.008038674969140262,0.5690117947676202,0.6205128205128205,0.6386720657348633,0.5668714434894636,0.6183150183150183,0.6423242449760437,0.5604621978393048,0.6124542124542125,0.6486124396324157,True
+crossattn,provided_word_aligned_50,3,0.6039377289377289,0.011682555697960702,0.5402080959251137,0.04931996722297363,0.6486262281735738,0.0247913008377115,0.5962298100136721,0.016314693282960167,0.6013125763125764,0.00904432495782142,0.5394035484113701,0.047085193081398864,0.4615384615384615,0.6492320696512858,0.023553864046032207,0.5917838426035978,0.015944508668008214,0.5437388259395578,0.6055860805860807,0.6467597643534342,0.5379114650369052,0.5998168498168498,0.6491282820701599,0.5365603542576471,0.5985347985347985,0.6518081625302632,False
+concat,equal_window_resampled_unaligned,3,0.586996336996337,0.010491244722884272,0.5005019574320758,0.013925639871727626,0.6615431904792786,0.005397772426619935,0.581966026863303,0.009642037378255953,0.5905677655677655,0.006480515023060099,0.5042920441199125,0.015580461355158729,0.4641572706698656,0.6651763810051813,0.004962154081458186,0.5766306314815771,0.010310184996660417,0.5042141077437426,0.5906593406593407,0.6621770620346069,0.503384597978282,0.5899267399267399,0.6627050677935282,0.5052774266377128,0.5911172161172161,0.6706470131874084,True
@@ -0,0 +1,199 @@
+method,representation,seed,condition,missing_rate,n_valid,accuracy,macro_f1,mae,pearson
+concat,provided_word_aligned_50,42,clean,0.0,728,0.6332417582417582,0.587941053090477,0.6350555419921875,0.6157338809096098
+concat,provided_word_aligned_50,42,text,0.1,728,0.6263736263736264,0.5758300795065502,0.6362224221229553,0.6099001459533547
+concat,provided_word_aligned_50,42,audio,0.1,728,0.6318681318681318,0.5859694579820366,0.6346138119697571,0.6186827904447096
+concat,provided_word_aligned_50,42,vision,0.1,728,0.6277472527472527,0.5825072763635762,0.6360324621200562,0.6155412177622099
+concat,provided_word_aligned_50,42,audio_vision,0.1,728,0.635989010989011,0.5919984665799197,0.6340617537498474,0.6171314707027704
+concat,provided_word_aligned_50,42,all_modalities,0.1,728,0.625,0.5767997019394343,0.6347602009773254,0.6127068856220025
+concat,provided_word_aligned_50,42,text,0.2,728,0.6112637362637363,0.5533514537785528,0.6472386121749878,0.5878231360637466
+concat,provided_word_aligned_50,42,audio,0.2,728,0.6346153846153846,0.5913737928222076,0.6337395906448364,0.6167897438460014
+concat,provided_word_aligned_50,42,vision,0.2,728,0.6277472527472527,0.5807552037464527,0.6369960308074951,0.6152903324501466
+concat,provided_word_aligned_50,42,audio_vision,0.2,728,0.6346153846153846,0.5889252158180606,0.6331195831298828,0.6187845560398619
+concat,provided_word_aligned_50,42,all_modalities,0.2,728,0.6263736263736264,0.5759393139946568,0.6418775916099548,0.6097362713984099
+concat,provided_word_aligned_50,42,text,0.3,728,0.6181318681318682,0.5669198053141907,0.6596490740776062,0.57596152605997
+concat,provided_word_aligned_50,42,audio,0.3,728,0.6414835164835165,0.5997485874560443,0.6367413997650146,0.6154119880721182
+concat,provided_word_aligned_50,42,vision,0.3,728,0.6332417582417582,0.5862230880240699,0.6370688080787659,0.61519025568542
+concat,provided_word_aligned_50,42,audio_vision,0.3,728,0.6428571428571429,0.5982133735032932,0.6347481608390808,0.6176032363345405
+concat,provided_word_aligned_50,42,all_modalities,0.3,728,0.625,0.5769720397379972,0.6341773867607117,0.6145213983185148
+concat,provided_word_aligned_50,3407,clean,0.0,728,0.6332417582417582,0.5937554950072336,0.6333680152893066,0.6138300943227009
+concat,provided_word_aligned_50,3407,text,0.1,728,0.6277472527472527,0.5848809977670061,0.6363487243652344,0.6056385975757032
+concat,provided_word_aligned_50,3407,audio,0.1,728,0.635989010989011,0.5968950441819239,0.6318458914756775,0.6158991632022357
+concat,provided_word_aligned_50,3407,vision,0.1,728,0.6332417582417582,0.5933513166880985,0.6328762769699097,0.6142965716303582
+concat,provided_word_aligned_50,3407,audio_vision,0.1,728,0.635989010989011,0.5972272747847399,0.6317789554595947,0.6131521212086648
+concat,provided_word_aligned_50,3407,all_modalities,0.1,728,0.6208791208791209,0.5757563714642701,0.6323474645614624,0.6174836749082849
+concat,provided_word_aligned_50,3407,text,0.2,728,0.6043956043956044,0.5490811521155289,0.6509508490562439,0.5895223408430469
+concat,provided_word_aligned_50,3407,audio,0.2,728,0.635989010989011,0.5977119556159886,0.6304426789283752,0.618383030295478
+concat,provided_word_aligned_50,3407,vision,0.2,728,0.6401098901098901,0.6034867151810619,0.628890335559845,0.6187934617340419
+concat,provided_word_aligned_50,3407,audio_vision,0.2,728,0.6346153846153846,0.600406263153788,0.633434534072876,0.6123211947292676
+concat,provided_word_aligned_50,3407,all_modalities,0.2,728,0.6153846153846154,0.5731546231546232,0.6435555815696716,0.6001160890533341
+concat,provided_word_aligned_50,3407,text,0.3,728,0.6071428571428571,0.5572172310238351,0.6613345146179199,0.5663984666907937
+concat,provided_word_aligned_50,3407,audio,0.3,728,0.6401098901098901,0.602821099359954,0.6328924298286438,0.6157652973695024
+concat,provided_word_aligned_50,3407,vision,0.3,728,0.6277472527472527,0.5882423141524775,0.6326207518577576,0.6153119550822198
+concat,provided_word_aligned_50,3407,audio_vision,0.3,728,0.6428571428571429,0.6062757576614914,0.6341086030006409,0.6170348507761535
+concat,provided_word_aligned_50,3407,all_modalities,0.3,728,0.6277472527472527,0.5839727159969967,0.646858811378479,0.5966126328673234
+concat,provided_word_aligned_50,2026,clean,0.0,728,0.6112637362637363,0.5592125291749822,0.6401797533035278,0.6097643383810852
+concat,provided_word_aligned_50,2026,text,0.1,728,0.6098901098901099,0.5538253146595995,0.6407675743103027,0.6035210214254456
+concat,provided_word_aligned_50,2026,audio,0.1,728,0.614010989010989,0.5630983962746592,0.6385665535926819,0.6110860092543575
+concat,provided_word_aligned_50,2026,vision,0.1,728,0.6181318681318682,0.5661140652625988,0.6405577659606934,0.6110400577216631
+concat,provided_word_aligned_50,2026,audio_vision,0.1,728,0.6098901098901099,0.5587567567850849,0.6381099224090576,0.6107705749941461
+concat,provided_word_aligned_50,2026,all_modalities,0.1,728,0.614010989010989,0.5581636747439855,0.6429373621940613,0.6088956569868328
+concat,provided_word_aligned_50,2026,text,0.2,728,0.6043956043956044,0.5429151983962918,0.6489536762237549,0.5886848356465891
+concat,provided_word_aligned_50,2026,audio,0.2,728,0.6112637362637363,0.5603026186889267,0.636631965637207,0.6105368566655438
+concat,provided_word_aligned_50,2026,vision,0.2,728,0.614010989010989,0.5607756511971072,0.6439995765686035,0.6079712044449676
+concat,provided_word_aligned_50,2026,audio_vision,0.2,728,0.6181318681318682,0.5696462960623787,0.6381081342697144,0.6114528702793328
+concat,provided_word_aligned_50,2026,all_modalities,0.2,728,0.6085164835164835,0.5508998902588577,0.6492584347724915,0.6012091713678648
+concat,provided_word_aligned_50,2026,text,0.3,728,0.5851648351648352,0.5160231153138954,0.6731752753257751,0.5521914823230472
+concat,provided_word_aligned_50,2026,audio,0.3,728,0.6181318681318682,0.5670636517410711,0.636723518371582,0.6097718270787369
+concat,provided_word_aligned_50,2026,vision,0.3,728,0.6085164835164835,0.5560149900702367,0.6435868144035339,0.6098501902389435
+concat,provided_word_aligned_50,2026,audio_vision,0.3,728,0.6098901098901099,0.5653848813301509,0.637891948223114,0.6075333619057974
+concat,provided_word_aligned_50,2026,all_modalities,0.3,728,0.6043956043956044,0.5496991187074715,0.6696622371673584,0.5843647212263229
