增加LiON的评估指标

data/external/all_amine_split_for_LiON/cv_0/args.json

@@ -0,0 +1,165 @@
+{
+    "activation": "ReLU",
+    "adding_bond_types": true,
+    "adding_h": false,
+    "aggregation": "mean",
+    "aggregation_norm": 100,
+    "atom_constraints": [],
+    "atom_descriptor_scaling": true,
+    "atom_descriptors": null,
+    "atom_descriptors_path": null,
+    "atom_descriptors_size": 0,
+    "atom_features_size": 0,
+    "atom_messages": false,
+    "atom_targets": [],
+    "batch_size": 50,
+    "bias": false,
+    "bias_solvent": false,
+    "bond_constraints": [],
+    "bond_descriptor_scaling": true,
+    "bond_descriptors": null,
+    "bond_descriptors_path": null,
+    "bond_descriptors_size": 0,
+    "bond_features_size": 0,
+    "bond_targets": [],
+    "cache_cutoff": 10000,
+    "checkpoint_dir": null,
+    "checkpoint_frzn": null,
+    "checkpoint_path": null,
+    "checkpoint_paths": null,
+    "class_balance": false,
+    "config_path": "../data/args_files/optimized_configs.json",
+    "constraints_path": null,
+    "crossval_index_dir": null,
+    "crossval_index_file": null,
+    "crossval_index_sets": null,
+    "cuda": true,
+    "data_path": "../data/crossval_splits/all_amine_split_for_paper/cv_0/train.csv",
+    "data_weights_path": "../data/crossval_splits/all_amine_split_for_paper/cv_0/train_weights.csv",
+    "dataset_type": "regression",
+    "depth": 4,
+    "depth_solvent": 3,
+    "device": {
+        "_string": "cuda",
+        "_type": "python_object (type = device)",
+        "_value": "gASVHwAAAAAAAACMBXRvcmNolIwGZGV2aWNllJOUjARjdWRhlIWUUpQu"
+    },
+    "dropout": 0.1,
+    "empty_cache": false,
+    "ensemble_size": 1,
+    "epochs": 50,
+    "evidential_regularization": 0,
+    "explicit_h": false,
+    "extra_metrics": [],
+    "features_generator": null,
+    "features_only": false,
+    "features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_0/train_extra_x.csv"
+    ],
+    "features_scaling": true,
+    "features_size": null,
+    "ffn_hidden_size": 600,
+    "ffn_num_layers": 3,
+    "final_lr": 0.0001,
+    "folds_file": null,
+    "freeze_first_only": false,
+    "frzn_ffn_layers": 0,
+    "gpu": null,
+    "grad_clip": null,
+    "hidden_size": 600,
+    "hidden_size_solvent": 300,
+    "ignore_columns": null,
+    "ignore_nan_metrics": false,
+    "init_lr": 0.0001,
+    "is_atom_bond_targets": false,
+    "keeping_atom_map": false,
+    "log_frequency": 10,
+    "loss_function": "mse",
+    "max_data_size": null,
+    "max_lr": 0.001,
+    "metric": "rmse",
+    "metrics": [
+        "rmse"
+    ],
+    "minimize_score": true,
+    "mpn_shared": false,
+    "multiclass_num_classes": 3,
+    "no_adding_bond_types": false,
+    "no_atom_descriptor_scaling": false,
+    "no_bond_descriptor_scaling": false,
+    "no_cache_mol": false,
+    "no_cuda": false,
+    "no_features_scaling": false,
+    "no_shared_atom_bond_ffn": false,
+    "num_folds": 1,
+    "num_lrs": 1,
+    "num_tasks": 1,
+    "num_workers": 8,
+    "number_of_molecules": 1,
+    "overwrite_default_atom_features": false,
+    "overwrite_default_bond_features": false,
+    "phase_features_path": null,
+    "pytorch_seed": 0,
+    "quantile_loss_alpha": 0.1,
+    "quantiles": [],
+    "quiet": false,
+    "reaction": false,
+    "reaction_mode": "reac_diff",
+    "reaction_solvent": false,
+    "reproducibility": {
+        "command_line": "python main_script.py train all_amine_split_for_paper",
+        "git_has_uncommitted_changes": true,
+        "git_root": "/media/andersonxps/wd_4tb/evan/LNP_ML",
+        "git_url": "https://github.com/jswitten/LNP_ML/tree/167822980dc26ba65c5c14539c4ce12b81b0b8f3",
+        "time": "Tue Jul 30 10:15:25 2024"
+    },
+    "resume_experiment": false,
+    "save_dir": "../data/crossval_splits/all_amine_split_for_paper/cv_0",
+    "save_preds": false,
+    "save_smiles_splits": false,
+    "seed": 42,
+    "separate_test_atom_descriptors_path": null,
+    "separate_test_bond_descriptors_path": null,
+    "separate_test_constraints_path": null,
+    "separate_test_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_0/test_extra_x.csv"
+    ],
+    "separate_test_path": "../data/crossval_splits/all_amine_split_for_paper/cv_0/test.csv",
+    "separate_test_phase_features_path": null,
+    "separate_val_atom_descriptors_path": null,
+    "separate_val_bond_descriptors_path": null,
+    "separate_val_constraints_path": null,
+    "separate_val_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_0/valid_extra_x.csv"
+    ],
+    "separate_val_path": "../data/crossval_splits/all_amine_split_for_paper/cv_0/valid.csv",
+    "separate_val_phase_features_path": null,
+    "shared_atom_bond_ffn": true,
+    "show_individual_scores": false,
+    "smiles_columns": [
+        "smiles"
+    ],
+    "spectra_activation": "exp",
+    "spectra_phase_mask_path": null,
+    "spectra_target_floor": 1e-08,
+    "split_key_molecule": 0,
+    "split_sizes": [
+        1.0,
+        0.0,
+        0.0
+    ],
+    "split_type": "random",
+    "target_columns": null,
+    "target_weights": null,
+    "task_names": [
+        "quantified_delivery"
+    ],
+    "test": false,
+    "test_fold_index": null,
+    "train_data_size": null,
+    "undirected": false,
+    "use_input_features": true,
+    "val_fold_index": null,
+    "warmup_epochs": 2.0,
+    "weights_ffn_num_layers": 2
+}

