bayes_opt pipeline has been implemented instead of wandb

bayes_opt_run.py

+from bayes_opt import BayesianOptimization
+from bayes_opt.logger import JSONLogger
+from bayes_opt.event import Events
+
+
+from pipeline_test_command_params_just_tls import RunCommand
+from pipeline_test_command_params_just_tls import main as pipeline_main
+
+def bayes_opt_main(
+    n_tiles,
+    slice_thickness,
+    find_stems_height,
+    find_stems_thickness,
+    graph_maximum_cumulative_gap,
+    add_leaves_voxel_length,
+    find_stems_min_points,
+    graph_edge_length,
+    add_leaves_edge_length
+):
+    n_tiles = int(n_tiles)
+    find_stems_min_points = int(find_stems_min_points)
+    return pipeline_main(
+        n_tiles,
+        slice_thickness,
+        find_stems_height,
+        find_stems_thickness,
+        graph_maximum_cumulative_gap,
+        add_leaves_voxel_length,
+        find_stems_min_points,
+        graph_edge_length,
+        add_leaves_edge_length
+    )
+
+pbounds = {
+    'n_tiles': (3, 3),
+    'slice_thickness': (0.25, 0.75),
+    'find_stems_height': (0.5, 2.0),
+    'find_stems_thickness': (0.1, 1.0),
+    'graph_maximum_cumulative_gap': (5, 20),
+    'add_leaves_voxel_length': (0.1, 0.5),
+    'find_stems_min_points': (50, 500),
+    'graph_edge_length': (0.1, 2.0),
+    'add_leaves_edge_length': (0.2, 1.5)
+}
+
+optimizer = BayesianOptimization(
+    f=bayes_opt_main,
+    pbounds=pbounds,
+    random_state=1,
+    allow_duplicate_points=True
+)
+
+logger = JSONLogger(path="./bayes_opt_run_logs.json")
+optimizer.subscribe(Events.OPTIMIZATION_STEP, logger)
+
+optimizer.maximize(
+    init_points=5,
+    n_iter=100
+)

helpers/find_param_importance.py

+# partially based on : https://medium.com/analytics-vidhya/feature-importance-explained-bfc8d874bcf
+import json
+import matplotlib.pyplot as plt
+import pandas as pd
+import argparse
+from sklearn.linear_model import LinearRegression
+from sklearn.preprocessing import StandardScaler
+
+class FindParamImportance:
+    def __init__(
+        self, 
+        logs_json_file, 
+        plot_file_path='feature_importance.png',
+         verbose=False
+         ) -> None:
+        self.logs_json_file = logs_json_file
+        self.plot_file_path = plot_file_path
+        self.verbose = verbose
+
+    def get_data(self):
+        runs = []
+        for line in open('logs.json', 'r'):
+            runs.append(json.loads(line))
+
+        # get header of the logs
+        header = ['target']
+
+        for key in runs[0]['params']:
+            header.append(key)
+
+        # create a dictionary from the header with empty lists
+        data = {}
+        data = data.fromkeys(header)
+        data = {key: [] for key in header}
+
+        # fill the dictionary with the data
+        for run in runs:
+            data['target'].append(run['target'])
+            for key in run['params']:
+                data[key].append(run['params'][key])
+
+        # create a dataframe from the dictionary
+        df = pd.DataFrame(data)
+
+        return df
+
+    def get_feature_importance(self, df):
+        # get the features
+        X = df.drop('target', axis=1)
+        # get the target
+        y = df['target']
+
+        ss = StandardScaler()
+        X_scaled = ss.fit_transform(X)
+
+        model=LinearRegression()
+        model.fit(X_scaled,y)
+        importance=model.coef_
+
+        # combine importance with the feature names
+        feature_importance = pd.DataFrame({'feature': X.columns, 'importance': abs(importance)})        
+        # sort the values
+        feature_importance.sort_values(by='importance', ascending=True, inplace=True)
+
+        if self.verbose:
+            print(feature_importance)
+
+        return feature_importance
+
+    def gen_plot_of_feature_importance(self, feature_importance):
+        plt.figure(figsize=(10, 6))
+        plt.barh(feature_importance['feature'], feature_importance['importance'])
+        plt.title('Feature Importance')
+
+        # save the plot
+        plt.savefig(self.plot_file_path)
+
+
+    def main(self):
+        df = self.get_data()
+        feature_importance = self.get_feature_importance(df)
+        self.gen_plot_of_feature_importance(feature_importance)
+        if self.verbose:
+            print('Done')
+            print('Plot saved to: ', self.plot_file_path)
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--logs_json_file', type=str, default='logs.json')
+    parser.add_argument('--plot_file_path', type=str, default='feature_importance.png')
+    parser.add_argument('--verbose', help="Print more information.", action="store_true")
+
+
+    args = parser.parse_args()
+    logs_json_file = args.logs_json_file
+    plot_file_path = args.plot_file_path
+    verbose = args.verbose
+
+    find_param_importance = FindParamImportance(
+        logs_json_file=logs_json_file,
+        plot_file_path=plot_file_path,
+        verbose=verbose
+    )
+
+    find_param_importance.main()
\ No newline at end of file

pipeline_test_command_params_just_tls.py

@@ -5,9 +5,6 @@ import wandb
 from metrics.instance_segmentation_metrics_in_folder import \
     InstanceSegmentationMetricsInFolder
 
-# wandb.login()
-
-# wandb.init(project="instance_segmentation_classic", entity="smart_forest")
 
 # define a class to run the command with arguments
 class RunCommand:
@@ -30,9 +27,12 @@ def main(
     graph_edge_length, 
     add_leaves_edge_length
     ):
+    
+    USE_WANDB = False
 
     # initialize the sweep
-    run = wandb.init(project="paper-sweep-nibio-model-just-tls", entity="smart_forest")
+    if USE_WANDB:
+        run = wandb.init(project="paper-sweep-nibio-model-just-tls", entity="smart_forest")
 
     # get files for the sweep
     print("Getting files for the sweep")
@@ -92,7 +92,10 @@ def main(
 
     # log the metric
     print("Logging the metric")
-    wandb.log({"f1_score": f1_score})
+    if USE_WANDB:
+        wandb.log({"f1_score": f1_score})
+
+    return f1_score
 
 if __name__ == "__main__":
     # use argparse to get the arguments