+gate,provided_word_aligned_50,42,clean,0.0,728,0.6332417582417582,0.5807021489645882,0.649620771408081,0.6141327125880438
+gate,provided_word_aligned_50,42,text,0.1,728,0.6291208791208791,0.5724806267179149,0.6482935547828674,0.607812656143153
+gate,provided_word_aligned_50,42,audio,0.1,728,0.6318681318681318,0.5804741658290519,0.64830082654953,0.6160345444489322
+gate,provided_word_aligned_50,42,vision,0.1,728,0.635989010989011,0.5855122033916645,0.6496425867080688,0.6131362897637801
+gate,provided_word_aligned_50,42,audio_vision,0.1,728,0.6332417582417582,0.5831283881314556,0.6489386558532715,0.6140409647711148
+gate,provided_word_aligned_50,42,all_modalities,0.1,728,0.6332417582417582,0.583341398636846,0.6480620503425598,0.6083974001811914
+gate,provided_word_aligned_50,42,text,0.2,728,0.6071428571428571,0.5464225853270284,0.6594239473342896,0.5807870469081623
+gate,provided_word_aligned_50,42,audio,0.2,728,0.6373626373626373,0.5879790951691172,0.6486459970474243,0.6137318643257155
+gate,provided_word_aligned_50,42,vision,0.2,728,0.6401098901098901,0.5906020456660217,0.6478570103645325,0.6146850468047298
+gate,provided_word_aligned_50,42,audio_vision,0.2,728,0.6428571428571429,0.5956295807308858,0.6458060145378113,0.615765199266891
+gate,provided_word_aligned_50,42,all_modalities,0.2,728,0.6222527472527473,0.5699698676231126,0.6509362459182739,0.6090527347361592
+gate,provided_word_aligned_50,42,text,0.3,728,0.5947802197802198,0.5351473922902494,0.6707870364189148,0.5698602169256346
+gate,provided_word_aligned_50,42,audio,0.3,728,0.6414835164835165,0.5932626146881034,0.6488574147224426,0.6137698439998989
+gate,provided_word_aligned_50,42,vision,0.3,728,0.635989010989011,0.5859542607988605,0.6494671106338501,0.6129307301381368
+gate,provided_word_aligned_50,42,audio_vision,0.3,728,0.6373626373626373,0.5906348593126334,0.6477416753768921,0.6122889569353058
+gate,provided_word_aligned_50,42,all_modalities,0.3,728,0.6304945054945055,0.5730354311449867,0.6414216160774231,0.6210376760306293
+gate,provided_word_aligned_50,3407,clean,0.0,728,0.6181318681318682,0.5815951114421278,0.6290252804756165,0.6278403558722048
+gate,provided_word_aligned_50,3407,text,0.1,728,0.6126373626373627,0.5710061419053845,0.6293500661849976,0.6178298909468922
+gate,provided_word_aligned_50,3407,audio,0.1,728,0.6208791208791209,0.5860650310799707,0.6279832124710083,0.6279555276295478
+gate,provided_word_aligned_50,3407,vision,0.1,728,0.6181318681318682,0.5824584906426339,0.629248321056366,0.6268484455037675
+gate,provided_word_aligned_50,3407,audio_vision,0.1,728,0.6195054945054945,0.5832702882513066,0.6273886561393738,0.6257845884321586
+gate,provided_word_aligned_50,3407,all_modalities,0.1,728,0.6098901098901099,0.5711323268553031,0.6288275122642517,0.6273843507994947
+gate,provided_word_aligned_50,3407,text,0.2,728,0.5947802197802198,0.5362174090491446,0.6508536338806152,0.5974496153645171
+gate,provided_word_aligned_50,3407,audio,0.2,728,0.6236263736263736,0.5893879935672744,0.6285275220870972,0.6276866860061261
+gate,provided_word_aligned_50,3407,vision,0.2,728,0.6126373626373627,0.5806822211441783,0.6233463287353516,0.6299624942730432
+gate,provided_word_aligned_50,3407,audio_vision,0.2,728,0.6236263736263736,0.5941062537491897,0.6226418614387512,0.631152312923411
+gate,provided_word_aligned_50,3407,all_modalities,0.2,728,0.6085164835164835,0.5685926111566246,0.6388359069824219,0.6172122010198969
+gate,provided_word_aligned_50,3407,text,0.3,728,0.5892857142857143,0.5319135217224389,0.6806973814964294,0.5598124161849717
+gate,provided_word_aligned_50,3407,audio,0.3,728,0.614010989010989,0.5787631246271929,0.6292544603347778,0.6264481106992486
+gate,provided_word_aligned_50,3407,vision,0.3,728,0.6112637362637363,0.5785955746513064,0.6262590885162354,0.6269980353567415
+gate,provided_word_aligned_50,3407,audio_vision,0.3,728,0.614010989010989,0.5854704803100748,0.6231685280799866,0.627471635956635
+gate,provided_word_aligned_50,3407,all_modalities,0.3,728,0.603021978021978,0.5649390371505222,0.6471090912818909,0.6058508041636224
+gate,provided_word_aligned_50,2026,clean,0.0,728,0.6126373626373627,0.5478299587833569,0.6391368508338928,0.6133363612721914
+gate,provided_word_aligned_50,2026,text,0.1,728,0.6085164835164835,0.5374777669587797,0.6402928233146667,0.61032865906815
+gate,provided_word_aligned_50,2026,audio,0.1,728,0.614010989010989,0.5504415308771583,0.6376320719718933,0.6137133407936685
+gate,provided_word_aligned_50,2026,vision,0.1,728,0.614010989010989,0.549111309400672,0.6384209990501404,0.6116423863441457
+gate,provided_word_aligned_50,2026,audio_vision,0.1,728,0.6112637362637363,0.548206981836245,0.6367278695106506,0.6134016265067737
+gate,provided_word_aligned_50,2026,all_modalities,0.1,728,0.6153846153846154,0.5510702709999151,0.6409717798233032,0.6055737170119238
+gate,provided_word_aligned_50,2026,text,0.2,728,0.6043956043956044,0.5387500187126036,0.6619775295257568,0.5870594418667304
+gate,provided_word_aligned_50,2026,audio,0.2,728,0.614010989010989,0.5520324803225694,0.6362826824188232,0.6150266095836183
+gate,provided_word_aligned_50,2026,vision,0.2,728,0.614010989010989,0.549868355076148,0.6375797986984253,0.6131134463734892
+gate,provided_word_aligned_50,2026,audio_vision,0.2,728,0.6098901098901099,0.5462836267094535,0.6391955614089966,0.6117577661672767
+gate,provided_word_aligned_50,2026,all_modalities,0.2,728,0.6195054945054945,0.556547508338603,0.642953634262085,0.6004007468165689
+gate,provided_word_aligned_50,2026,text,0.3,728,0.5810439560439561,0.4977788775985414,0.6796172261238098,0.55000542409833
+gate,provided_word_aligned_50,2026,audio,0.3,728,0.614010989010989,0.5490560178594751,0.6381493806838989,0.6113943643034875
+gate,provided_word_aligned_50,2026,vision,0.3,728,0.614010989010989,0.5555904891120482,0.6359769105911255,0.6131701769471742
+gate,provided_word_aligned_50,2026,audio_vision,0.3,728,0.6126373626373627,0.5581596077386942,0.6396954655647278,0.6082282425073746
+gate,provided_word_aligned_50,2026,all_modalities,0.3,728,0.5934065934065934,0.5286316785844464,0.6709842085838318,0.5729928980874585
+crossattn,provided_word_aligned_50,42,clean,0.0,728,0.6043956043956044,0.48998863119202335,0.6759882569313049,0.5777291731774127
+crossattn,provided_word_aligned_50,42,text,0.1,728,0.603021978021978,0.49273388898137477,0.6812499165534973,0.5694193464166294
+crossattn,provided_word_aligned_50,42,audio,0.1,728,0.6112637362637363,0.504821770685027,0.6705242395401001,0.5815934451841169
+crossattn,provided_word_aligned_50,42,vision,0.1,728,0.6043956043956044,0.4937041426648234,0.6775157451629639,0.5754628320692556
+crossattn,provided_word_aligned_50,42,audio_vision,0.1,728,0.6112637362637363,0.5009137863541273,0.6698321104049683,0.5814767002889362
+crossattn,provided_word_aligned_50,42,all_modalities,0.1,728,0.6016483516483516,0.49390809329904145,0.6687620282173157,0.5750133906760472
+crossattn,provided_word_aligned_50,42,text,0.2,728,0.5837912087912088,0.4692083792640309,0.6945626139640808,0.5412206795833109
+crossattn,provided_word_aligned_50,42,audio,0.2,728,0.6057692307692307,0.4919398568662255,0.6648613810539246,0.5871766995932145
+crossattn,provided_word_aligned_50,42,vision,0.2,728,0.603021978021978,0.4986750074706885,0.6775456666946411,0.5770632994127847
+crossattn,provided_word_aligned_50,42,audio_vision,0.2,728,0.6112637362637363,0.506035010012657,0.6634278297424316,0.5873091839764368
+crossattn,provided_word_aligned_50,42,all_modalities,0.2,728,0.5906593406593407,0.48358818056113656,0.6767656803131104,0.5728545007104048
