Feature to label

code/kappa_data/feature_to_label.py

@@ -0,0 +1,70 @@
+import pandas as pd
+import numpy as np 
+import os
+
+def add_label(df):
+    # add a column to save the results of the decision rules
+    df['label'] = np.nan
+    # test wheather the new pixel-to-mm-ratio is accurate
+    # the old ratio is 4.2 and the new one 2.05
+    # hence mutliplying the value/4.2 is the same as value/2.05
+    df['width'] = (df['auto_width']*4.2)*0.205
+    # TODO: evaluate threshold for lenght and width
+    t_vthin = 16 # =Suppe
+    t_thin = 20
+    t_medium = 22
+    t_thick = 24
+    t_vthick = 26 # =Dicke
+
+    t_bruch = 210
+
+
+    for i in range(df.shape[0]):
+        # First ask the 4 'short' questions which lead to easy decisions
+        # 1. check for Bruch
+        if (df['auto_length'][i] < t_bruch or df['is_bruch'][i] == 1):
+            df['label'][i] = 'Bruch'
+        # 2. check for Suppe
+        elif (df['width'][i] < t_vthin or df['very_thin'][i]==1):
+            df['label'][i] = 'Suppe'
+        # 3. check for Dicke
+        elif (df['width'][i] > t_vthick or df['very_thick'][i]==1):
+            df['label'][i] = 'Dicke'
+        # 4. check for Hohle
+        elif (df['is_hollow'][i] == 1):
+            df['label'][i] = 'Hohle'
+        # check for 1A quality and afterwards decide between Anna, Bona, Clara
+        elif (df['has_blume'][i]==0 and df['has_rost_head'][i]==0 and df['has_rost_body'][i]==0 and df['is_violet'][i]==0 and df['is_bended'][i]==0):
+            if df['width'][i] < t_thin or df['thin'][i] == 1:
+                df['label'][i] = '1A_Clara'
+            elif df['width'][i] < t_medium or df['medium_thick'][i] == 1:
+                df['label'][i] = '1A_Bona'
+            else:
+                df['label'][i] = '1A_Anna'
+        elif (df['has_blume'][i]==0 and df['has_rost_head'][i]==0 and df['has_rost_body'][i]==0 and df['is_violet'][i]==0):
+            df['label'][i] = '1A_Krumme'
+        elif (df['has_blume'][i]==0 and df['has_rost_head'][i]==0 and df['has_rost_body'][i]==0 and df['is_bended'][i]==0):
+            df['label'][i] = '1A_Violett'
+        elif (df['has_rost_head'][i]==1 or df['has_rost_body'][i]==1):
+            df['label'][i] = 'Rost'
+        elif (df['has_blume'][i]==1):
+            df['label'][i] = 'Blume'
+        elif (df['width'][i] > t_thick or df['medium_thick'][i] == 1 or df['thick'][i] == 1 ):
+            df['label'][i] = '2A'
+        else:
+            df['label'][i] = '2B'
+    return df
+
+
+if __name__ == "__main__":
+    path = '.' # path to label files for which a class should be estimated
+    print("Path is current folder")
+    # get all csv files in path
+    csvs = [x for x in os.listdir(path) if x.endswith('.csv')]
+    # open files and add the corresponding labels to each image id
+    for csv in csvs:
+        full_path = path + csv
+        df = pd.read_csv(full_path, delimiter=';')
+        new_df = add_label(df)
+        new_file_name = full_path + 'labeltestnew.csv'
+        export_csv = new_df.to_csv(new_file_name)