data/external/all_amine_split_for_LiON/cv_0/fold_0/test_scores.json

@@ -0,0 +1,5 @@
+{
+    "rmse": [
+        0.8880622451903801
+    ]
+}

data/external/all_amine_split_for_LiON/cv_0/test_scores.csv

@@ -0,0 +1,2 @@
+Task,Mean rmse,Standard deviation rmse,Fold 0 rmse
+quantified_delivery,0.8880622451903801,0.0,0.8880622451903801

data/external/all_amine_split_for_LiON/cv_1/args.json

@@ -0,0 +1,165 @@
+{
+    "activation": "ReLU",
+    "adding_bond_types": true,
+    "adding_h": false,
+    "aggregation": "mean",
+    "aggregation_norm": 100,
+    "atom_constraints": [],
+    "atom_descriptor_scaling": true,
+    "atom_descriptors": null,
+    "atom_descriptors_path": null,
+    "atom_descriptors_size": 0,
+    "atom_features_size": 0,
+    "atom_messages": false,
+    "atom_targets": [],
+    "batch_size": 50,
+    "bias": false,
+    "bias_solvent": false,
+    "bond_constraints": [],
+    "bond_descriptor_scaling": true,
+    "bond_descriptors": null,
+    "bond_descriptors_path": null,
+    "bond_descriptors_size": 0,
+    "bond_features_size": 0,
+    "bond_targets": [],
+    "cache_cutoff": 10000,
+    "checkpoint_dir": null,
+    "checkpoint_frzn": null,
+    "checkpoint_path": null,
+    "checkpoint_paths": null,
+    "class_balance": false,
+    "config_path": "../data/args_files/optimized_configs.json",
+    "constraints_path": null,
+    "crossval_index_dir": null,
+    "crossval_index_file": null,
+    "crossval_index_sets": null,
+    "cuda": true,
+    "data_path": "../data/crossval_splits/all_amine_split_for_paper/cv_1/train.csv",
+    "data_weights_path": "../data/crossval_splits/all_amine_split_for_paper/cv_1/train_weights.csv",
+    "dataset_type": "regression",
+    "depth": 4,
+    "depth_solvent": 3,
+    "device": {
+        "_string": "cuda",
+        "_type": "python_object (type = device)",
+        "_value": "gASVHwAAAAAAAACMBXRvcmNolIwGZGV2aWNllJOUjARjdWRhlIWUUpQu"
+    },
+    "dropout": 0.1,
+    "empty_cache": false,
+    "ensemble_size": 1,
+    "epochs": 50,
+    "evidential_regularization": 0,
+    "explicit_h": false,
+    "extra_metrics": [],
+    "features_generator": null,
+    "features_only": false,
+    "features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_1/train_extra_x.csv"
+    ],
+    "features_scaling": true,
+    "features_size": null,
+    "ffn_hidden_size": 600,
+    "ffn_num_layers": 3,
+    "final_lr": 0.0001,
+    "folds_file": null,
+    "freeze_first_only": false,
+    "frzn_ffn_layers": 0,
+    "gpu": null,
+    "grad_clip": null,
+    "hidden_size": 600,
+    "hidden_size_solvent": 300,
+    "ignore_columns": null,
+    "ignore_nan_metrics": false,
+    "init_lr": 0.0001,
+    "is_atom_bond_targets": false,
+    "keeping_atom_map": false,
+    "log_frequency": 10,
+    "loss_function": "mse",
+    "max_data_size": null,
+    "max_lr": 0.001,
+    "metric": "rmse",
+    "metrics": [
+        "rmse"
+    ],
+    "minimize_score": true,
+    "mpn_shared": false,
+    "multiclass_num_classes": 3,
+    "no_adding_bond_types": false,
+    "no_atom_descriptor_scaling": false,
+    "no_bond_descriptor_scaling": false,
+    "no_cache_mol": false,
+    "no_cuda": false,
+    "no_features_scaling": false,
+    "no_shared_atom_bond_ffn": false,
+    "num_folds": 1,
+    "num_lrs": 1,
+    "num_tasks": 1,
+    "num_workers": 8,
+    "number_of_molecules": 1,
+    "overwrite_default_atom_features": false,
+    "overwrite_default_bond_features": false,
+    "phase_features_path": null,
+    "pytorch_seed": 0,
+    "quantile_loss_alpha": 0.1,
+    "quantiles": [],
+    "quiet": false,
+    "reaction": false,
+    "reaction_mode": "reac_diff",
+    "reaction_solvent": false,
+    "reproducibility": {
+        "command_line": "python main_script.py train all_amine_split_for_paper",
+        "git_has_uncommitted_changes": true,
+        "git_root": "/media/andersonxps/wd_4tb/evan/LNP_ML",
+        "git_url": "https://github.com/jswitten/LNP_ML/tree/167822980dc26ba65c5c14539c4ce12b81b0b8f3",
+        "time": "Tue Jul 30 10:21:40 2024"
+    },
+    "resume_experiment": false,
+    "save_dir": "../data/crossval_splits/all_amine_split_for_paper/cv_1",
+    "save_preds": false,
+    "save_smiles_splits": false,
+    "seed": 42,
+    "separate_test_atom_descriptors_path": null,
+    "separate_test_bond_descriptors_path": null,
+    "separate_test_constraints_path": null,
+    "separate_test_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_1/test_extra_x.csv"
+    ],
+    "separate_test_path": "../data/crossval_splits/all_amine_split_for_paper/cv_1/test.csv",
+    "separate_test_phase_features_path": null,
+    "separate_val_atom_descriptors_path": null,
+    "separate_val_bond_descriptors_path": null,
+    "separate_val_constraints_path": null,
+    "separate_val_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_1/valid_extra_x.csv"
+    ],
+    "separate_val_path": "../data/crossval_splits/all_amine_split_for_paper/cv_1/valid.csv",
+    "separate_val_phase_features_path": null,
+    "shared_atom_bond_ffn": true,
+    "show_individual_scores": false,
+    "smiles_columns": [
+        "smiles"
+    ],
+    "spectra_activation": "exp",
+    "spectra_phase_mask_path": null,
+    "spectra_target_floor": 1e-08,
+    "split_key_molecule": 0,
+    "split_sizes": [
+        1.0,
+        0.0,
+        0.0
+    ],
+    "split_type": "random",
+    "target_columns": null,
+    "target_weights": null,
+    "task_names": [
+        "quantified_delivery"
+    ],
+    "test": false,
+    "test_fold_index": null,
+    "train_data_size": null,
+    "undirected": false,
+    "use_input_features": true,
+    "val_fold_index": null,
+    "warmup_epochs": 2.0,
+    "weights_ffn_num_layers": 2
+}