+crossattn,provided_word_aligned_50,42,text,0.3,728,0.5769230769230769,0.4615384615384615,0.7029426693916321,0.5265077727922097
+crossattn,provided_word_aligned_50,42,audio,0.3,728,0.6057692307692307,0.4999911736715954,0.6637417078018188,0.585929811841307
+crossattn,provided_word_aligned_50,42,vision,0.3,728,0.6002747252747253,0.488070946081782,0.6788135766983032,0.5753310885314236
+crossattn,provided_word_aligned_50,42,audio_vision,0.3,728,0.6071428571428571,0.5068731682791942,0.6635814309120178,0.5881054554275786
+crossattn,provided_word_aligned_50,42,all_modalities,0.3,728,0.592032967032967,0.48234759307110725,0.667457640171051,0.5803081998379233
+crossattn,provided_word_aligned_50,3407,clean,0.0,728,0.6153846153846154,0.5885764436484813,0.6276583671569824,0.6085564971364105
+crossattn,provided_word_aligned_50,3407,text,0.1,728,0.6112637362637363,0.5835380749854434,0.6310563683509827,0.599629174971605
+crossattn,provided_word_aligned_50,3407,audio,0.1,728,0.6222527472527473,0.5972489054492567,0.623246431350708,0.6125511992350159
+crossattn,provided_word_aligned_50,3407,vision,0.1,728,0.6098901098901099,0.583342725650418,0.6262677311897278,0.6099948694514816
+crossattn,provided_word_aligned_50,3407,audio_vision,0.1,728,0.6112637362637363,0.5867178901767943,0.623745322227478,0.6093859569553041
+crossattn,provided_word_aligned_50,3407,all_modalities,0.1,728,0.614010989010989,0.5888196778236806,0.6245514750480652,0.6117287980833325
+crossattn,provided_word_aligned_50,3407,text,0.2,728,0.6043956043956044,0.5717745964093709,0.6391856074333191,0.5848733951460722
+crossattn,provided_word_aligned_50,3407,audio,0.2,728,0.6112637362637363,0.5894389643701733,0.6257225871086121,0.6074896792266403
+crossattn,provided_word_aligned_50,3407,vision,0.2,728,0.6112637362637363,0.5851332896742695,0.6264434456825256,0.6127778365291627
+crossattn,provided_word_aligned_50,3407,audio_vision,0.2,728,0.614010989010989,0.5949110192264351,0.6187586188316345,0.6182033017472267
+crossattn,provided_word_aligned_50,3407,all_modalities,0.2,728,0.6016483516483516,0.5773507156134905,0.634911835193634,0.592528289751998
+crossattn,provided_word_aligned_50,3407,text,0.3,728,0.6002747252747253,0.5711675869572423,0.6530678868293762,0.5606507699299987
+crossattn,provided_word_aligned_50,3407,audio,0.3,728,0.614010989010989,0.5913067138879948,0.6232653856277466,0.608846448174426
+crossattn,provided_word_aligned_50,3407,vision,0.3,728,0.6071428571428571,0.5819589420485362,0.6271139979362488,0.6113427502800617
+crossattn,provided_word_aligned_50,3407,audio_vision,0.3,728,0.6098901098901099,0.5897044167286073,0.6164263486862183,0.6193569599948916
+crossattn,provided_word_aligned_50,3407,all_modalities,0.3,728,0.6181318681318682,0.5940082804540815,0.6317148804664612,0.5948538231621964
+crossattn,provided_word_aligned_50,2026,clean,0.0,728,0.592032967032967,0.5420592129348364,0.6422320604324341,0.6024037597271933
+crossattn,provided_word_aligned_50,2026,text,0.1,728,0.5975274725274725,0.5454707818440054,0.6418640613555908,0.6024112136660502
+crossattn,provided_word_aligned_50,2026,audio,0.1,728,0.5934065934065934,0.5436021768815864,0.6399821043014526,0.6049551097626092
+crossattn,provided_word_aligned_50,2026,vision,0.1,728,0.5892857142857143,0.5384039205789096,0.6415307521820068,0.6011185044495995
+crossattn,provided_word_aligned_50,2026,audio_vision,0.1,728,0.603021978021978,0.554150981347194,0.6382296681404114,0.6055650886270252
+crossattn,provided_word_aligned_50,2026,all_modalities,0.1,728,0.6002747252747253,0.5487055723716843,0.6430385112762451,0.5965045050053945
+crossattn,provided_word_aligned_50,2026,text,0.2,728,0.5906593406593407,0.5288690186841035,0.6494663953781128,0.5950657404432197
+crossattn,provided_word_aligned_50,2026,audio,0.2,728,0.5906593406593407,0.5424377950535354,0.6427614688873291,0.601656698677867
+crossattn,provided_word_aligned_50,2026,vision,0.2,728,0.5934065934065934,0.5455594737723953,0.6451690793037415,0.5999133532386576
+crossattn,provided_word_aligned_50,2026,audio_vision,0.2,728,0.5961538461538461,0.5478165327681003,0.6367671489715576,0.6098925146891432
+crossattn,provided_word_aligned_50,2026,all_modalities,0.2,728,0.5892857142857143,0.5359341358069671,0.6405748724937439,0.5950458235969649
+crossattn,provided_word_aligned_50,2026,text,0.3,728,0.5755494505494505,0.5101004599189818,0.6680996417999268,0.565804882523542
+crossattn,provided_word_aligned_50,2026,audio,0.3,728,0.603021978021978,0.5546585571147281,0.6443181037902832,0.6002440646348579
+crossattn,provided_word_aligned_50,2026,vision,0.3,728,0.5989010989010989,0.5523658108429751,0.6397479772567749,0.6069549742183601
+crossattn,provided_word_aligned_50,2026,audio_vision,0.3,728,0.5879120879120879,0.5420598732858583,0.6358648538589478,0.6059215495922518
+crossattn,provided_word_aligned_50,2026,all_modalities,0.3,728,0.5810439560439561,0.5222533299835609,0.6609663367271423,0.5802332350553648
+concat,provided_word_aligned_50,42,audio_vision_shifted_1_to_10_slots,0.0,728,0.6222527472527473,0.5650157195157686,0.6228974461555481,0.6112627581844078
+concat,provided_word_aligned_50,3407,audio_vision_shifted_1_to_10_slots,0.0,728,0.6263736263736264,0.5755677417356012,0.6278499364852905,0.6111890270289838
+concat,provided_word_aligned_50,2026,audio_vision_shifted_1_to_10_slots,0.0,728,0.5947802197802198,0.5294295564063006,0.6285219192504883,0.6104475375451327
+concat,equal_window_resampled_unaligned,42,clean,0.0,728,0.5892857142857143,0.4853955760688928,0.6553117632865906,0.587849225352828
+concat,equal_window_resampled_unaligned,42,text,0.1,728,0.5892857142857143,0.48234450535531,0.6542671918869019,0.5821794113270566
+concat,equal_window_resampled_unaligned,42,audio,0.1,728,0.5879120879120879,0.4840888292569874,0.6544017195701599,0.589859444837917
+concat,equal_window_resampled_unaligned,42,vision,0.1,728,0.5906593406593407,0.48684880290869176,0.6560104489326477,0.5873086725722289
+concat,equal_window_resampled_unaligned,42,audio_vision,0.1,728,0.5961538461538461,0.49502890457555865,0.6592482924461365,0.5895610005108267
+concat,equal_window_resampled_unaligned,42,all_modalities,0.1,728,0.5865384615384616,0.4839129480007753,0.661249577999115,0.5814457099031457
+concat,equal_window_resampled_unaligned,42,text,0.2,728,0.5741758241758241,0.4641572706698656,0.6567733883857727,0.5645263734507251
+concat,equal_window_resampled_unaligned,42,audio,0.2,728,0.5934065934065934,0.49411312080823294,0.6519993543624878,0.5919285095409706
+concat,equal_window_resampled_unaligned,42,vision,0.2,728,0.5851648351648352,0.4807564908587733,0.6555609107017517,0.5864882167667788
+concat,equal_window_resampled_unaligned,42,audio_vision,0.2,728,0.5947802197802198,0.486680557422301,0.6588707566261292,0.5950950548477731
+concat,equal_window_resampled_unaligned,42,all_modalities,0.2,728,0.5824175824175825,0.4826855628872202,0.6649248003959656,0.5778664904235934
+concat,equal_window_resampled_unaligned,42,text,0.3,728,0.5769230769230769,0.47568745986340694,0.6655409932136536,0.5549458556193816
+concat,equal_window_resampled_unaligned,42,audio,0.3,728,0.5865384615384616,0.49062463717636134,0.6570892930030823,0.5903076532449915
+concat,equal_window_resampled_unaligned,42,vision,0.3,728,0.5934065934065934,0.49023200117754157,0.6581456661224365,0.5844521656955831
+concat,equal_window_resampled_unaligned,42,audio_vision,0.3,728,0.6057692307692307,0.5036551721604455,0.6700985431671143,0.5948534685103303
+concat,equal_window_resampled_unaligned,42,all_modalities,0.3,728,0.5989010989010989,0.49385933420149347,0.6675514578819275,0.5765029580102057
