analysis/method1b.log

/workspace/ego6d_rag/method1b.py:18: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  fnet = Forecaster(cin=12, P=1).to(DEV); fnet.enc.load_state_dict(torch.load(AN / 'p3_fcx/shared_encoder.pt', map_location=DEV)); fnet.eval()
/workspace/ego6d_rag/method1b.py:19: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  cnet = CamFormer().to(DEV); cnet.load_state_dict(torch.load(AN / 'p3_cf/camformer.pt', map_location=DEV)); cnet.eval()
precompute train (CamFormer+forecasting modes, scene-leave-out) ...
/opt/conda/envs/trajgaze/lib/python3.10/site-packages/torch/nn/modules/transformer.py:409: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at ../aten/src/ATen/NestedTensorImpl.cpp:178.)
  output = torch._nested_tensor_from_mask(output, src_key_padding_mask.logical_not(), mask_check=False)
  pooled 149,127; training mass-gated dual head ...
  ep  0 Lpos=1.189 Lsel=1.469 Lroom=1.627 Lact=0.463 beta=0.86
  ep 30 Lpos=0.959 Lsel=1.090 Lroom=0.784 Lact=0.211 beta=0.28
  ep 60 Lpos=0.922 Lsel=1.044 Lroom=0.721 Lact=0.216 beta=0.28
  ep 90 Lpos=0.902 Lsel=1.018 Lroom=0.672 Lact=0.219 beta=0.27
  ep119 Lpos=0.890 Lsel=1.003 Lroom=0.634 Lact=0.221 beta=0.27

venue   mass  soft  hard oracle  selA  room  actF1
Loc_18  3.88  4.02  3.83   2.16  0.32 0.250  0.256
Loc_19  2.31  3.71  4.09   0.65  0.30 0.198  0.218
Loc_20  1.67  1.77  1.85   0.59  0.41 0.497  0.230
Loc_21  1.12  1.08  0.85   0.58  0.65 0.731  0.282
Loc_22  4.49  5.39  6.54   0.83  0.22 0.261  0.306
Loc_23  5.06  4.86  6.06   2.27  0.23 0.283  0.307
Loc_24  3.09  3.07  3.59   0.81  0.36 0.422  0.262
Loc_25  2.77  2.60  3.01   0.66  0.37 0.433  0.314
Loc_26  4.04  3.52  3.11   0.65  0.45 0.307  0.206
Loc_27  1.60  2.54  1.54   0.76  0.52 0.421  0.268
Loc_28  5.31  4.71  4.69   1.16  0.35 0.279  0.275
Loc_29  3.87  4.22  4.76   0.94  0.24 0.294  0.254
MACRO   3.27  3.46  3.66   1.01  0.37 0.365  0.265

[refs] FCx top-1 3.31 | Method1 soft 3.43/hard 3.77 selA 0.37 | mass-baseline 3.27 | oracle ~1.0
[sem refs] FCx probe room .462/actF1 .282 | CamFormer room .453/actF1 .353