Tested model:
=============
------- notif_encoder:
NotifEncoder(
  (model): Sequential(
    (0): Linear(in_features=29, out_features=64, bias=True)
    (1): ReLU()
    (2): Linear(in_features=64, out_features=32, bias=True)
    (3): Dropout(p=0.2, inplace=False)
  )
)

------- evnt_encoder:
EvntEncoder(
  (model): Sequential(
    (0): Linear(in_features=11, out_features=64, bias=True)
    (1): ReLU()
    (2): Linear(in_features=64, out_features=32, bias=True)
    (3): Dropout(p=0.2, inplace=False)
  )
)

------- embedder:
AttentionEmbedder(
  (adapter): Linear(in_features=32, out_features=32, bias=True)
  (representer): AttentionRepresenter(
    (models): ParameterList(
        (0): Object of type: AttentionRepresenterBlock
        (1): Object of type: AttentionRepresenterBlock
      (0): AttentionRepresenterBlock(
        (first_part_norm): BatchNorm1d(22, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (first_part_Q): Linear(in_features=32, out_features=160, bias=True)
        (first_part_K): Linear(in_features=32, out_features=160, bias=True)
        (first_part_V): Linear(in_features=32, out_features=160, bias=True)
        (first_part_att): MultiheadAttention(
          (out_proj): NonDynamicallyQuantizableLinear(in_features=160, out_features=160, bias=True)
        )
        (first_part_final): Sequential(
          (0): Linear(in_features=160, out_features=32, bias=True)
          (1): Dropout(p=0.2, inplace=False)
        )
        (second_part_norm): BatchNorm1d(22, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (second_part_mlp): Sequential(
          (0): Linear(in_features=32, out_features=64, bias=True)
          (1): GELU(approximate='none')
          (2): Dropout(p=0.2, inplace=False)
          (3): Linear(in_features=64, out_features=32, bias=True)
          (4): GELU(approximate='none')
          (5): Dropout(p=0.2, inplace=False)
        )
      )
      (1): AttentionRepresenterBlock(
        (first_part_norm): BatchNorm1d(22, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (first_part_Q): Linear(in_features=32, out_features=160, bias=True)
        (first_part_K): Linear(in_features=32, out_features=160, bias=True)
        (first_part_V): Linear(in_features=32, out_features=160, bias=True)
        (first_part_att): MultiheadAttention(
          (out_proj): NonDynamicallyQuantizableLinear(in_features=160, out_features=160, bias=True)
        )
        (first_part_final): Sequential(
          (0): Linear(in_features=160, out_features=32, bias=True)
          (1): Dropout(p=0.2, inplace=False)
        )
        (second_part_norm): BatchNorm1d(22, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (second_part_mlp): Sequential(
          (0): Linear(in_features=32, out_features=64, bias=True)
          (1): GELU(approximate='none')
          (2): Dropout(p=0.2, inplace=False)
          (3): Linear(in_features=64, out_features=32, bias=True)
          (4): GELU(approximate='none')
          (5): Dropout(p=0.2, inplace=False)
        )
      )
    )
  )
  (mlp): Sequential(
    (0): Linear(in_features=64, out_features=64, bias=True)
    (1): ReLU()
    (2): Dropout(p=0.2, inplace=False)
    (3): Linear(in_features=64, out_features=32, bias=True)
  )
)

(Training dataset class occurences: {3: 176, 1: 46, 4: 45, 2: 1}.)
svm finished: True; SV cnt 74;
Trained SVM with LINEAR kernel with loss 720449.2712342693 and 74 outliers.
svm finished: False; SV cnt 40;
Trained SVM with POLYNOMIAL (deg 2) kernel with loss 400000.0 and 40 outliers.
svm finished: True; SV cnt 74;
Trained SVM with GAUSSIAN (sigma 1) kernel with loss 400000.0 and 40 outliers.
Results of classification of specially annotated notifications:
  IDs of clssified notificaions: [11168292, 11169523, 11173814, 9961382, 9962202, 9971171, 10168557, 10213020, 10275769]
  classes of classified notifications: [3, 3, 1, 3, 3, 1, 3, 2, 3]
  SVM with LINEAR kernel: tensor([-1., -1., -1., -1., -1., -1., -1., -1., -1.], dtype=torch.float64)
  SVM with POLYNOMIAL kernel: tensor([0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=torch.float64)
  SVM with GAUSSIAN kernel: tensor([-1., -1., -1., -1., -1., -1., -1., -1., -1.], dtype=torch.float64)
  logistic regression: tensor([[1.0000e+00, 8.7569e-14],
        [1.0000e+00, 8.7569e-14],
        [1.0000e+00, 8.7569e-14],
        [1.0000e+00, 8.7569e-14],
        [1.0000e+00, 8.7569e-14],
        [1.0000e+00, 8.7569e-14],
        [1.0000e+00, 8.7572e-14],
        [1.0000e+00, 8.7572e-14],
        [1.0000e+00, 8.7572e-14]], grad_fn=<SoftmaxBackward0>)
  MLP classifier: tensor([[1.0000e+00, 2.7591e-16],
        [1.0000e+00, 2.7591e-16],
        [1.0000e+00, 2.7591e-16],
        [1.0000e+00, 2.7591e-16],
        [1.0000e+00, 2.7591e-16],
        [1.0000e+00, 2.7591e-16],
        [1.0000e+00, 2.7592e-16],
        [1.0000e+00, 2.7592e-16],
        [1.0000e+00, 2.7592e-16]], grad_fn=<SoftmaxBackward0>)
Results on test set:
  SVM with LINEAR kernel: 0.0 (0/13)
  SVM with POLYNOMIAL kernel: 0.0 (0/13)
  SVM with GAUSSIAN kernel: 0.0 (0/13)
  logistic regression: 1.0 (13/13)
  MLP classifier: 1.0 (13/13)
  (class occurence counts: {3: 11, 4: 2})