+concat,equal_window_resampled_unaligned,42,audio_vision_shifted_1_to_10_slots,0.0,728,0.5961538461538461,0.5023963012985687,0.6553080081939697,0.5869747480356168
+concat,equal_window_resampled_unaligned,3407,clean,0.0,728,0.5961538461538461,0.512827084557042,0.664772629737854,0.5872103830577866
+concat,equal_window_resampled_unaligned,3407,text,0.1,728,0.6002747252747253,0.5143254275091239,0.6675248146057129,0.5790469117683982
+concat,equal_window_resampled_unaligned,3407,audio,0.1,728,0.5961538461538461,0.5141876873569328,0.6630795001983643,0.5887049023755868
+concat,equal_window_resampled_unaligned,3407,vision,0.1,728,0.592032967032967,0.5021484550452703,0.6637163758277893,0.588692970574274
+concat,equal_window_resampled_unaligned,3407,audio_vision,0.1,728,0.6002747252747253,0.5172277536075813,0.6633241176605225,0.592489640322385
+concat,equal_window_resampled_unaligned,3407,all_modalities,0.1,728,0.6043956043956044,0.5264327838402229,0.6659616231918335,0.5874171690472151
+concat,equal_window_resampled_unaligned,3407,text,0.2,728,0.5810439560439561,0.49649824913662327,0.6659781336784363,0.5747007869349695
+concat,equal_window_resampled_unaligned,3407,audio,0.2,728,0.6002747252747253,0.5210696576225557,0.662110447883606,0.5898815380715464
+concat,equal_window_resampled_unaligned,3407,vision,0.2,728,0.603021978021978,0.5134655842561412,0.6635054349899292,0.5874898942649338
+concat,equal_window_resampled_unaligned,3407,audio_vision,0.2,728,0.6043956043956044,0.5213539297376665,0.665846586227417,0.591038037009385
+concat,equal_window_resampled_unaligned,3407,all_modalities,0.2,728,0.592032967032967,0.5111232435941855,0.6659784317016602,0.5810186064947893
+concat,equal_window_resampled_unaligned,3407,text,0.3,728,0.5824175824175825,0.49292032420488613,0.6802199482917786,0.5488280644778166
+concat,equal_window_resampled_unaligned,3407,audio,0.3,728,0.603021978021978,0.5278870449050491,0.6631943583488464,0.587517131217198
+concat,equal_window_resampled_unaligned,3407,vision,0.3,728,0.6043956043956044,0.5144897492123951,0.6652408838272095,0.5849148208986014
+concat,equal_window_resampled_unaligned,3407,audio_vision,0.3,728,0.6057692307692307,0.5222120983952546,0.6709288954734802,0.5894788895188137
+concat,equal_window_resampled_unaligned,3407,all_modalities,0.3,728,0.5934065934065934,0.5118478541545217,0.6919083595275879,0.558672500491863
+concat,equal_window_resampled_unaligned,3407,audio_vision_shifted_1_to_10_slots,0.0,728,0.6057692307692307,0.5237566082920623,0.6653115153312683,0.5864640082461416
+concat,equal_window_resampled_unaligned,2026,clean,0.0,728,0.5755494505494505,0.5032832116702929,0.6645451784133911,0.5708384721792944
+concat,equal_window_resampled_unaligned,2026,text,0.1,728,0.5686813186813187,0.4953554567971044,0.6647433638572693,0.5624346756639816
+concat,equal_window_resampled_unaligned,2026,audio,0.1,728,0.592032967032967,0.524084368831769,0.6619700789451599,0.5727849021508449
+concat,equal_window_resampled_unaligned,2026,vision,0.1,728,0.5837912087912088,0.5121488634353761,0.6646548509597778,0.5731723699341335
+concat,equal_window_resampled_unaligned,2026,audio_vision,0.1,728,0.5879120879120879,0.5138457090126345,0.6634346842765808,0.5783368477304374
+concat,equal_window_resampled_unaligned,2026,all_modalities,0.1,728,0.5837912087912088,0.5112311206228025,0.6690692901611328,0.5634143872695312
+concat,equal_window_resampled_unaligned,2026,text,0.2,728,0.570054945054945,0.4938343318957674,0.6707639694213867,0.5459273537200591
+concat,equal_window_resampled_unaligned,2026,audio,0.2,728,0.5934065934065934,0.5230327607731714,0.6592981815338135,0.5755735913907287
+concat,equal_window_resampled_unaligned,2026,vision,0.2,728,0.5906593406593407,0.5211672989417647,0.6658343076705933,0.5720666083948681
+concat,equal_window_resampled_unaligned,2026,audio_vision,0.2,728,0.5989010989010989,0.5276783237584716,0.6637739539146423,0.583433626473117
+concat,equal_window_resampled_unaligned,2026,all_modalities,0.2,728,0.5851648351648352,0.5131525873114892,0.6693573594093323,0.5605391525623993
+concat,equal_window_resampled_unaligned,2026,text,0.3,728,0.5508241758241759,0.4775661528696233,0.6854801177978516,0.5184409269913169
+concat,equal_window_resampled_unaligned,2026,audio,0.3,728,0.5989010989010989,0.5333065606332371,0.6566126942634583,0.576381050540554
+concat,equal_window_resampled_unaligned,2026,vision,0.3,728,0.592032967032967,0.5197981617020956,0.6703022122383118,0.5691319241770694
+concat,equal_window_resampled_unaligned,2026,audio_vision,0.3,728,0.6016483516483516,0.5344217204682321,0.6686394214630127,0.5850381809653504
+concat,equal_window_resampled_unaligned,2026,all_modalities,0.3,728,0.5728021978021978,0.4906531284411469,0.6887523531913757,0.5344899699772931
+concat,equal_window_resampled_unaligned,2026,audio_vision_shifted_1_to_10_slots,0.0,728,0.5851648351648352,0.5171187040239319,0.6643120646476746,0.5689291630270369
@@ -0,0 +1,19 @@
+[project]
+name = "deep-learning-q2-q3-selection"
+version = "0.1.0"
+requires-python = ">=3.14"
+dependencies = [
+    "matplotlib>=3.11.2",
+    "numpy>=2.5.3",
+    "scikit-learn>=1.9.1",
+    "torch>=2.14.0",
+    "transformers>=5.17.0",
+]
+
+[tool.uv.sources]
+torch = { index = "pytorch" }
+
+[[tool.uv.index]]
+name = "pytorch"
+url = "https://download.pytorch.org/whl/cu130"
+explicit = true
@@ -0,0 +1 @@
+"""Q2 robustness and Q3 explanation-selection experiments."""
@@ -0,0 +1,213 @@
+from __future__ import annotations
+
+import pickle
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+import numpy as np
+
+
+ROOT = Path(__file__).resolve().parents[3]
+ATTACHMENT2 = ROOT / "E题数据" / "附件2-数据集特征文件"
+MODALITIES = ("text", "audio", "vision")
+
+
+@dataclass
+class Split:
+    x: tuple[np.ndarray, np.ndarray, np.ndarray]
+    mask: np.ndarray  # N x T x 3
+    y_cls: np.ndarray
+    y_reg: np.ndarray
+    ids: list[str]
+
+    @property
+    def n(self) -> int:
+        return len(self.y_cls)
+
+    @property
+    def steps(self) -> int:
+        return int(self.x[0].shape[1])
+
+
+@dataclass
+class RobustStats:
+    center: tuple[np.ndarray, np.ndarray, np.ndarray]
+    scale: tuple[np.ndarray, np.ndarray, np.ndarray]
+
+    def save(self, path: Path) -> None:
+        path.parent.mkdir(parents=True, exist_ok=True)
+        np.savez_compressed(
+            path,
+            text_center=self.center[0], text_scale=self.scale[0],
+            audio_center=self.center[1], audio_scale=self.scale[1],
+            vision_center=self.center[2], vision_scale=self.scale[2],
+        )
+
+    @classmethod
+    def load(cls, path: Path) -> "RobustStats":
+        with np.load(path) as data:
+            return cls(
+                tuple(data[f"{m}_center"].astype(np.float32) for m in MODALITIES),
+                tuple(data[f"{m}_scale"].astype(np.float32) for m in MODALITIES),
+            )
+
+
+def _unpickle(path: Path) -> dict[str, Any]:
+    with path.open("rb") as stream:
+        return pickle.load(stream, encoding="latin1")
+
+
+def _ids_and_targets(part: dict[str, Any]) -> tuple[list[str], np.ndarray, np.ndarray]:
+    ids = [str(x) for x in part["id"]]
+    y_cls = np.asarray(part["classification_labels"], dtype=np.int64).reshape(-1)
+    y_reg = np.asarray(part["regression_labels"], dtype=np.float32).reshape(-1)
+    return ids, y_cls, y_reg
+
+
+def _text_mask(part: dict[str, Any]) -> np.ndarray:
+    tokens = np.asarray(part["text_bert"])
+    if tokens.ndim != 3 or tokens.shape[1] < 2:
+        raise ValueError(f"unexpected text_bert shape: {tokens.shape}")
+    # MOSEI text_bert rows are input_ids, input_mask, segment_ids.
