77from sklearn .ensemble import RandomForestClassifier
88from sklearn .utils import class_weight
99from sklearn .metrics import classification_report
10+ from sklearn .metrics import confusion_matrix , accuracy_score
11+ from io import StringIO
1012
1113@st .cache_data
12- def run_random_forest (attribute , n_trees ):
14+ def run_random_forest (attribute , n_trees , random_seed = None ):
1315 # initialize a log to print out in the app later
1416 log = ""
1517
18+ df_oob = pd .DataFrame () # Placeholder
19+ df_important_features = pd .DataFrame () # Placeholder
20+
21+ # Placeholder for classification report and label mapping
22+ class_report = "Classification report here"
23+ label_mapping = "Label mapping here"
24+
1625 labels = st .session_state .md [[attribute ]]
1726 rf_data = pd .concat ([st .session_state .data , labels ], axis = 1 )
1827
@@ -22,63 +31,125 @@ def run_random_forest(attribute, n_trees):
2231 labels = enc .fit_transform (labels )
2332 labels = np .array ([x [0 ] + 1 for x in labels ])
2433
34+ class_names = enc .categories_ [0 ] #getting the class names
35+
2536 # Extract the feature intensities as np 2D array
2637 features = np .array (st .session_state .data )
2738
2839
2940 # Split the data into training and test sets
30- train_features , test_features , train_labels , test_labels = train_test_split (features , labels , test_size = 0.25 , random_state = 123 )
31-
32- print (f'Training Features Shape: { train_features .shape } )
33- print (f'Training Labels Shape: { train_labels .shape } )
34- print (f'Testing Features Shape: { test_features .shape } )
35- print (f'Testing Labels Shape: { test_labels .shape } )
41+ train_features , test_features , train_labels , test_labels = train_test_split (features ,
42+ labels ,
43+ test_size = 0.25 ,
44+ random_state = random_seed ,
45+ stratify = labels )
46+
47+ # Collecting info about feature and label shapes for logging
48+ log += f"Training Features Shape: { train_features .shape } \n "
49+ log += f"Training Labels Shape: { train_labels .shape } \n "
50+ log += f"Testing Features Shape: { test_features .shape } \n "
51+ log += f"Testing Labels Shape: { test_labels .shape } \n "
3652
3753 # Balance the weights of the attribute of interest to account for unbalanced sample sizes per group
3854 sklearn_weights = class_weight .compute_class_weight (
3955 class_weight = 'balanced' ,
4056 classes = np .unique (train_labels ),
4157 y = train_labels )
58+
4259 weights = {}
4360 for i ,w in enumerate (np .unique (train_labels )):
4461 weights [w ] = sklearn_weights [i ]
4562
4663 # Set up the random forest classifier with 100 tress, balanded weights, and a random state to make it reproducible
47- rf = RandomForestClassifier (n_estimators = n_trees , class_weight = weights , random_state = 123 )
64+ rf = RandomForestClassifier (n_estimators = n_trees , class_weight = 'balanced' , random_state = random_seed )
65+
4866 # Fit the classifier to the training set
4967 rf .fit (train_features , train_labels )
5068
5169 # Use the random forest classifier to predict the sample areas in the test set
52- predictions = rf .predict (test_features )
53- print (f'Classifier mean accuracy score: { round (rf .score (test_features , test_labels )* 100 , 2 )} )
70+ predictions_test = rf .predict (test_features )
71+ predictions_train = rf .predict (train_features )
72+
73+ classifier_accuracy = round (rf .score (test_features , test_labels )* 100 , 2 )
74+ log += f"Classifier mean accuracy score: { classifier_accuracy } \n "
5475
76+ # Calculate confusion matrices
77+ test_confusion_matrix = confusion_matrix (test_labels , predictions_test , labels = range (len (class_names )))
78+ train_confusion_matrix = confusion_matrix (train_labels , predictions_train , labels = range (len (class_names )))
79+
80+ test_confusion_df = pd .DataFrame (test_confusion_matrix , index = class_names , columns = class_names )
81+ train_confusion_df = pd .DataFrame (train_confusion_matrix , index = class_names , columns = class_names )