data/external/all_amine_split_for_LiON/cv_1/fold_0/test_scores.json

@@ -0,0 +1,5 @@
+{
+    "rmse": [
+        1.01673724295223
+    ]
+}

data/external/all_amine_split_for_LiON/cv_1/test_scores.csv

@@ -0,0 +1,2 @@
+Task,Mean rmse,Standard deviation rmse,Fold 0 rmse
+quantified_delivery,1.01673724295223,0.0,1.01673724295223

data/external/all_amine_split_for_LiON/cv_2/args.json

@@ -0,0 +1,165 @@
+{
+    "activation": "ReLU",
+    "adding_bond_types": true,
+    "adding_h": false,
+    "aggregation": "mean",
+    "aggregation_norm": 100,
+    "atom_constraints": [],
+    "atom_descriptor_scaling": true,
+    "atom_descriptors": null,
+    "atom_descriptors_path": null,
+    "atom_descriptors_size": 0,
+    "atom_features_size": 0,
+    "atom_messages": false,
+    "atom_targets": [],
+    "batch_size": 50,
+    "bias": false,
+    "bias_solvent": false,
+    "bond_constraints": [],
+    "bond_descriptor_scaling": true,
+    "bond_descriptors": null,
+    "bond_descriptors_path": null,
+    "bond_descriptors_size": 0,
+    "bond_features_size": 0,
+    "bond_targets": [],
+    "cache_cutoff": 10000,
+    "checkpoint_dir": null,
+    "checkpoint_frzn": null,
+    "checkpoint_path": null,
+    "checkpoint_paths": null,
+    "class_balance": false,
+    "config_path": "../data/args_files/optimized_configs.json",
+    "constraints_path": null,
+    "crossval_index_dir": null,
+    "crossval_index_file": null,
+    "crossval_index_sets": null,
+    "cuda": true,
+    "data_path": "../data/crossval_splits/all_amine_split_for_paper/cv_2/train.csv",
+    "data_weights_path": "../data/crossval_splits/all_amine_split_for_paper/cv_2/train_weights.csv",
+    "dataset_type": "regression",
+    "depth": 4,
+    "depth_solvent": 3,
+    "device": {
+        "_string": "cuda",
+        "_type": "python_object (type = device)",
+        "_value": "gASVHwAAAAAAAACMBXRvcmNolIwGZGV2aWNllJOUjARjdWRhlIWUUpQu"
+    },
+    "dropout": 0.1,
+    "empty_cache": false,
+    "ensemble_size": 1,
+    "epochs": 50,
+    "evidential_regularization": 0,
+    "explicit_h": false,
+    "extra_metrics": [],
+    "features_generator": null,
+    "features_only": false,
+    "features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_2/train_extra_x.csv"
+    ],
+    "features_scaling": true,
+    "features_size": null,
+    "ffn_hidden_size": 600,
+    "ffn_num_layers": 3,
+    "final_lr": 0.0001,
+    "folds_file": null,
+    "freeze_first_only": false,
+    "frzn_ffn_layers": 0,
+    "gpu": null,
+    "grad_clip": null,
+    "hidden_size": 600,
+    "hidden_size_solvent": 300,
+    "ignore_columns": null,
+    "ignore_nan_metrics": false,
+    "init_lr": 0.0001,
+    "is_atom_bond_targets": false,
+    "keeping_atom_map": false,
+    "log_frequency": 10,
+    "loss_function": "mse",
+    "max_data_size": null,
+    "max_lr": 0.001,
+    "metric": "rmse",
+    "metrics": [
+        "rmse"
+    ],
+    "minimize_score": true,
+    "mpn_shared": false,
+    "multiclass_num_classes": 3,
+    "no_adding_bond_types": false,
+    "no_atom_descriptor_scaling": false,
+    "no_bond_descriptor_scaling": false,
+    "no_cache_mol": false,
+    "no_cuda": false,
+    "no_features_scaling": false,
+    "no_shared_atom_bond_ffn": false,
+    "num_folds": 1,
+    "num_lrs": 1,
+    "num_tasks": 1,
+    "num_workers": 8,
+    "number_of_molecules": 1,
+    "overwrite_default_atom_features": false,
+    "overwrite_default_bond_features": false,
+    "phase_features_path": null,
+    "pytorch_seed": 0,
+    "quantile_loss_alpha": 0.1,
+    "quantiles": [],
+    "quiet": false,
+    "reaction": false,
+    "reaction_mode": "reac_diff",
+    "reaction_solvent": false,
+    "reproducibility": {
+        "command_line": "python main_script.py train all_amine_split_for_paper",
+        "git_has_uncommitted_changes": true,
+        "git_root": "/media/andersonxps/wd_4tb/evan/LNP_ML",
+        "git_url": "https://github.com/jswitten/LNP_ML/tree/167822980dc26ba65c5c14539c4ce12b81b0b8f3",
+        "time": "Tue Jul 30 10:28:04 2024"
+    },
+    "resume_experiment": false,
+    "save_dir": "../data/crossval_splits/all_amine_split_for_paper/cv_2",
+    "save_preds": false,
+    "save_smiles_splits": false,
+    "seed": 42,
+    "separate_test_atom_descriptors_path": null,
+    "separate_test_bond_descriptors_path": null,
+    "separate_test_constraints_path": null,
+    "separate_test_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_2/test_extra_x.csv"
+    ],
+    "separate_test_path": "../data/crossval_splits/all_amine_split_for_paper/cv_2/test.csv",
+    "separate_test_phase_features_path": null,
+    "separate_val_atom_descriptors_path": null,
+    "separate_val_bond_descriptors_path": null,
+    "separate_val_constraints_path": null,
+    "separate_val_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_2/valid_extra_x.csv"
+    ],
+    "separate_val_path": "../data/crossval_splits/all_amine_split_for_paper/cv_2/valid.csv",
+    "separate_val_phase_features_path": null,
+    "shared_atom_bond_ffn": true,
+    "show_individual_scores": false,
+    "smiles_columns": [
+        "smiles"
+    ],
+    "spectra_activation": "exp",
+    "spectra_phase_mask_path": null,
+    "spectra_target_floor": 1e-08,
+    "split_key_molecule": 0,
+    "split_sizes": [
+        1.0,
+        0.0,
+        0.0
+    ],
+    "split_type": "random",
+    "target_columns": null,
+    "target_weights": null,
+    "task_names": [
+        "quantified_delivery"
+    ],
+    "test": false,
+    "test_fold_index": null,
+    "train_data_size": null,
+    "undirected": false,
+    "use_input_features": true,
+    "val_fold_index": null,
+    "warmup_epochs": 2.0,
+    "weights_ffn_num_layers": 2
+}

