best val_rmse 0.0000

config.json

@@ -1,5 +1,5 @@
 {
-  "best_val_loss": 0.041541337966918945,
+  "best_val_loss": 0.0,
   "comet_ai_key": null,
   "context_observations": {
     "add_rem": true,
@@ -18,7 +18,7 @@
     "route": "oral",
     "time": 0.0
   },
-  "experiment_dir": "/work/ojedamarin/Projects/Pharma/Results/comet/node_pk_compartments/c6cf93e707f34c8ea42a7f879759b712",
+  "experiment_dir": "/work/ojedamarin/Projects/Pharma/Results/comet/node_pk_compartments/825c9c6329a941d4a74ea56913982ba2",
   "experiment_indentifier": null,
   "experiment_name": "node_pk_compartments",
   "hf_model_card_path": [
@@ -143,7 +143,7 @@
       "lenuzza",
       "Lenuzza2016.csv"
     ],
-    "pretraining_epochs": 800,
+    "pretraining_epochs": 2000,
     "pretraining_protocol": "none",
     "recreate_tempfile": true,
     "return_split_versions": true,
@@ -155,10 +155,11 @@
       "simulated_ou_as_rates"
     ],
     "test_protocol": "simulated",
-    "test_size": 50,
-    "train_size": 2000,
+    "test_size": 100,
+    "tqdm_progress": false,
+    "train_size": 10000,
     "val_protocol": "simulated",
-    "val_size": 50,
+    "val_size": 100,
     "z_score_normalization": false
   },
   "model_type": "node_pk",
@@ -166,8 +167,9 @@
   "name_str": "NeuralProcessPK",
   "network": {
     "activation": "ReLU",
-    "aggregator_num_heads": 8,
+    "aggregator_num_heads": 4,
     "aggregator_type": "attention",
+    "combine_latent_mode": "sum",
     "cov_proj_dim": 16,
     "decoder_attention_layers": 2,
     "decoder_hidden_dim": 512,
@@ -181,7 +183,7 @@
     "exclusive_node_step": true,
     "individual_encoder_name": "RNNContextEncoder",
     "individual_encoder_number_of_heads": 4,
-    "init_hidden_num_layers": 4,
+    "init_hidden_num_layers": 2,
     "input_encoding_hidden_dim": 128,
     "loss_name": "nll",
     "node_step": true,
@@ -197,11 +199,9 @@
     "use_kl_s": true,
     "zi_latent_dim": 512
   },
-  "run_index": 7,
+  "run_index": 0,
   "tags": [
-    "NP",
-    "OU",
-    "KL_ZI_PRED",
+    "YOLO-0",
     "N-0"
   ],
   "target_observations": {
@@ -217,13 +217,14 @@
   },
   "train": {
     "amsgrad": false,
-    "batch_size": 64,
+    "batch_size": 128,
     "betas": [
       0.9,
       0.999
     ],
-    "epochs": 1000,
+    "epochs": 2500,
     "eps": 1e-08,
+    "eval_empirical_every_pct": 0.1,
     "gradient_clip_val": 1.0,
     "learning_rate": 0.0001,
     "log_image_every_epoch": 25,
@@ -239,6 +240,7 @@
       "eta_min": 5e-05,
       "last_epoch": -1
     },
+    "shuffle_val": true,
     "weight_decay": 0.0001
   },
   "transformers_version": "4.52.4",

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47c30c02f894a83ad7e250db280b8d30da39e25cea30ec513bf2418057c6c4f9
-size 48200922
+oid sha256:4f92823e643df01d1b5e4bcde8942b8d81702b75231443afd84f82608fbe124e
+size 45044802