Use Flaml zero-shot automl

to mine good hyperparameter configurations offline
ersilia-os · May 9, 2024 · f5d5ad8 · f5d5ad8
1 parent d2ff580
commit f5d5ad8
Show file tree

Hide file tree

Showing 9 changed files with 18 additions and 7 deletions.
diff --git a/assets/evaluation_data.joblib b/assets/evaluation_data.joblib
diff --git a/assets/evaluation_metrics.json b/assets/evaluation_metrics.json
@@ -1,7  1,7 @@
 {
-    "Mean Squared Error": 0.3952,
-    "Root Mean Squared Error": 0.6287,
-    "Mean Absolute Error": 0.4569,
-    "R-squared Score": 0.7018,
-    "Explained Variance Score": 0.7018
     "Mean Squared Error": 0.3913,
     "Root Mean Squared Error": 0.6256,
     "Mean Absolute Error": 0.4472,
     "R-squared Score": 0.7047,
     "Explained Variance Score": 0.7053
 }
diff --git a/assets/evaluation_scatter_plot.png b/assets/evaluation_scatter_plot.png
diff --git a/assets/interpretability_bar_plot.png b/assets/interpretability_bar_plot.png
diff --git a/assets/interpretability_beeswarm_plot.png b/assets/interpretability_beeswarm_plot.png
diff --git a/assets/interpretability_sample1.png b/assets/interpretability_sample1.png
diff --git a/requirements.txt b/requirements.txt
@@ -9,6  9,8 @@ mordred==1.2.0
 pandas<=2.1
 matplotlib==3.8.3
 featurewiz==0.5.7
 tensorflow==2.16.1
 flaml==2.1.2
 imblearn
 pydantic
 aiohttp 

diff --git a/scripts/testing.py b/scripts/testing.py
@@ -41,7  41,7 @@
     smiles_valid_transformed = descriptor.transform(smiles_valid)
 
     # Instantiate the regressor
-    regressor = Regressor(output_folder, algorithm='catboost')
     regressor = Regressor(output_folder, algorithm='xgboost')
 
     # Train the model 
     regressor.fit(smiles_train_transformed, y_train) 

diff --git a/xai4chem/regressor.py b/xai4chem/regressor.py
@@ -5,12  5,15 @@
 import matplotlib.pyplot as plt
 import json
 import optuna
-from xgboost import XGBRegressor
 # from xgboost import XGBRegressor
 from catboost import CatBoostRegressor 
 from sklearn import metrics
 from sklearn.model_selection import train_test_split
 from sklearn.feature_selection import SelectKBest, mutual_info_regression
 from featurewiz import FeatureWiz
 from sklearn.svm import SVR
 from flaml.default import LGBMRegressor, XGBRegressor
 from flaml.default import preprocess_and_suggest_hyperparams
 
 
 class Regressor:
@@ -124,6  127,12 @@ def fit(self, X_train, y_train, default_params=True):
             else:
                 self.model = CatBoostRegressor()
             self.model.fit(X_train, y_train, verbose=False)
         elif self.algorithm == 'svr': 
             self.model = SVR()
             self.model.fit(X_train, y_train)
         elif self.algorithm == 'lgbm': 
             self.model = LGBMRegressor()
             self.model.fit(X_train, y_train)            
         else:
             raise ValueError("Invalid Algorithm. Supported Algorithms: xgboost, catboost")