data/external/all_amine_split_for_LiON/cv_2/fold_0/test_scores.json

@@ -0,0 +1,5 @@
+{
+    "rmse": [
+        0.8788072588544181
+    ]
+}

data/external/all_amine_split_for_LiON/cv_2/test_scores.csv

@@ -0,0 +1,2 @@
+Task,Mean rmse,Standard deviation rmse,Fold 0 rmse
+quantified_delivery,0.8788072588544181,0.0,0.8788072588544181

data/external/all_amine_split_for_LiON/cv_3/args.json

@@ -0,0 +1,165 @@
+{
+    "activation": "ReLU",
+    "adding_bond_types": true,
+    "adding_h": false,
+    "aggregation": "mean",
+    "aggregation_norm": 100,
+    "atom_constraints": [],
+    "atom_descriptor_scaling": true,
+    "atom_descriptors": null,
+    "atom_descriptors_path": null,
+    "atom_descriptors_size": 0,
+    "atom_features_size": 0,
+    "atom_messages": false,
+    "atom_targets": [],
+    "batch_size": 50,
+    "bias": false,
+    "bias_solvent": false,
+    "bond_constraints": [],
+    "bond_descriptor_scaling": true,
+    "bond_descriptors": null,
+    "bond_descriptors_path": null,
+    "bond_descriptors_size": 0,
+    "bond_features_size": 0,
+    "bond_targets": [],
+    "cache_cutoff": 10000,
+    "checkpoint_dir": null,
+    "checkpoint_frzn": null,
+    "checkpoint_path": null,
+    "checkpoint_paths": null,
+    "class_balance": false,
+    "config_path": "../data/args_files/optimized_configs.json",
+    "constraints_path": null,
+    "crossval_index_dir": null,
+    "crossval_index_file": null,
+    "crossval_index_sets": null,
+    "cuda": true,
+    "data_path": "../data/crossval_splits/all_amine_split_for_paper/cv_3/train.csv",
+    "data_weights_path": "../data/crossval_splits/all_amine_split_for_paper/cv_3/train_weights.csv",
+    "dataset_type": "regression",
+    "depth": 4,
+    "depth_solvent": 3,
+    "device": {
+        "_string": "cuda",
+        "_type": "python_object (type = device)",
+        "_value": "gASVHwAAAAAAAACMBXRvcmNolIwGZGV2aWNllJOUjARjdWRhlIWUUpQu"
+    },
+    "dropout": 0.1,
+    "empty_cache": false,
+    "ensemble_size": 1,
+    "epochs": 50,
+    "evidential_regularization": 0,
+    "explicit_h": false,
+    "extra_metrics": [],
+    "features_generator": null,
+    "features_only": false,
+    "features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_3/train_extra_x.csv"
+    ],
+    "features_scaling": true,
+    "features_size": null,
+    "ffn_hidden_size": 600,
+    "ffn_num_layers": 3,
+    "final_lr": 0.0001,
+    "folds_file": null,
+    "freeze_first_only": false,
+    "frzn_ffn_layers": 0,
+    "gpu": null,
+    "grad_clip": null,
+    "hidden_size": 600,
+    "hidden_size_solvent": 300,
+    "ignore_columns": null,
+    "ignore_nan_metrics": false,
+    "init_lr": 0.0001,
+    "is_atom_bond_targets": false,
+    "keeping_atom_map": false,
+    "log_frequency": 10,
+    "loss_function": "mse",
+    "max_data_size": null,
+    "max_lr": 0.001,
+    "metric": "rmse",
+    "metrics": [
+        "rmse"
+    ],
+    "minimize_score": true,
+    "mpn_shared": false,
+    "multiclass_num_classes": 3,
+    "no_adding_bond_types": false,
+    "no_atom_descriptor_scaling": false,
+    "no_bond_descriptor_scaling": false,
+    "no_cache_mol": false,
+    "no_cuda": false,
+    "no_features_scaling": false,
+    "no_shared_atom_bond_ffn": false,
+    "num_folds": 1,
+    "num_lrs": 1,
+    "num_tasks": 1,
+    "num_workers": 8,
+    "number_of_molecules": 1,
+    "overwrite_default_atom_features": false,
+    "overwrite_default_bond_features": false,
+    "phase_features_path": null,
+    "pytorch_seed": 0,
+    "quantile_loss_alpha": 0.1,
+    "quantiles": [],
+    "quiet": false,
+    "reaction": false,
+    "reaction_mode": "reac_diff",
+    "reaction_solvent": false,
+    "reproducibility": {
+        "command_line": "python main_script.py train all_amine_split_for_paper",
+        "git_has_uncommitted_changes": true,
+        "git_root": "/media/andersonxps/wd_4tb/evan/LNP_ML",
+        "git_url": "https://github.com/jswitten/LNP_ML/tree/167822980dc26ba65c5c14539c4ce12b81b0b8f3",
+        "time": "Tue Jul 30 10:34:31 2024"
+    },
+    "resume_experiment": false,
+    "save_dir": "../data/crossval_splits/all_amine_split_for_paper/cv_3",
+    "save_preds": false,
+    "save_smiles_splits": false,
+    "seed": 42,
+    "separate_test_atom_descriptors_path": null,
+    "separate_test_bond_descriptors_path": null,
+    "separate_test_constraints_path": null,
+    "separate_test_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_3/test_extra_x.csv"
+    ],
+    "separate_test_path": "../data/crossval_splits/all_amine_split_for_paper/cv_3/test.csv",
+    "separate_test_phase_features_path": null,
+    "separate_val_atom_descriptors_path": null,
+    "separate_val_bond_descriptors_path": null,
+    "separate_val_constraints_path": null,
+    "separate_val_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_3/valid_extra_x.csv"
+    ],
+    "separate_val_path": "../data/crossval_splits/all_amine_split_for_paper/cv_3/valid.csv",
+    "separate_val_phase_features_path": null,
+    "shared_atom_bond_ffn": true,
+    "show_individual_scores": false,
+    "smiles_columns": [
+        "smiles"
+    ],
+    "spectra_activation": "exp",
+    "spectra_phase_mask_path": null,
+    "spectra_target_floor": 1e-08,
+    "split_key_molecule": 0,
+    "split_sizes": [
+        1.0,
+        0.0,
+        0.0
+    ],
+    "split_type": "random",
+    "target_columns": null,
+    "target_weights": null,
+    "task_names": [
+        "quantified_delivery"
+    ],
+    "test": false,
+    "test_fold_index": null,
+    "train_data_size": null,
+    "undirected": false,
+    "use_input_features": true,
+    "val_fold_index": null,
+    "warmup_epochs": 2.0,
+    "weights_ffn_num_layers": 2
+}

data/external/all_amine_split_for_LiON/cv_3/fold_0/test_scores.json

@@ -0,0 +1,5 @@
+{
+    "rmse": [
+        0.9245934905333985
+    ]
+}

data/external/all_amine_split_for_LiON/cv_3/test_scores.csv

@@ -0,0 +1,2 @@
+Task,Mean rmse,Standard deviation rmse,Fold 0 rmse
+quantified_delivery,0.9245934905333985,0.0,0.9245934905333985