+    return tokens[:, 1, :].astype(bool)
+
+
+def load_aligned(path: Path | None = None) -> dict[str, Split]:
+    path = path or ATTACHMENT2 / "aligned_50.pkl"
+    raw = _unpickle(path)
+    result: dict[str, Split] = {}
+    for name in ("train", "valid"):
+        part = raw[name]
+        xs = tuple(np.asarray(part[m], dtype=np.float32) for m in MODALITIES)
+        masks = [
+            _text_mask(part),
+            np.any(np.isfinite(xs[1]) & (xs[1] != 0), axis=-1),
+            np.any(np.isfinite(xs[2]) & (xs[2] != 0), axis=-1),
+        ]
+        mask = np.stack(masks, axis=-1)
+        ids, y_cls, y_reg = _ids_and_targets(part)
+        if any(x.shape[1] != 50 for x in xs):
+            raise ValueError(f"{name} aligned feature tensors must have 50 slots")
+        result[name] = Split(xs, mask, y_cls, y_reg, ids)
+    train_videos = {x.split("$_$", 1)[0] for x in result["train"].ids}
+    valid_videos = {x.split("$_$", 1)[0] for x in result["valid"].ids}
+    overlap = train_videos & valid_videos
+    if overlap:
+        raise ValueError(f"official train/valid split leaks {len(overlap)} source video ids")
+    return result
+
+
+def _resample_rows_to_50(values: np.ndarray, lengths: list[int] | np.ndarray) -> tuple[np.ndarray, np.ndarray]:
+    n, source_steps, dim = values.shape
+    output = np.zeros((n, 50, dim), dtype=np.float32)
+    mask = np.zeros((n, 50), dtype=bool)
+    lengths_arr = np.asarray(lengths, dtype=np.int64).reshape(-1)
+    for i in range(n):
+        length = int(np.clip(lengths_arr[i], 0, source_steps))
+        if length == 0:
+            continue
+        source = np.nan_to_num(values[i, :length], nan=0.0, posinf=0.0, neginf=0.0)
+        observed = np.any(source != 0, axis=-1)
+        for j in range(50):
+            left = int(np.floor(j * length / 50))
+            right = max(left + 1, int(np.ceil((j + 1) * length / 50)))
+            right = min(right, length)
+            use = observed[left:right]
+            if use.any():
+                output[i, j] = source[left:right][use].mean(axis=0)
+                mask[i, j] = True
+    return output, mask
+
+
+def load_fixed_window(path: Path | None = None) -> dict[str, Split]:
+    """Build a matched 50-slot equal-window control from the unaligned file."""
+    path = path or ATTACHMENT2 / "unaligned_50.pkl"
+    raw = _unpickle(path)
+    result: dict[str, Split] = {}
+    for name in ("train", "valid"):
+        part = raw[name]
+        text = np.asarray(part["text"], dtype=np.float32)
+        audio, audio_mask = _resample_rows_to_50(part["audio"], part["audio_lengths"])
+        vision, vision_mask = _resample_rows_to_50(part["vision"], part["vision_lengths"])
+        text_mask = _text_mask(part)
+        xs = (text, audio, vision)
+        mask = np.stack((text_mask, audio_mask, vision_mask), axis=-1)
+        ids, y_cls, y_reg = _ids_and_targets(part)
+        result[name] = Split(xs, mask, y_cls, y_reg, ids)
+    return result
+
+
+def fit_robust_stats(split: Split) -> RobustStats:
+    centers: list[np.ndarray] = []
+    scales: list[np.ndarray] = []
+    for modality in range(3):
+        observed = split.mask[:, :, modality].reshape(-1)
+        values = split.x[modality].reshape(-1, split.x[modality].shape[-1])[observed]
+        if not len(values):
+            raise ValueError(f"no observed values for {MODALITIES[modality]}")
+        values = np.nan_to_num(values, nan=0.0, posinf=0.0, neginf=0.0)
+        center = np.median(values, axis=0)
+        mad = np.median(np.abs(values - center), axis=0)
+        scale = 1.4826 * mad
+        std = np.std(values, axis=0)
+        scale = np.where(scale > 1e-6, scale, std)
+        scale = np.where(scale > 1e-6, scale, 1.0)
+        centers.append(center.astype(np.float32))
+        scales.append(scale.astype(np.float32))
+    return RobustStats(tuple(centers), tuple(scales))
+
+
+def apply_robust_stats(split: Split, stats: RobustStats) -> Split:
+    xs: list[np.ndarray] = []
+    for modality in range(3):
+        values = (split.x[modality] - stats.center[modality]) / stats.scale[modality]
+        values = np.nan_to_num(values, nan=0.0, posinf=0.0, neginf=0.0)
+        values *= split.mask[:, :, modality, None]
+        xs.append(values.astype(np.float32, copy=False))
+    return Split(tuple(xs), split.mask.copy(), split.y_cls, split.y_reg, split.ids)
+
+
+def corrupt_masks(
+    base: np.ndarray,
+    ratio: float,
+    modalities: tuple[int, ...],
+    seed: int,
+) -> np.ndarray:
+    result = base.copy()
+    rng = np.random.default_rng(seed)
+    n, steps, _ = result.shape
+    width = max(1, min(steps, int(round(ratio * steps))))
+    starts = rng.integers(0, steps - width + 1, size=n)
+    for row, start in enumerate(starts.tolist()):
+        result[row, start:start + width, list(modalities)] = False
+    return result
+
+
+def augment_masks(base: np.ndarray, rng: np.random.Generator) -> np.ndarray:
+    result = base.copy()
+    n, steps, _ = result.shape
+    for row in range(n):
+        if rng.random() >= 0.85:
+            continue
+        count = int(rng.integers(1, 4))
+        modalities = rng.choice(3, size=count, replace=False)
+        ratio = float(rng.choice((0.10, 0.20, 0.30)))
+        width = max(1, int(round(ratio * steps)))
+        start = int(rng.integers(0, steps - width + 1))
+        result[row, start:start + width, modalities] = False
+    return result
+
+
+def shift_audio_vision(split: Split, seed: int, max_shift: int = 10) -> Split:
+    rng = np.random.default_rng(seed)
+    xs = [x.copy() for x in split.x]
+    masks = split.mask.copy()
+    for row in range(split.n):
+        for modality in (1, 2):
+            shift = int(rng.integers(1, max_shift + 1))
+            if rng.random() < 0.5:
+                shift = -shift
+            xs[modality][row] = np.roll(xs[modality][row], shift, axis=0)
+            masks[row, :, modality] = np.roll(masks[row, :, modality], shift)
+    return Split(tuple(xs), masks, split.y_cls, split.y_reg, split.ids)
@@ -0,0 +1,29 @@
+from __future__ import annotations
+
+import argparse
+import csv
+from pathlib import Path
+
+from .train_compare import _plot, _summary, _write_csv
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Rebuild Q2 summary tables from saved validation predictions")
+    parser.add_argument("--output-dir", default=str(Path(__file__).resolve().parents[1] / "outputs" / "algorithm_selection"))
+    args = parser.parse_args()
+    output = Path(args.output_dir)
+    with (output / "validation_metrics_by_condition.csv").open(encoding="utf-8-sig", newline="") as stream:
+        rows = list(csv.DictReader(stream))
+    for row in rows:
+        for key in ("missing_rate", "accuracy", "macro_f1", "mae", "pearson", "n_valid"):
+            row[key] = float(row[key])
+        row["seed"] = int(row["seed"])
+    summary = _summary(rows)
+    _write_csv(output / "summary.csv", summary)
+    aligned = [row for row in summary if row["representation"] == "provided_word_aligned_50"]
+    _plot(aligned, rows, output / "missing_rate_comparison.png")
+    print(f"rebuilt summary table and plot from {len(rows)} saved validation rows")
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,109 @@
+from __future__ import annotations
+
+import torch
+from torch import nn
+
+
+class AlignedFusionModel(nn.Module):
+    def __init__(
+        self,
+        kind: str,
+        dims: tuple[int, int, int],
+        steps: int = 50,
+        hidden: int = 128,
+        dropout: float = 0.15,
+    ) -> None:
+        super().__init__()
+        if kind not in {"concat", "gate", "crossattn"}:
+            raise ValueError(f"unknown model kind: {kind}")
+        self.kind = kind
+        self.hidden = hidden
+        self.projections = nn.ModuleList(
+            nn.Sequential(nn.Linear(size, hidden), nn.GELU(), nn.LayerNorm(hidden))
+            for size in dims
+        )
+        self.position = nn.Parameter(torch.randn(1, steps, hidden) * 0.02)
+        self.modality = nn.Parameter(torch.randn(1, 1, 3, hidden) * 0.02)
+        self.dropout = nn.Dropout(dropout)
+
+        if kind == "concat":
+            self.fusion = nn.Sequential(
+                nn.Linear(hidden * 3 + 3, hidden), nn.GELU(), nn.LayerNorm(hidden), nn.Dropout(dropout)
+            )
+        elif kind == "gate":
+            self.gate_score = nn.Sequential(nn.Linear(hidden, hidden // 2), nn.Tanh(), nn.Linear(hidden // 2, 1))
+            self.fusion = nn.Sequential(
+                nn.Linear(hidden + 3, hidden), nn.GELU(), nn.LayerNorm(hidden), nn.Dropout(dropout)
+            )
+        else:
+            layer = nn.TransformerEncoderLayer(
+                d_model=hidden,
+                nhead=4,
+                dim_feedforward=hidden * 2,
+                dropout=dropout,
+                activation="gelu",
+                batch_first=True,
+                norm_first=True,
+            )
+            self.cross_encoder = nn.TransformerEncoder(layer, num_layers=2, enable_nested_tensor=False)
+            self.fusion = nn.Sequential(
+                nn.Linear(hidden + 3, hidden), nn.GELU(), nn.LayerNorm(hidden), nn.Dropout(dropout)
+            )
+
+        self.temporal = nn.GRU(
+            input_size=hidden,
+            hidden_size=hidden // 2,
+            num_layers=1,
+            batch_first=True,
+            bidirectional=True,
+        )
+        self.head = nn.Sequential(nn.Linear(hidden, hidden // 2), nn.GELU(), nn.Dropout(dropout))
+        self.classifier = nn.Linear(hidden // 2, 3)
+        self.regressor = nn.Linear(hidden // 2, 1)
