update single_core_regen settings new runs

data/single_core_regen.db

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7231dea2c9107f443de9122fdc971d9ce6df93db2ee27a9d68a5e22c986373eb
-size 937984
+oid sha256:f3510d41f9f0605e438a09767c43edda38162601292be1207f50747117ae5479
+size 9863168

src/single-core-regen/regen.py

@@ -1,4 +1,6 @@
 from datetime import datetime
+
+import optuna
 from hypertraining.hypertraining import HyperTraining
 from hypertraining.settings import (
     GlobalSettings,
@@ -10,59 +12,72 @@ from hypertraining.settings import (
 )
 
 global_settings = GlobalSettings(
-    seed = 42,
+    seed=42,
 )
 
 data_settings = DataSettings(
-    config_path     = "data/*-128-16384-100000-0-0-17-0-PAM4-0.ini",
-    dtype           = ("complex128", "complex64", "float64", "float32"),
-    symbols         = (1, 16),
-    model_input_dim = (1, 32),
-    shuffle         = True,
-    in_out_delay    = 0,
-    xy_delay        = 0,
-    drop_first      = 1000,
-    train_split     = 0.8,
+    config_path="data/*-128-16384-100000-0-0-17-0-PAM4-0.ini",
+    dtype="complex64",
+    # symbols         = (9, 20), # 13 symbol @ 10GBd <-> 1.3ns <-> 0.26m of fiber
+    symbols=13,  # study: single_core_regen_20241123_011232
+    # output_size     = (11, 32), # ballpark 26 taps -> 2 taps per input symbol -> 1 tap every 0.01m (model has 52 inputs)
+    output_size=26,  # study: single_core_regen_20241123_011232 (model_input_dim/2)
+    shuffle=True,
+    in_out_delay=0,
+    xy_delay=0,
+    drop_first=128 * 100,
+    train_split=0.8,
 )
 
 pytorch_settings = PytorchSettings(
-    epochs              = 25,
-    batchsize           = 2**10,
-    device              = "cuda",
-    dataloader_workers  = 2,
-    dataloader_prefetch = 2,
-    summary_dir         = ".runs",
-    write_every         = 2**5,
-    model_dir           = ".models",
+    epochs=10,
+    batchsize=2**10,
+    device="cuda",
+    dataloader_workers=2,
+    dataloader_prefetch=4,
+    summary_dir=".runs",
+    write_every=2**5,
+    save_models=True,
+    model_dir=".models",
 )
 
 model_settings = ModelSettings(
-    output_dim = 2,
-    model_n_layers = (2, 8),
-    unit_count  = (2, 16),
-    model_activation_func    = ("ModReLU")#, "ZReLU", "Mag")#, "CReLU", "Identity"),
+    output_dim=2,
+    # n_hidden_layers = (3, 8),
+    n_hidden_layers=(4, 6),  # study: single_core_regen_20241123_011232
+    n_hidden_nodes=(4,20),
+    # overrides={
+    #     "n_hidden_nodes_0": (14, 20),  # study: single_core_regen_20241123_011232
+    #     "n_hidden_nodes_1":  (8, 16),
+    #     "n_hidden_nodes_2":  (10, 16),
+    #     # "n_hidden_nodes_3":  (4, 20),  # study: single_core_regen_20241123_135749
+    #     "n_hidden_nodes_4": (2, 8),
+    #     "n_hidden_nodes_5": (10, 16),
+    #     },  
+    # model_activation_func    = ("ModReLU", "Mag", "Identity")
+    model_activation_func="Mag",  # study: single_core_regen_20241123_011232
 )
 
 optimizer_settings = OptimizerSettings(
-    optimizer           = ("Adam", "RMSprop"),#, "SGD"),
+    optimizer="Adam",
     # learning_rate       = (1e-5, 1e-1),
-    learning_rate=1e-3,
-    # scheduler           = "ReduceLROnPlateau",
-    # scheduler_kwargs    = {"mode": "min", "factor": 0.5, "patience": 10}
+    learning_rate=5e-4,
 )
 
 optuna_settings = OptunaSettings(
-    n_trials        = 4096,
-    n_threads       = 16,
-    timeout         = 600,
-    directions      = ("minimize","minimize"),
-    metrics_names   = ("n_params","mse"),
-
-    limit_examples  = True,
-    n_train_batches = 100,
-    n_valid_batches = 100,
-    storage         = "sqlite:///data/single_core_regen.db",
-    study_name      = f"single_core_regen_{datetime.now().strftime('%Y%m%d_%H%M%S')}",
+    n_trials=512,
+    n_workers=14,
+    timeout=3600,
+    directions=("maximize", "minimize"),
+    metrics_names=("neg_log_mse","n_nodes"),
+    limit_examples=True,
+    n_train_batches=500,
+    # n_valid_batches = 100,
+    storage="sqlite:///data/single_core_regen.db",
+    study_name=f"single_core_regen_{datetime.now().strftime('%Y%m%d_%H%M%S')}",
+    n_trials_filter=(optuna.trial.TrialState.COMPLETE, optuna.trial.TrialState.PRUNED),
+    pruner="MedianPruner",
+    pruner_kwargs=None
 )
 
 
@@ -78,8 +93,6 @@ if __name__ == "__main__":
 
     hyper_training.setup_study()
 
-    # hyper_training.resume_latest_study()
-
     hyper_training.run_study()
     # best_trial = hyper_training.study.best_trial