data/external/all_amine_split_for_LiON/cv_4/args.json

@@ -0,0 +1,165 @@
+{
+    "activation": "ReLU",
+    "adding_bond_types": true,
+    "adding_h": false,
+    "aggregation": "mean",
+    "aggregation_norm": 100,
+    "atom_constraints": [],
+    "atom_descriptor_scaling": true,
+    "atom_descriptors": null,
+    "atom_descriptors_path": null,
+    "atom_descriptors_size": 0,
+    "atom_features_size": 0,
+    "atom_messages": false,
+    "atom_targets": [],
+    "batch_size": 50,
+    "bias": false,
+    "bias_solvent": false,
+    "bond_constraints": [],
+    "bond_descriptor_scaling": true,
+    "bond_descriptors": null,
+    "bond_descriptors_path": null,
+    "bond_descriptors_size": 0,
+    "bond_features_size": 0,
+    "bond_targets": [],
+    "cache_cutoff": 10000,
+    "checkpoint_dir": null,
+    "checkpoint_frzn": null,
+    "checkpoint_path": null,
+    "checkpoint_paths": null,
+    "class_balance": false,
+    "config_path": "../data/args_files/optimized_configs.json",
+    "constraints_path": null,
+    "crossval_index_dir": null,
+    "crossval_index_file": null,
+    "crossval_index_sets": null,
+    "cuda": true,
+    "data_path": "../data/crossval_splits/all_amine_split_for_paper/cv_4/train.csv",
+    "data_weights_path": "../data/crossval_splits/all_amine_split_for_paper/cv_4/train_weights.csv",
+    "dataset_type": "regression",
+    "depth": 4,
+    "depth_solvent": 3,
+    "device": {
+        "_string": "cuda",
+        "_type": "python_object (type = device)",
+        "_value": "gASVHwAAAAAAAACMBXRvcmNolIwGZGV2aWNllJOUjARjdWRhlIWUUpQu"
+    },
+    "dropout": 0.1,
+    "empty_cache": false,
+    "ensemble_size": 1,
+    "epochs": 50,
+    "evidential_regularization": 0,
+    "explicit_h": false,
+    "extra_metrics": [],
+    "features_generator": null,
+    "features_only": false,
+    "features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_4/train_extra_x.csv"
+    ],
+    "features_scaling": true,
+    "features_size": null,
+    "ffn_hidden_size": 600,
+    "ffn_num_layers": 3,
+    "final_lr": 0.0001,
+    "folds_file": null,
+    "freeze_first_only": false,
+    "frzn_ffn_layers": 0,
+    "gpu": null,
+    "grad_clip": null,
+    "hidden_size": 600,
+    "hidden_size_solvent": 300,
+    "ignore_columns": null,
+    "ignore_nan_metrics": false,
+    "init_lr": 0.0001,
+    "is_atom_bond_targets": false,
+    "keeping_atom_map": false,
+    "log_frequency": 10,
+    "loss_function": "mse",
+    "max_data_size": null,
+    "max_lr": 0.001,
+    "metric": "rmse",
+    "metrics": [
+        "rmse"
+    ],
+    "minimize_score": true,
+    "mpn_shared": false,
+    "multiclass_num_classes": 3,
+    "no_adding_bond_types": false,
+    "no_atom_descriptor_scaling": false,
+    "no_bond_descriptor_scaling": false,
+    "no_cache_mol": false,
+    "no_cuda": false,
+    "no_features_scaling": false,
+    "no_shared_atom_bond_ffn": false,
+    "num_folds": 1,
+    "num_lrs": 1,
+    "num_tasks": 1,
+    "num_workers": 8,
+    "number_of_molecules": 1,
+    "overwrite_default_atom_features": false,
+    "overwrite_default_bond_features": false,
+    "phase_features_path": null,
+    "pytorch_seed": 0,
+    "quantile_loss_alpha": 0.1,
+    "quantiles": [],
+    "quiet": false,
+    "reaction": false,
+    "reaction_mode": "reac_diff",
+    "reaction_solvent": false,
+    "reproducibility": {
+        "command_line": "python main_script.py train all_amine_split_for_paper",
+        "git_has_uncommitted_changes": true,
+        "git_root": "/media/andersonxps/wd_4tb/evan/LNP_ML",
+        "git_url": "https://github.com/jswitten/LNP_ML/tree/167822980dc26ba65c5c14539c4ce12b81b0b8f3",
+        "time": "Tue Jul 30 10:40:44 2024"
+    },
+    "resume_experiment": false,
+    "save_dir": "../data/crossval_splits/all_amine_split_for_paper/cv_4",
+    "save_preds": false,
+    "save_smiles_splits": false,
+    "seed": 42,
+    "separate_test_atom_descriptors_path": null,
+    "separate_test_bond_descriptors_path": null,
+    "separate_test_constraints_path": null,
+    "separate_test_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_4/test_extra_x.csv"
+    ],
+    "separate_test_path": "../data/crossval_splits/all_amine_split_for_paper/cv_4/test.csv",
+    "separate_test_phase_features_path": null,
+    "separate_val_atom_descriptors_path": null,
+    "separate_val_bond_descriptors_path": null,
+    "separate_val_constraints_path": null,
+    "separate_val_features_path": [
+        "../data/crossval_splits/all_amine_split_for_paper/cv_4/valid_extra_x.csv"
+    ],
+    "separate_val_path": "../data/crossval_splits/all_amine_split_for_paper/cv_4/valid.csv",
+    "separate_val_phase_features_path": null,
+    "shared_atom_bond_ffn": true,
+    "show_individual_scores": false,
+    "smiles_columns": [
+        "smiles"
+    ],
+    "spectra_activation": "exp",
+    "spectra_phase_mask_path": null,
+    "spectra_target_floor": 1e-08,
+    "split_key_molecule": 0,
+    "split_sizes": [
+        1.0,
+        0.0,
+        0.0
+    ],
+    "split_type": "random",
+    "target_columns": null,
+    "target_weights": null,
+    "task_names": [
+        "quantified_delivery"
+    ],
+    "test": false,
+    "test_fold_index": null,
+    "train_data_size": null,
+    "undirected": false,
+    "use_input_features": true,
+    "val_fold_index": null,
+    "warmup_epochs": 2.0,
+    "weights_ffn_num_layers": 2
+}