+
+    def forward(self, xs: tuple[torch.Tensor, torch.Tensor, torch.Tensor], masks: torch.Tensor):
+        masks = masks.bool()
+        pos = self.position[:, :masks.shape[1]]
+        encoded = []
+        for modality, (projection, x) in enumerate(zip(self.projections, xs)):
+            token = projection(x)
+            token = token + pos + self.modality[:, :, modality, :]
+            token = token * masks[:, :, modality, None]
+            encoded.append(token)
+        stack = torch.stack(encoded, dim=2)  # B x T x M x D
+        availability = masks.to(stack.dtype)
+        gate_weights = None
+
+        if self.kind == "concat":
+            fused = self.fusion(torch.cat((stack.flatten(2), availability), dim=-1))
+        elif self.kind == "gate":
+            scores = self.gate_score(stack).squeeze(-1)
+            scores = scores.masked_fill(~masks, -1e4)
+            gate_weights = torch.softmax(scores, dim=-1) * availability
+            gate_weights = gate_weights / gate_weights.sum(dim=-1, keepdim=True).clamp_min(1e-8)
+            weighted = (stack * gate_weights[..., None]).sum(dim=2)
+            fused = self.fusion(torch.cat((weighted, availability), dim=-1))
+        else:
+            batch, steps, modalities, hidden = stack.shape
+            flat = stack.reshape(batch, steps * modalities, hidden)
+            valid = masks.reshape(batch, steps * modalities).clone()
+            empty = ~valid.any(dim=1)
+            if empty.any():
+                valid[empty, 0] = True
+                flat[empty, 0] = 0.0
+            attended = self.cross_encoder(flat, src_key_padding_mask=~valid)
+            attended = attended.reshape(batch, steps, modalities, hidden)
+            observed_count = availability.sum(dim=2, keepdim=True)
+            pooled = (attended * availability[..., None]).sum(dim=2) / observed_count.clamp_min(1.0)
+            fused = self.fusion(torch.cat((pooled, availability), dim=-1))
+
+        temporal, _ = self.temporal(self.dropout(fused))
+        time_weight = masks.any(dim=-1).to(temporal.dtype)
+        empty_time = time_weight.sum(dim=1, keepdim=True) <= 0
+        if empty_time.any():
+            time_weight[empty_time.squeeze(1), 0] = 1.0
+        pooled = (temporal * time_weight[..., None]).sum(dim=1) / time_weight.sum(dim=1, keepdim=True).clamp_min(1.0)
+        hidden = self.head(pooled)
+        logits = self.classifier(hidden)
+        intensity = 3.0 * torch.tanh(self.regressor(hidden).squeeze(-1))
+        return {"logits": logits, "intensity": intensity, "gate": gate_weights}
@@ -0,0 +1,490 @@
+from __future__ import annotations
+
+import argparse
+import csv
+import hashlib
+import json
+import math
+import random
+import shutil
+import time
+from collections import Counter
+from pathlib import Path
+from typing import Any
+
+import matplotlib.pyplot as plt
+import numpy as np
+import torch
+import torch.nn.functional as F
+from sklearn.metrics import accuracy_score, f1_score, mean_absolute_error
+from torch import nn
+
+from .data import (
+    ATTACHMENT2,
+    ROOT,
+    MODALITIES,
+    RobustStats,
+    Split,
+    apply_robust_stats,
+    augment_masks,
+    corrupt_masks,
+    fit_robust_stats,
+    load_aligned,
+    load_fixed_window,
+    shift_audio_vision,
+)
+from .models import AlignedFusionModel
+
+
+PATTERNS = {
+    "text": (0,),
+    "audio": (1,),
+    "vision": (2,),
+    "audio_vision": (1, 2),
+    "all_modalities": (0, 1, 2),
+}
+KINDS = ("concat", "gate", "crossattn")
+
+
+def seed_everything(seed: int) -> None:
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(seed)
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+
+def _tensor_split(split: Split, device: torch.device) -> tuple[tuple[torch.Tensor, ...], torch.Tensor, torch.Tensor, torch.Tensor]:
+    xs = tuple(torch.as_tensor(x, dtype=torch.float32, device=device) for x in split.x)
+    mask = torch.as_tensor(split.mask, dtype=torch.bool, device=device)
+    y_cls = torch.as_tensor(split.y_cls, dtype=torch.long, device=device)
+    y_reg = torch.as_tensor(split.y_reg, dtype=torch.float32, device=device)
+    return xs, mask, y_cls, y_reg
+
+
+def _loss(output: dict[str, torch.Tensor], y_cls: torch.Tensor, y_reg: torch.Tensor) -> torch.Tensor:
+    class_loss = F.cross_entropy(output["logits"], y_cls)
+    intensity_loss = F.smooth_l1_loss(output["intensity"] / 3.0, y_reg / 3.0)
+    return class_loss + 0.5 * intensity_loss
+
+
+@torch.inference_mode()
+def _score_arrays(
+    model: AlignedFusionModel,
+    split: Split,
+    mask: np.ndarray,
+    device: torch.device,
+    batch_size: int = 128,
+) -> tuple[dict[str, float], dict[str, np.ndarray]]:
+    model.eval()
+    predictions: dict[str, list[np.ndarray]] = {"logits": [], "intensity": []}
+    xs = split.x
+    for start in range(0, split.n, batch_size):
+        end = min(start + batch_size, split.n)
+        xb = tuple(torch.as_tensor(x[start:end], dtype=torch.float32, device=device) for x in xs)
+        mb = torch.as_tensor(mask[start:end], dtype=torch.bool, device=device)
+        output = model(xb, mb)
+        predictions["logits"].append(output["logits"].float().cpu().numpy())
+        predictions["intensity"].append(output["intensity"].float().cpu().numpy())
+    logits = np.concatenate(predictions["logits"], axis=0)
+    intensity = np.clip(np.concatenate(predictions["intensity"], axis=0), -3.0, 3.0)
+    pred_cls = logits.argmax(axis=-1)
+    pearson = _pearson(split.y_reg, intensity)
+    metrics = {
+        "accuracy": float(accuracy_score(split.y_cls, pred_cls)),
+        "macro_f1": float(f1_score(split.y_cls, pred_cls, labels=[0, 1, 2], average="macro", zero_division=0)),
+        "mae": float(mean_absolute_error(split.y_reg, intensity)),
+        "pearson": pearson,
+    }
+    return metrics, {"logits": logits, "intensity": intensity, "class": pred_cls}
+
+
+def _pearson(y: np.ndarray, pred: np.ndarray) -> float:
+    a = np.asarray(y, dtype=np.float64)
+    b = np.asarray(pred, dtype=np.float64)
+    if a.std() < 1e-12 or b.std() < 1e-12:
+        return 0.0
+    return float(np.corrcoef(a, b)[0, 1])
+
+
+def _validation_loss(model: AlignedFusionModel, valid: Split, device: torch.device, batch_size: int) -> float:
+    model.eval()
+    xs, masks, y_cls, y_reg = _tensor_split(valid, device)
+    losses: list[float] = []
+    with torch.inference_mode():
+        for start in range(0, valid.n, batch_size):
+            idx = slice(start, min(start + batch_size, valid.n))
+            output = model(tuple(x[idx] for x in xs), masks[idx])
+            losses.append(float(_loss(output, y_cls[idx], y_reg[idx]).item()))
+    return float(np.average(losses, weights=[min(batch_size, valid.n - i) for i in range(0, valid.n, batch_size)]))
+
+
+def _write_csv(path: Path, rows: list[dict[str, Any]]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    if not rows:
+        return
+    fields = list(dict.fromkeys(key for row in rows for key in row))
+    with path.open("w", newline="", encoding="utf-8-sig") as stream:
+        writer = csv.DictWriter(stream, fieldnames=fields)
+        writer.writeheader()
+        writer.writerows(rows)
+
+
+def _train_one(
+    kind: str,
+    train: Split,
+    valid: Split,
+    output_dir: Path,
+    device: torch.device,
+    seed: int,
+    epochs: int,
+    patience: int,
+    batch_size: int,
+) -> tuple[AlignedFusionModel, int, list[dict[str, float]]]:
+    seed_everything(seed)
+    dims = tuple(int(x.shape[-1]) for x in train.x)
+    model = AlignedFusionModel(kind, dims=dims).to(device)
+    optimizer = torch.optim.AdamW(model.parameters(), lr=1.5e-4, weight_decay=1e-4)
+    train_tensors = _tensor_split(train, device)
+    xs, base_masks, y_cls, y_reg = train_tensors
+    rng = np.random.default_rng(seed + 809)
+    best_loss = math.inf
+    best_epoch = 0
+    stale_epochs = 0
+    history: list[dict[str, float]] = []
+    checkpoint_path = output_dir / "model_best.pt"
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    for epoch in range(1, epochs + 1):
+        model.train()
+        order = rng.permutation(train.n)
+        batch_losses: list[float] = []
+        for start in range(0, train.n, batch_size):
+            ids_np = order[start:start + batch_size]
+            ids = torch.as_tensor(ids_np, dtype=torch.long, device=device)
+            masks_np = augment_masks(train.mask[ids_np], rng)
+            masks = torch.as_tensor(masks_np, dtype=torch.bool, device=device)
+            output = model(tuple(x.index_select(0, ids) for x in xs), masks)
+            loss = _loss(output, y_cls.index_select(0, ids), y_reg.index_select(0, ids))
+            optimizer.zero_grad(set_to_none=True)
+            loss.backward()
+            nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)
+            optimizer.step()
+            batch_losses.append(float(loss.detach().item()))
+        valid_loss = _validation_loss(model, valid, device, batch_size)
+        row = {"epoch": float(epoch), "train_loss": float(np.mean(batch_losses)), "valid_clean_loss": valid_loss}
+        history.append(row)
+        print(f"[{kind}] epoch={epoch:02d} train={row['train_loss']:.4f} valid={valid_loss:.4f}", flush=True)
+        if valid_loss < best_loss - 1e-4:
+            best_loss = valid_loss
+            best_epoch = epoch
+            stale_epochs = 0
+            torch.save({"kind": kind, "dims": dims, "state_dict": model.state_dict(), "seed": seed, "best_epoch": epoch}, checkpoint_path)
+        else:
+            stale_epochs += 1
+            if stale_epochs >= patience:
