| lstm-time_concat_soft | lstm-no_time | lstm-time_concat_mlp | lstm-time_mask | |
|---|---|---|---|---|
| time encoding | soft one-hot | no time | mlp projection | time mask |
| min_measurement | 3 | 3 | 3 | 3 |
| cycle_len | 15 | 15 | 30 | 20 |
| batch_size | 4 | 12 | 20 | 32 |
| class_weight | 0.55 | 0.55 | 0.45 | 0.4 |
| optimizer | radam | adam | adam | adam |
| max_lr | 0.0008929 | 0.001375 | 0.000599 | 0.0007358 |
| weight_decay | 6.62E-05 | 7.58E-05 | 1.24E-06 | 7.30E-05 |
| patient_model-hidden_dims | 28 | 4 | 24 | 24 |
| patient_model-output_dim | 16 | 56 | 40 | 56 |
| patient_model-activation | gelu | relu | gelu | tanh |
| time_model-hidden_dims | None | None | 24 | None |
| time_model-projection_size | 24 | None | None | 28 |
| time_model-output_dim | 16 | 1 | 40 | None |
| event_model-continuous_hidden_dims | 20 | 32 | 28 | 8 |
| event_model-continuous_output_dim | 56 | 16 | 48 | 56 |
| event_model-tf_activation | relu | relu | gelu | gelu |
| temporal_model-hidden_size | 64 | 56 | 16 | 64 |
| temporal_model-num_layers | 3 | 2 | 1 | 1 |
| temporal_model-dropout | 0.5 | 0.1 | 0.1 | 0.5 |
| temporal_model-hidden_act | relu | tanh | relu | gelu |
| temporal_model-hidden_dropout_prob | 0.1 | 0.2 | 0.3 | 0.3 |
| classifier_model-hidden_dim | 64 | 32 | 64 | 24 |
| classifier_model-output_dropout | 0.5 | 0.1 | 0.4 | 0.3 |
| clstm-forget_output | clstm-output | clstm-forget | |
|---|---|---|---|
| temporal gate | forget + output | output | forget |
| min_measurement | 3 | 3 | 3 |
| cycle_len | 40 | 20 | 15 |
| batch_size | 8 | 16 | 32 |
| class_weight | 0.55 | 0.55 | 0.6 |
| optimizer | adam | adam | sgd |
| max_lr | 0.006557 | 0.003244 | 0.009452 |
| weight_decay | 3.45E-06 | 1.60E-05 | 2.99E-05 |
| patient_model-hidden_dims | 24 | 28 | 4 |
| patient_model-output_dim | 16 | 32 | 24 |
| patient_model-activation | relu | relu | gelu |
| time_model-output_dim | 1 | 1 | 1 |
| event_model-continuous_hidden_dims | 24 | 28 | 8 |
| event_model-continuous_output_dim | 56 | 64 | 16 |
| event_model-tf_activation | tanh | gelu | tanh |
| temporal_model-hidden_size | 8 | 48 | 56 |
| temporal_model-timedecay_size | 5 | 1 | 2 |
| temporal_model-dropout | 0.4 | 0.4 | 0.4 |
| temporal_model-hidden_act | tanh | tanh | tanh |
| classifier_model-hidden_dim | 32 | 24 | 40 |
| classifier_model-output_dropout | 0.3 | 0.1 | 0.2 |
| attention-time_encode | attention-time_concat_soft | attention-no_time | attention-time_concat_mlp | attention-time_mask | |
|---|---|---|---|---|---|
| time encoding | positional time | soft one-hot | no time | mlp projection | time mask |
| min_measurement | 3 | 3 | 3 | 3 | 3 |
| cycle_len | 15 | 15 | 30 | 35 | 5 |
| batch_size | 16 | 12 | 32 | 4 | 4 |
| class_weight | 0.7 | 0.4 | 0.5 | 0.35 | 0.65 |
| optimizer | adam | radam | sgd | adam | radam |
| max_lr | 0.001972 | 0.001261 | 0.000539 | 0.0001125 | 0.0008968 |
| weight_decay | 1.70E-06 | 7.75E-06 | 1.57E-06 | 3.23E-05 | 1.53E-06 |
| patient_model-hidden_dims | 16 | 32 | 24 | 16 | 12 |
| patient_model-output_dim | None | 56 | 40 | 32 | 32 |
| patient_model-activation | relu | gelu | relu | relu | tanh |
| time_model-hidden_dims | None | None | None | 16 | None |
| time_model-projection_size | None | 12 | None | None | 4 |
| time_model-output_dim | None | 56 | None | 32 | None |
| event_model-continuous_hidden_dims | 20 | 16 | 16 | 12 | 20 |
| event_model-continuous_output_dim | 24 | 40 | 64 | 24 | 64 |
| event_model-tf_activation | gelu | relu | gelu | tanh | tanh |
| temporal_model-num_layers | 3 | 4 | 4 | 5 | 1 |
| temporal_model-attn_dropout_prob | 0.4 | 0 | 0.2 | 0.2 | 0.2 |
| temporal_model-feed_forward_hidden | 8 | 56 | 56 | 40 | 16 |
| temporal_model-hidden_act | tanh | gelu | gelu | relu | relu |
| temporal_model-hidden_dropout_prob | 0.2 | 0.1 | 0 | 0.3 | 0.1 |
| classifier_model-hidden_dim | 16 | 24 | 24 | 16 | 24 |
| classifier_model-output_dropout | 0.4 | 0.5 | 0.3 | 0.1 | 0.3 |