data/external/all_amine_split_for_LiON/cv_4/fold_0/test_scores.json

@@ -0,0 +1,5 @@
+{
+    "rmse": [
+        0.8268900471469541
+    ]
+}

data/external/all_amine_split_for_LiON/cv_4/test_scores.csv

@@ -0,0 +1,2 @@
+Task,Mean rmse,Standard deviation rmse,Fold 0 rmse
+quantified_delivery,0.8268900471469541,0.0,0.8268900471469541

data/external/all_amine_split_for_LiON/evaluation_results.json

@@ -0,0 +1,81 @@
+{
+  "data_dir": "/Users/ryde/Documents/workspaces/\u8102\u8d28\u5206\u5b50\u836f\u7269\u9012\u9001\u6027\u80fd\u9884\u6d4b/\u6700\u65b0\u6574\u7406/lnp-ml/data/external/all_amine_split_for_LiON",
+  "y_col": "quantified_delivery",
+  "n_splits": 5,
+  "split_results": {
+    "cv_0": {
+      "n_samples": 2037,
+      "mse": 0.7886545513325788,
+      "rmse": 0.8880622451903801,
+      "mae": 0.6877543784155847,
+      "r2": 0.0932581633950823,
+      "correlation": 0.392311374929795
+    },
+    "cv_1": {
+      "n_samples": 1658,
+      "mse": 1.033754621206102,
+      "rmse": 1.01673724295223,
+      "mae": 0.7387616255484185,
+      "r2": 0.1125028601022715,
+      "correlation": 0.36625872345742133
+    },
+    "cv_2": {
+      "n_samples": 1615,
+      "mse": 0.7723021982152163,
+      "rmse": 0.8788072588544181,
+      "mae": 0.6676348056050336,
+      "r2": 0.21571146545185693,
+      "correlation": 0.48232889534779555
+    },
+    "cv_3": {
+      "n_samples": 1754,
+      "mse": 0.8548731227367337,
+      "rmse": 0.9245934905333985,
+      "mae": 0.6985704943044474,
+      "r2": 0.1788528564997307,
+      "correlation": 0.43034534464045127
+    },
+    "cv_4": {
+      "n_samples": 1520,
+      "mse": 0.6837471500706921,
+      "rmse": 0.8268900471469541,
+      "mae": 0.6290387454956646,
+      "r2": 0.27613207572098897,
+      "correlation": 0.5356790397266105
+    }
+  },
+  "overall": {
+    "n_samples": 8584,
+    "mse": 0.8278734457746315,
+    "rmse": 0.9098755111412943,
+    "mae": 0.6856342259095253,
+    "r2": 0.1721265542253685,
+    "correlation": 0.43182567556105245
+  },
+  "summary_stats": {
+    "rmse": {
+      "mean": 0.9070180569354761,
+      "std": 0.0631234750727322,
+      "min": 0.8268900471469541,
+      "max": 1.01673724295223
+    },
+    "r2": {
+      "mean": 0.17529148423398608,
+      "std": 0.06706289614045773,
+      "min": 0.0932581633950823,
+      "max": 0.27613207572098897
+    },
+    "mae": {
+      "mean": 0.6843520098738297,
+      "std": 0.036092020790716446,
+      "min": 0.6290387454956646,
+      "max": 0.7387616255484185
+    },
+    "correlation": {
+      "mean": 0.4413846756204148,
+      "std": 0.061223885190061875,
+      "min": 0.36625872345742133,
+      "max": 0.5356790397266105
+    }
+  }
+}