+                break
+
+    saved = torch.load(checkpoint_path, map_location=device, weights_only=False)
+    model.load_state_dict(saved["state_dict"])
+    model.eval()
+    _write_csv(output_dir / "training_history.csv", history)
+    return model, best_epoch, history
+
+
+def _conditions(valid: Split, seed: int) -> list[tuple[str, float, np.ndarray]]:
+    result = [("clean", 0.0, valid.mask.copy())]
+    for rate in (0.10, 0.20, 0.30):
+        for pattern_id, (pattern, mods) in enumerate(PATTERNS.items()):
+            result.append((pattern, rate, corrupt_masks(valid.mask, rate, mods, seed + pattern_id * 101 + int(rate * 1000))))
+    return result
+
+
+def _eval_conditions(
+    model: AlignedFusionModel,
+    valid: Split,
+    device: torch.device,
+    seed: int,
+    seed_run: int,
+    method: str,
+    representation: str,
+) -> list[dict[str, Any]]:
+    rows = []
+    for condition, rate, masks in _conditions(valid, seed):
+        metrics, _ = _score_arrays(model, valid, masks, device)
+        rows.append({"method": method, "representation": representation, "seed": seed_run, "condition": condition,
+                     "missing_rate": rate, "n_valid": valid.n, **metrics})
+        print(f"[{method}/{representation}] {condition:14s} rate={rate:.1f} "
+              f"F1={metrics['macro_f1']:.3f} MAE={metrics['mae']:.3f} "
+              f"P={metrics['pearson']:.3f}", flush=True)
+    return rows
+
+
+def _summary(rows: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    groups = list(dict.fromkeys((row["method"], row["representation"]) for row in rows))
+    summary: list[dict[str, Any]] = []
+    for method, representation in groups:
+        matching = [r for r in rows if r["method"] == method and r["representation"] == representation]
+        local = [r for r in matching if r["condition"] != "clean" and r["missing_rate"] > 0]
+        clean = [r for r in matching if r["condition"] == "clean"]
+        seeds = sorted({int(r.get("seed", 0)) for r in matching})
+
+        def per_seed_mean(selected: list[dict[str, Any]], metric: str) -> list[float]:
+            return [float(np.mean([r[metric] for r in selected if int(r.get("seed", 0)) == seed]))
+                    for seed in seeds if any(int(r.get("seed", 0)) == seed for r in selected)]
+
+        clean_f1 = per_seed_mean(clean, "macro_f1")
+        clean_accuracy = per_seed_mean(clean, "accuracy")
+        clean_mae = per_seed_mean(clean, "mae")
+        clean_pearson = per_seed_mean(clean, "pearson")
+        corrupt_f1 = per_seed_mean(local, "macro_f1")
+        corrupt_accuracy = per_seed_mean(local, "accuracy")
+        corrupt_mae = per_seed_mean(local, "mae")
+        corrupt_pearson = per_seed_mean(local, "pearson")
+        row: dict[str, Any] = {
+            "method": method,
+            "representation": representation,
+            "n_seeds": len(seeds),
+            "clean_accuracy": float(np.mean(clean_accuracy)),
+            "clean_accuracy_sd": float(np.std(clean_accuracy, ddof=1)) if len(clean_accuracy) > 1 else 0.0,
+            "clean_macro_f1": float(np.mean(clean_f1)),
+            "clean_macro_f1_sd": float(np.std(clean_f1, ddof=1)) if len(clean_f1) > 1 else 0.0,
+            "clean_mae": float(np.mean(clean_mae)),
+            "clean_mae_sd": float(np.std(clean_mae, ddof=1)) if len(clean_mae) > 1 else 0.0,
+            "clean_pearson": float(np.mean(clean_pearson)),
+            "clean_pearson_sd": float(np.std(clean_pearson, ddof=1)) if len(clean_pearson) > 1 else 0.0,
+            "corrupt_accuracy_mean": float(np.mean(corrupt_accuracy)),
+            "corrupt_accuracy_sd": float(np.std(corrupt_accuracy, ddof=1)) if len(corrupt_accuracy) > 1 else 0.0,
+            "corrupt_macro_f1_mean": float(np.mean(corrupt_f1)),
+            "corrupt_macro_f1_sd": float(np.std(corrupt_f1, ddof=1)) if len(corrupt_f1) > 1 else 0.0,
+            "corrupt_macro_f1_worst": float(np.min([r["macro_f1"] for r in local])),
+            "corrupt_mae_mean": float(np.mean(corrupt_mae)),
+            "corrupt_mae_sd": float(np.std(corrupt_mae, ddof=1)) if len(corrupt_mae) > 1 else 0.0,
+            "corrupt_pearson_mean": float(np.mean(corrupt_pearson)),
+            "corrupt_pearson_sd": float(np.std(corrupt_pearson, ddof=1)) if len(corrupt_pearson) > 1 else 0.0,
+        }
+        for rate in (0.10, 0.20, 0.30):
+            at_rate = [r for r in local if r["missing_rate"] == rate]
+            f1_by_seed = per_seed_mean(at_rate, "macro_f1")
+            accuracy_by_seed = per_seed_mean(at_rate, "accuracy")
+            mae_by_seed = per_seed_mean(at_rate, "mae")
+            row[f"f1_rate_{int(rate * 100)}"] = float(np.mean(f1_by_seed))
+            row[f"accuracy_rate_{int(rate * 100)}"] = float(np.mean(accuracy_by_seed))
+            row[f"mae_rate_{int(rate * 100)}"] = float(np.mean(mae_by_seed))
+        summary.append(row)
+    for row in summary:
+        row["pareto_nondominated"] = not any(
+            other is not row and other["representation"] == row["representation"]
+            and other["corrupt_macro_f1_mean"] >= row["corrupt_macro_f1_mean"]
+            and other["corrupt_mae_mean"] <= row["corrupt_mae_mean"]
+            and other["corrupt_pearson_mean"] >= row["corrupt_pearson_mean"]
+            and (
+                other["corrupt_macro_f1_mean"] > row["corrupt_macro_f1_mean"]
+                or other["corrupt_mae_mean"] < row["corrupt_mae_mean"]
+                or other["corrupt_pearson_mean"] > row["corrupt_pearson_mean"]
+            )
+            for other in summary
+        )
+    return summary
+
+
+def _plot(summary: list[dict[str, Any]], rows: list[dict[str, Any]], path: Path) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    colors = {"concat": "#4e79a7", "gate": "#f28e2b", "crossattn": "#59a14f"}
+    fig, axes = plt.subplots(1, 2, figsize=(11, 4.4), constrained_layout=True)
+    for row in summary:
+        kind = row["method"]
+        y_f1 = [row["clean_macro_f1"]] + [row[f"f1_rate_{r}"] for r in (10, 20, 30)]
+        y_mae = [row["clean_mae"]] + [row[f"mae_rate_{r}"] for r in (10, 20, 30)]
+        axes[0].plot([0, 10, 20, 30], y_f1, marker="o", label=kind, color=colors.get(kind))
+        axes[1].plot([0, 10, 20, 30], y_mae, marker="o", label=kind, color=colors.get(kind))
+    axes[0].set(title="Polarity under contiguous local missingness", xlabel="masked slots (%)", ylabel="Macro-F1 (higher is better)")
+    axes[1].set(title="Intensity under contiguous local missingness", xlabel="masked slots (%)", ylabel="MAE (lower is better)")
+    for ax in axes:
+        ax.grid(alpha=0.25)
+        ax.legend(frameon=False)
+    fig.savefig(path, dpi=180)
+    plt.close(fig)
+
+
+def _sha256(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as stream:
+        for block in iter(lambda: stream.read(1024 * 1024), b""):
+            digest.update(block)
+    return digest.hexdigest()
+
+
+def _run(args: argparse.Namespace) -> None:
+    seed_everything(args.seeds[0])
+    if args.device == "auto":
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    else:
+        device = torch.device(args.device)
+    torch.set_num_threads(args.threads)
+    output = Path(args.output_dir)
+    output.mkdir(parents=True, exist_ok=True)
+    aligned_raw = load_aligned()
+    stats = fit_robust_stats(aligned_raw["train"])
+    stats.save(output / "aligned_robust_stats.npz")
+    aligned = {k: apply_robust_stats(v, stats) for k, v in aligned_raw.items()}
+    audit = {
+        "source": str(ATTACHMENT2 / "aligned_50.pkl"),
+        "train_samples": aligned["train"].n,
+        "valid_samples": aligned["valid"].n,
+        "train_classes": np.bincount(aligned["train"].y_cls, minlength=3).tolist(),
+        "valid_classes": np.bincount(aligned["valid"].y_cls, minlength=3).tolist(),
+        "mean_observed_slots": {
+            MODALITIES[m]: float(aligned["train"].mask[:, :, m].sum(axis=1).mean()) for m in range(3)
+        },
+        "train_valid_video_overlap": 0,
+    }
+    with (output / "data_audit.json").open("w", encoding="utf-8") as stream:
+        json.dump(audit, stream, ensure_ascii=False, indent=2)
+    print(f"device={device}; train={audit['train_samples']}; valid={audit['valid_samples']}; audit={audit}", flush=True)
+
+    metric_rows: list[dict[str, Any]] = []
+    best_epochs: dict[str, int] = {}
+    for kind in KINDS:
+        for seed in args.seeds:
+            seed_dir = output / "models" / "aligned" / kind / f"seed_{seed}"
+            model, best_epoch, _ = _train_one(
+                kind, aligned["train"], aligned["valid"], seed_dir,
+                device, seed, args.epochs, args.patience, args.batch_size,
+            )
+            best_epochs[f"{kind}_seed_{seed}"] = best_epoch
+            metric_rows.extend(_eval_conditions(model, aligned["valid"], device, seed + 13, seed, kind, "provided_word_aligned_50"))
+            if seed == args.seeds[0]:
+                shutil.copy2(seed_dir / "model_best.pt", output / "models" / "aligned" / kind / "model_best.pt")
+            del model
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+
+    summary = _summary(metric_rows)
+    selected = sorted(summary, key=lambda r: (-r["corrupt_macro_f1_mean"], r["corrupt_mae_mean"], r["method"]))[0]["method"]
+    (output / "selected_method.txt").write_text(
+        f"Macro-F1-first validation selection: {selected}. See summary.csv for the full multi-metric tradeoff.\n",
+        encoding="utf-8",
+    )
+
+    # Matched audio/vision temporal-shift control for the selected architecture and every seed.