82+
83+ test_accuracy = accuracy_score (test_labels , predictions_test )
84+ train_accuracy = accuracy_score (train_labels , predictions_train )
85+
5586 # Report of the accuracy of predictions on the test set
56- print ( classification_report (test_labels , predictions ) )
87+ class_report = classification_report (test_labels , predictions_test )
5788
5889 # Print the sample areas corresponding to the numbers in the report
59- print ("Sample areas corresponding to the numbers:" )
60- for i ,cat in enumerate (enc .categories_ [0 ]):
61- print (f"{ i + 1.0 } { cat } )
90+ label_mapping = "\n " .join ([f"{ i + 1.0 } { cat } for i , cat in enumerate (enc .categories_ [0 ])])
6291
6392 # Most important model quality plot
6493 # OOB error lines should flatline. If it doesn't flatline add more trees
65- rf = RandomForestClassifier (class_weight = weights , warm_start = True , oob_score = True , random_state = 123 )
94+ rf_oob = RandomForestClassifier (class_weight = weights , warm_start = True , oob_score = True , random_state = 123 )
6695 errors = []
6796 tree_range = np .arange (1 ,500 , 10 )
6897 for i in tree_range :
69- rf .set_params (n_estimators = i )
70- rf .fit (train_features , train_labels )
71- errors .append (1 - rf .oob_score_ )
98+ rf_oob .set_params (n_estimators = i )
99+ rf_oob .fit (train_features , train_labels )
100+ errors .append (1 - rf_oob .oob_score_ )
72101
73102
74103 df_oob = pd .DataFrame ({"n trees" : tree_range , "error rate" : errors })
75104
76105 # Extract the important features in the model
77- df_important_features = pd .DataFrame (rf .feature_importances_ , index = st .session_state .data .columns ).sort_values (by = 0 , ascending = False )
106+ df_important_features = pd .DataFrame (rf .feature_importances_ ,
107+ index = st .session_state .data .columns ).sort_values (by = 0 , ascending = False )
78108 df_important_features .columns = ["importance" ]
79109
80-
81- return df_oob , df_important_features
110+ return df_oob , df_important_features , log , class_report , label_mapping , test_confusion_df , train_confusion_df , test_accuracy , train_accuracy
82111
83112def get_oob_fig (df ):
84- return px .line (df , x = "n trees" , y = "error rate" , title = "out-of-bag (OOB) error" )
113+ return px .line (df , x = "n trees" , y = "error rate" , title = "out-of-bag (OOB) error" )
114+
115+ def classification_report_to_df (report ):
116+
117+ # Split the report into lines
118+ lines = report .split ("\n " )
119+
120+ # Prepare a dictionary to hold the data
121+ report_data = {"class" : [], "precision" : [], "recall" : [], "f1-score" : [], "support" : []}
122+
123+ for line in lines [2 :- 3 ]: # Skip the header and summary lines
124+ parts = line .split ()
125+ # Ensure that the line contains the expected number of parts
126+ if len (parts ) == 5 :
127+ report_data ["class" ].append (parts [0 ])
128+ report_data ["precision" ].append (parts [1 ])
129+ report_data ["recall" ].append (parts [2 ])
130+ report_data ["f1-score" ].append (parts [3 ])
131+ report_data ["support" ].append (parts [4 ])
132+
133+ # Convert the dictionary to a DataFrame
134+ report_df = pd .DataFrame (report_data )
135+
136+ # Convert numeric columns from strings to floats
137+ report_df [["precision" , "recall" , "f1-score" ]] = report_df [["precision" , "recall" , "f1-score" ]].astype (float )
138+ report_df ["support" ] = report_df ["support" ].astype (int )
139+
140+ return report_df
141+
142+ def label_mapping_to_df (label_mapping_str ):
143+
144+ # Split the string into lines
145+ lines = label_mapping_str .split ("\n " )
146+
147+ # Split each line into index and label, then collect into a list of tuples
148+ mapping = [line .split (" ," ) for line in lines if line ] # Ensure the line is not empty
149+
150+ # Convert the list of tuples into a DataFrame
151+ mapping_df = pd .DataFrame (mapping , columns = ['class' , 'Label' ])
152+ mapping_df ['class' ] = mapping_df ['class' ].astype (str )
153+ return mapping_df
154+
155+
0 commit comments