optimal hyperparameters for cloud_parametric

Author: robin-janssen

datasets/cloud_parametric/surrogates_config.py

@@ -0,0 +1,140 @@
+from dataclasses import dataclass
+
+from torch import nn
+
+
+@dataclass
+class MultiONetConfig:
+    """Model config for MultiONet for the cloud_parametric dataset"""
+
+    # cloud_parametric_final_multionet, trial 27
+    scheduler: str = "poly"
+    optimizer: str = "AdamW"
+    loss_function: nn.Module = nn.MSELoss()
+    activation: nn.Module = nn.ELU()
+    branch_hidden_layers: int = 4
+    hidden_size: int = 100
+    learning_rate: float = 0.00184
+    output_factor: int = 40
+    params_branch: bool = True
+    poly_power: float = 1.52
+    regularization_factor: float = 0.0171
+    trunk_hidden_layers: int = 4
+
+
+@dataclass
+class LatentNeuralODEConfig:
+    """Model config for LatentNeuralODE for the cloud_parametric dataset"""
+
+    # cloud_parametric_final_latentneuralode, trial 299
+    scheduler: str = "schedulefree"
+    optimizer: str = "SGD"
+    loss_function: nn.Module = nn.SmoothL1Loss()
+    activation: nn.Module = nn.GELU()
+    beta: float = 9.61
+    coder_layers: int = 2
+    coder_width: int = 180
+    encode_params: bool = False
+    latent_features: int = 10
+    learning_rate: float = 0.00693
+    momentum: float = 0.613
+    ode_layers: int = 3
+    ode_tanh_reg: bool = False
+    ode_width: int = 220
+    regularization_factor: float = 0.000134
+
+
+@dataclass
+class FullyConnectedConfig:
+    """Model config for FullyConnected for the cloud_parametric dataset"""
+
+    # cloud_parametric_final_fullyconnected, trial 61
+    scheduler: str = "poly"
+    optimizer: str = "AdamW"
+    loss_function: nn.Module = nn.SmoothL1Loss()
+    activation: nn.Module = nn.ELU()
+    beta: float = 0.299
+    hidden_size: int = 290
+    learning_rate: float = 0.00331
+    num_hidden_layers: int = 5
+    poly_power: float = 1.88
+    regularization_factor: float = 0.113
+
+
+@dataclass
+class LatentPolyConfig:
+    """Model config for LatentPoly for the cloud_parametric dataset"""
+
+    # cloud_parametric_final_latentpoly, trial 13
+    scheduler: str = "poly"
+    optimizer: str = "AdamW"
+    loss_function: nn.Module = nn.SmoothL1Loss()
+    activation: nn.Module = nn.ReLU()
+    beta: float = 2.88
+    coder_layers: bool = True
+    coder_width: int = 170
+    coeff_network: bool = False
+    degree: int = 9
+    latent_features: int = 10
+    learning_rate: float = 0.00551
+    poly_power: float = 1.65
+    regularization_factor: float = 0.00629
+
+
+# @dataclass
+# class MultiONetConfig:
+#     """Model config for MultiONet for the simple_ode dataset"""
+
+#     # cloud, trial 69
+#     branch_hidden_layers: int = 1
+#     trunk_hidden_layers: int = 9
+#     hidden_size: int = 225
+#     output_factor: int = 63
+#     learning_rate: float = 4e-5  # optimal for ~4000 epochs
+#     activation: nn.Module = nn.Tanh()
+
+
+# @dataclass
+# class LatentNeuralODEConfig:
+#     """Model config for LatentNeuralODE for the simple_ode dataset"""
+
+#     # cloudparams, trial 40
+#     latent_features: int = 3
+#     coder_layers: int = 3
+#     coder_width: int = 377
+#     learning_rate: float = 3e-4
+#     ode_layers: int = 5
+#     ode_width: int = 167
+#     regularization_factor: float = 0.000127
+#     encode_params: bool = False
+#     optimizer: str = "sgd"
+#     momentum: float = 0.226
+#     scheduler: str = "cosine"
+#     eta_min: float = 0.0222
+#     ode_tanh_reg: bool = False
+#     activation: nn.Module = nn.SiLU()
+#     model_version: str = "v2"
+
+
+# @dataclass
+# class FullyConnectedConfig:
+#     """Model config for FullyConnected for the simple_ode dataset"""
+
+#     # cloud, trial 44
+#     hidden_size: int = 261
+#     num_hidden_layers: int = 1
+#     learning_rate: float = 1e-4
+#     activation: nn.Module = nn.LeakyReLU()
+
+
+# @dataclass
+# class LatentPolyConfig:
+#     """Model config for LatentPoly for the simple_ode dataset"""
+
+#     # cloud, trial 92
+#     latent_features: int = 9
+#     degree: int = 5
+#     learning_rate: float = 3e-4
+#     coder_layers: int = 1
+#     coder_width: int = 86
+#     activation: nn.Module = nn.Mish()