+    for seed in args.seeds:
+        aligned_payload = torch.load(output / "models" / "aligned" / selected / f"seed_{seed}" / "model_best.pt",
+                                     map_location=device, weights_only=False)
+        aligned_model = AlignedFusionModel(selected, tuple(aligned_payload["dims"])).to(device)
+        aligned_model.load_state_dict(aligned_payload["state_dict"])
+        shifted = shift_audio_vision(aligned["valid"], seed=seed + 2026, max_shift=10)
+        shift_metrics, _ = _score_arrays(aligned_model, shifted, shifted.mask, device)
+        metric_rows.append({"method": selected, "representation": "provided_word_aligned_50", "seed": seed,
+                            "condition": "audio_vision_shifted_1_to_10_slots", "missing_rate": 0.0,
+                            "n_valid": shifted.n, **shift_metrics})
+        del aligned_model
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+    # Same selected fusion architecture, but equal-window audio/vision pooling of the unaligned source.
+    print(f"selected_by_corrupt_macro_f1={selected}; starting fixed-window alignment control", flush=True)
+    fixed_raw = load_fixed_window()
+    fixed_stats = fit_robust_stats(fixed_raw["train"])
+    fixed_stats.save(output / "fixed_window_robust_stats.npz")
+    fixed = {k: apply_robust_stats(v, fixed_stats) for k, v in fixed_raw.items()}
+    for seed in args.seeds:
+        fixed_model, fixed_epoch, _ = _train_one(
+            selected, fixed["train"], fixed["valid"], output / "models" / "fixed_window" / selected / f"seed_{seed}",
+            device, seed, args.epochs, args.patience, args.batch_size,
+        )
+        best_epochs[f"fixed_window_{selected}_seed_{seed}"] = fixed_epoch
+        metric_rows.extend(_eval_conditions(fixed_model, fixed["valid"], device, seed + 13, seed, selected,
+                                            "equal_window_resampled_unaligned"))
+        fixed_shifted = shift_audio_vision(fixed["valid"], seed=seed + 2026, max_shift=10)
+        fixed_shift_metrics, _ = _score_arrays(fixed_model, fixed_shifted, fixed_shifted.mask, device)
+        metric_rows.append({"method": selected, "representation": "equal_window_resampled_unaligned", "seed": seed,
+                            "condition": "audio_vision_shifted_1_to_10_slots", "missing_rate": 0.0,
+                            "n_valid": fixed_shifted.n, **fixed_shift_metrics})
+        del fixed_model
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+    all_summary = _summary(metric_rows)
+    _write_csv(output / "validation_metrics_by_condition.csv", metric_rows)
+    _write_csv(output / "summary.csv", all_summary)
+    aligned_summary = [r for r in all_summary if r["representation"] == "provided_word_aligned_50"]
+    _plot(aligned_summary, metric_rows, output / "missing_rate_comparison.png")
+    alignment_rows = []
+    for rep in ("provided_word_aligned_50", "equal_window_resampled_unaligned"):
+        for condition in ("clean", "audio_vision_shifted_1_to_10_slots"):
+            match = [r for r in metric_rows if r["method"] == selected and r["representation"] == rep
+                     and r["condition"] == condition]
+            if match:
+                row = {"method": selected, "representation": rep, "condition": condition,
+                       "n_valid": aligned["valid"].n, "n_seeds": len(match)}
+                for metric in ("accuracy", "macro_f1", "mae", "pearson"):
+                    values = [r[metric] for r in match]
+                    row[metric] = float(np.mean(values))
+                    row[f"{metric}_sd"] = float(np.std(values, ddof=1)) if len(values) > 1 else 0.0
+                alignment_rows.append(row)
+        corrupt = [r for r in metric_rows if r["method"] == selected and r["representation"] == rep
+                   and r["condition"] != "clean" and r["missing_rate"] > 0]
+        if corrupt:
+            per_seed = []
+            for seed in args.seeds:
+                local = [r for r in corrupt if int(r["seed"]) == seed]
+                if local:
+                    per_seed.append({metric: float(np.mean([r[metric] for r in local])) for metric in
+                                     ("accuracy", "macro_f1", "mae", "pearson")})
+            alignment_rows.append({
+                "method": selected, "representation": rep, "condition": "all_local_corruption_mean",
+                "missing_rate": float(np.mean([r["missing_rate"] for r in corrupt])),
+                "n_valid": aligned["valid"].n, "n_seeds": len(per_seed),
+                **{metric: float(np.mean([r[metric] for r in per_seed])) for metric in ("accuracy", "macro_f1", "mae", "pearson")},
+                **{f"{metric}_sd": float(np.std([r[metric] for r in per_seed], ddof=1)) if len(per_seed) > 1 else 0.0
+                   for metric in ("accuracy", "macro_f1", "mae", "pearson")},
+            })
+    _write_csv(output / "alignment_transfer_ablation.csv", alignment_rows)
+
+    source_path = ATTACHMENT2 / "aligned_50.pkl"
+    manifest = {
+        "source_feature": str(source_path),
+        "source_sha256": _sha256(source_path),
+        "device": str(device),
+        "cuda_name": torch.cuda.get_device_name(0) if device.type == "cuda" else None,
+        "seeds": args.seeds,
+        "epochs_max": args.epochs,
+        "patience": args.patience,
+        "batch_size": args.batch_size,
+        "best_epochs": best_epochs,
+        "selected_macro_f1_first": selected,
+        "selection_policy": "report Macro-F1, MAE, and Pearson separately; selected model maximizes mean validation Macro-F1 across 15 contiguous corruption conditions, then uses MAE and lexical model name only as tie-breaks",
+        "models": list(KINDS),
+        "corruption_rates": [0.10, 0.20, 0.30],
+        "corruption_patterns": list(PATTERNS),
+        "feature_scaling": "training split median/MAD; fallback to standard deviation for zero-MAD dimensions",
+        "test_labels_used": False,
+        "alignment_transfer_limit": "The official aligned_50 data use a 50-slot wordpiece sequence with no per-slot seconds or stored Q1 B1 time_bounds. The fixed-window comparison is a downstream alignment control, not a re-run of Q1 B1 on the full dataset.",
+        "python": __import__("sys").version,
+        "torch": torch.__version__,
+        "numpy": np.__version__,
+        "created_unix": time.time(),
+    }
+    with (output / "run_manifest.json").open("w", encoding="utf-8") as stream:
+        json.dump(manifest, stream, ensure_ascii=False, indent=2)
+    print(f"saved selection artifacts to {output}; selected={selected}; seeds={args.seeds}", flush=True)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Q2 local-missingness model and alignment transfer selection")
+    parser.add_argument("--seeds", type=int, nargs="+", default=[42, 3407, 2026])
+    parser.add_argument("--epochs", type=int, default=32)
+    parser.add_argument("--patience", type=int, default=6)
+    parser.add_argument("--batch-size", type=int, default=64)
+    parser.add_argument("--threads", type=int, default=4)
+    parser.add_argument("--device", default="auto")
+    parser.add_argument("--output-dir", default=str(Path(__file__).resolve().parents[1] / "outputs" / "algorithm_selection"))
+    args = parser.parse_args()
+    _run(args)
+
+
+if __name__ == "__main__":
+    main()
				`@@ -0,0 +1 @@`
				`Macro-F1-first validation selection: concat. See summary.csv for the full multi-metric tradeoff.`
				`@@ -0,0 +1 @@`
				`"""Q2 robustness and Q3 explanation-selection experiments."""`