best val_rmse 0.2602

config.json

@@ -1,5 +1,5 @@
 {
-  "best_val_loss": 0.3713146150112152,
+  "best_val_loss": 0.2602488696575165,
   "comet_ai_key": null,
   "context_observations": {
     "add_rem": true,
@@ -18,7 +18,7 @@
     "route": "oral",
     "time": 0.0
   },
-  "experiment_dir": "/home/cesarali/Pharma/generative_pk/results/comet/node_pk_compartments/bf45c221eaf04550ba11d9f4fa61a39e",
+  "experiment_dir": "/home/cesarali/Pharma/generative_pk/results/comet/node_pk_compartments/86a8b13fe5d344da94c10dc3309139d0",
   "experiment_indentifier": null,
   "experiment_name": "node_pk_compartments",
   "hf_model_card_path": [
@@ -145,7 +145,7 @@
     ],
     "pretraining_epochs": 90,
     "pretraining_protocol": "none",
-    "recreate_tempfile": false,
+    "recreate_tempfile": true,
     "return_split_versions": true,
     "split_seed": 42,
     "split_strategy": "study",
@@ -155,11 +155,11 @@
       "simulated_ou_as_rates"
     ],
     "test_protocol": "simulated",
-    "test_size": 5,
+    "test_size": 10,
     "tqdm_progress": false,
-    "train_size": 10,
+    "train_size": 1000,
     "val_protocol": "simulated",
-    "val_size": 5,
+    "val_size": 10,
     "z_score_normalization": false
   },
   "model_type": "node_pk",
@@ -172,14 +172,14 @@
     "combine_latent_mode": "mlp",
     "cov_proj_dim": 16,
     "decoder_attention_layers": 2,
-    "decoder_hidden_dim": 32,
-    "decoder_name": "RNNDecoder",
+    "decoder_hidden_dim": 128,
+    "decoder_name": "TransformerDecoder",
     "decoder_num_layers": 2,
-    "decoder_rnn_hidden_dim": 20,
+    "decoder_rnn_hidden_dim": 200,
     "drift_activation": "Tanh",
-    "drift_num_layers": 2,
+    "drift_num_layers": 3,
     "dropout": 0.1,
-    "encoder_rnn_hidden_dim": 20,
+    "encoder_rnn_hidden_dim": 200,
     "exclusive_node_step": false,
     "individual_encoder_name": "RNNContextEncoder",
     "individual_encoder_number_of_heads": 4,
@@ -189,17 +189,17 @@
     "node_step": true,
     "norm": "layer",
     "output_head_num_layers": 2,
-    "rnn_decoder_number_of_layers": 2,
+    "rnn_decoder_number_of_layers": 4,
     "rnn_individual_encoder_number_of_layers": 2,
-    "time_obs_encoder_hidden_dim": 32,
-    "time_obs_encoder_output_dim": 32,
+    "time_obs_encoder_hidden_dim": 128,
+    "time_obs_encoder_output_dim": 128,
     "use_attention": true,
     "use_invariance_loss": true,
     "use_kl_i": true,
     "use_kl_i_np": true,
     "use_kl_init": true,
     "use_kl_s": true,
-    "zi_latent_dim": 20
+    "zi_latent_dim": 200
   },
   "run_index": 0,
   "tags": [
@@ -226,7 +226,7 @@
     ],
     "epochs": 5,
     "eps": 1e-08,
-    "eval_empirical_every_pct": 0.25,
+    "eval_empirical_every_pct": 0.5,
     "gradient_clip_val": 1.0,
     "learning_rate": 0.0001,
     "log_image_every_epoch": 2,

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:921fa4f0ba379906d23167c1850bd228f41b7ae0bb297873ec6909f58751da5a
-size 168701
+oid sha256:09ec0c73f5d2131327a6ea4ad9012db37b9377ee3d2f8421839b60c4e1a3b830
+size 6044367