scripts/evaluate_external.py

@@ -0,0 +1,204 @@
+"""评估外部数据 cross-validation 结果的脚本"""
+
+import json
+from pathlib import Path
+from typing import Dict
+
+import numpy as np
+import pandas as pd
+import typer
+from loguru import logger
+from sklearn.metrics import mean_squared_error, r2_score
+
+from lnp_ml.config import EXTERNAL_DATA_DIR
+
+
+app = typer.Typer()
+
+
+def evaluate_split(
+    test_path: Path,
+    preds_path: Path,
+    y_col: str = "quantified_delivery",
+) -> Dict[str, float]:
+    """
+    评估单个 split 的预测结果。
+    
+    Args:
+        test_path: groundtruth CSV 路径（包含 smiles 和 y_col）
+        preds_path: 预测结果 CSV 路径（包含 smiles 和 y_col）
+        y_col: Y 值列名
+        
+    Returns:
+        Dict with rmse, r2, mse, mae, n_samples
+    """
+    # 读取数据
+    test_df = pd.read_csv(test_path)
+    preds_df = pd.read_csv(preds_path)
+    
+    # 检查列是否存在
+    if y_col not in test_df.columns:
+        raise ValueError(f"Column '{y_col}' not found in {test_path}")
+    if y_col not in preds_df.columns:
+        raise ValueError(f"Column '{y_col}' not found in {preds_path}")
+    
+    # 直接按行对齐（test.csv 和 preds.csv 顺序一致）
+    assert len(test_df) == len(preds_df), f"行数不匹配: test={len(test_df)}, preds={len(preds_df)}"
+    
+    # 验证 smiles 列一致（如果存在）
+    if "smiles" in test_df.columns and "smiles" in preds_df.columns:
+        if not (test_df["smiles"].values == preds_df["smiles"].values).all():
+            logger.warning("smiles 列顺序不一致，请检查数据")
+    
+    y_true = test_df[y_col].values
+    y_pred = preds_df[y_col].values
+    
+    # 计算指标
+    mse = float(mean_squared_error(y_true, y_pred))
+    rmse = float(np.sqrt(mse))
+    r2 = float(r2_score(y_true, y_pred))
+    mae = float(np.mean(np.abs(y_true - y_pred)))
+    correlation = float(np.corrcoef(y_true, y_pred)[0, 1])
+    
+    return {
+        "n_samples": len(y_true),
+        "mse": mse,
+        "rmse": rmse,
+        "mae": mae,
+        "r2": r2,
+        "correlation": correlation,
+    }
+
+
+@app.command()
+def main(
+    data_dir: Path = EXTERNAL_DATA_DIR / "all_amine_split_for_LiON",
+    output_path: Path = EXTERNAL_DATA_DIR / "all_amine_split_for_LiON" / "evaluation_results.json",
+    y_col: str = "quantified_delivery",
+):
+    """
+    评估 all_amine_split_for_LiON 的 cross-validation 结果。
+    
+    计算每个 split 和整体的 RMSE、R² 指标。
+    """
+    logger.info(f"Evaluating cross-validation results from {data_dir}")
+    
+    # 找到所有 cv_* 目录
+    cv_dirs = sorted([d for d in data_dir.iterdir() if d.is_dir() and d.name.startswith("cv_")])
+    
+    if not cv_dirs:
+        logger.error(f"No cv_* directories found in {data_dir}")
+        raise typer.Exit(1)
+    
+    logger.info(f"Found {len(cv_dirs)} splits: {[d.name for d in cv_dirs]}")
+    
+    # 评估每个 split
+    split_results = {}
+    all_y_true = []
+    all_y_pred = []
+    
+    for cv_dir in cv_dirs:
+        split_name = cv_dir.name
+        test_path = cv_dir / "test.csv"
+        preds_path = cv_dir / "preds.csv"
+        
+        if not test_path.exists():
+            logger.warning(f"  {split_name}: test.csv not found, skipping")
+            continue
+        if not preds_path.exists():
+            logger.warning(f"  {split_name}: preds.csv not found, skipping")
+            continue
+        
