refactor(predict): Adjust hyperparameter sampling probabilities for improved model tuning

Code/src/hysteretic_curves/predict_hysteretic_curves.py

@@ -602,11 +602,11 @@ def train_final_full_trainpool_cached(  # type: ignore
 # Hyperparameter sampler
 # -------------------------
 def sample_params(device: str = "cpu"):
-    window_size = int(np.random.choice([40, 60, 80]))
-    hidden_dim = int(np.random.choice([32, 64, 128, 256], p=[0.25, 0.35, 0.30, 0.10]))
-    dense_dim = int(np.random.choice([16, 32, 64, 128], p=[0.20, 0.40, 0.30, 0.10]))
-    num_layers = int(np.random.choice([1, 2, 3], p=[0.45, 0.40, 0.15]))
-    dropout = float(np.random.choice([0.0, 0.1, 0.2, 0.3]))
+    window_size = int(np.random.choice([40, 60], p=[0.6, 0.4]))
+    hidden_dim = int(np.random.choice([32, 64, 128], p=[0.30, 0.45, 0.25]))
+    dense_dim = int(np.random.choice([16, 32, 64], p=[0.25, 0.50, 0.25]))
+    num_layers = int(np.random.choice([1, 2], p=[0.55, 0.45]))
+    dropout = float(np.random.choice([0.0, 0.1, 0.2]))
 
     lr = float(10 ** np.random.uniform(-4.0, -2.6))
     weight_decay = float(10 ** np.random.uniform(-6.0, -3.5))