+        # 评估
+        metrics = evaluate_split(test_path, preds_path, y_col)
+        split_results[split_name] = metrics
+        
+        logger.info(
+            f"  {split_name}: n={metrics['n_samples']}, "
+            f"RMSE={metrics['rmse']:.4f}, R²={metrics['r2']:.4f}"
+        )
+        
+        # 收集所有数据用于计算整体指标
+        test_df = pd.read_csv(test_path)
+        preds_df = pd.read_csv(preds_path)
+        all_y_true.extend(test_df[y_col].tolist())
+        all_y_pred.extend(preds_df[y_col].tolist())
+    
+    # 计算整体指标
+    all_y_true = np.array(all_y_true)
+    all_y_pred = np.array(all_y_pred)
+    
+    overall_mse = float(mean_squared_error(all_y_true, all_y_pred))
+    overall_rmse = float(np.sqrt(overall_mse))
+    overall_r2 = float(r2_score(all_y_true, all_y_pred))
+    overall_mae = float(np.mean(np.abs(all_y_true - all_y_pred)))
+    overall_correlation = float(np.corrcoef(all_y_true, all_y_pred)[0, 1])
+    
+    overall_results = {
+        "n_samples": len(all_y_true),
+        "mse": overall_mse,
+        "rmse": overall_rmse,
+        "mae": overall_mae,
+        "r2": overall_r2,
+        "correlation": overall_correlation,
+    }
+    
+    # 计算 split 指标的均值和标准差
+    split_metrics = list(split_results.values())
+    summary_stats = {}
+    for metric in ["rmse", "r2", "mae", "correlation"]:
+        values = [s[metric] for s in split_metrics]
+        summary_stats[metric] = {
+            "mean": float(np.mean(values)),
+            "std": float(np.std(values)),
+            "min": float(np.min(values)),
+            "max": float(np.max(values)),
+        }
+    
+    # 汇总结果
+    results = {
+        "data_dir": str(data_dir),
+        "y_col": y_col,
+        "n_splits": len(split_results),
+        "split_results": split_results,
+        "overall": overall_results,
+        "summary_stats": summary_stats,
+    }
+    
+    # 打印结果
+    logger.info("\n" + "=" * 60)
+    logger.info("CROSS-VALIDATION EVALUATION RESULTS")
+    logger.info("=" * 60)
+    
+    logger.info(f"\n[Per-Split Results]")
+    for split_name, metrics in sorted(split_results.items()):
+        logger.info(
+            f"  {split_name}: RMSE={metrics['rmse']:.4f}, "
+            f"R²={metrics['r2']:.4f}, "
+            f"MAE={metrics['mae']:.4f}, "
+            f"Corr={metrics['correlation']:.4f}"
+        )
+    
+    logger.info(f"\n[Summary Statistics (across {len(split_results)} splits)]")
+    for metric, stats in summary_stats.items():
+        logger.info(
+            f"  {metric.upper():12s}: "
+            f"mean={stats['mean']:.4f} ± {stats['std']:.4f} "
+            f"(min={stats['min']:.4f}, max={stats['max']:.4f})"
+        )
+    
+    logger.info(f"\n[Overall (all {overall_results['n_samples']} samples pooled)]")
+    logger.info(f"  RMSE:        {overall_results['rmse']:.4f}")
+    logger.info(f"  R²:          {overall_results['r2']:.4f}")
+    logger.info(f"  MAE:         {overall_results['mae']:.4f}")
+    logger.info(f"  Correlation: {overall_results['correlation']:.4f}")
+    
+    # 保存结果
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    with open(output_path, "w") as f:
+        json.dump(results, f, indent=2)
+    logger.success(f"\nSaved results to {output_path}")
+
+
+if __name__ == "__main__":
+    app()