Sort classification_report of sklearn.metrics

Since sklearn.metrics.classification_report returns a string, we can not sort the data as is. The solution, if we want to use this function, is to rewrite it or to parse the output chain and reorder the data to our liking.

Since it is not usually necessary to be very efficient when using this type of function, aimed only at presenting data to the user in a pleasant way, the simplest is the second option.

Since the structure of the output chain is constant and only varies in the number of spaces and decimals depending on the input parameters, it is not complicated to do this using str.split() . First we obtain the lines using \n as a separator and then we will obtain the data of the column according to the one we want to order. Based on this column, we reorder the lines and reconstruct the chain with str.join() .

One function to do this could be:

from sklearn.metrics import classification_report  #No olvidar este import



def sorted_classification_report(y_true, y_pred, labels=None, target_names=None, sample_weight=None, digits=2, sorted_by = None , reverse = False):
    lines = classification_report(y_true, y_pred, labels, target_names, sample_weight, digits).split('\n')
    datos = lines[2:-3]    

    if sorted_by:
        try:
            col = ('precision', 'recall', 'f1-score', 'support').index(sorted_by)-4
            keydict = dict(zip(datos, (float(row.split()[col]) for row in datos)))
            datos.sort(key = lambda x:  keydict.get(x), reverse=reverse)
        except ValueError:
            raise ValueError("invalid value for 'sorted_by' argument: '{}'.".format(sorted_by))

    return '\n'.join(lines[:2] + datos + lines[-3:])

The function receives the same parameters as classification_report plus two others:

• sorted_by: This parameter must receive the name of the column depending on the one you want to order, that is, one of the following options:

  -  None       # La salida queda ordenada según la etiqueta de la fila.
  - 'precision' # Ordena las filas de acuerdo a la columna "precision".
  - 'recall'    # Ordena las filas de acuerdo a la columna "recall".
  - 'f1-score'  # Ordena las filas de acuerdo a la columna "f1-score".
  - 'support'   # Ordena las filas de acuerdo a la columna "support".
  

• reversed :

  - False       # Ordena de menor a mayor.
  - True        # Ordena de mayor a menos.
  

Example of use sorting by f1-score from highest to lowest:

y_true = [0, 1, 2, 2, 2]
y_pred = [0, 0, 2, 2, 1]
target_names = ['class 0', 'class 1', 'class 2']

print(sorted_classification_report(y_true, y_pred, target_names=target_names, digits = 2, sorted_by= 'f1-score' , reverse = True))

Exit:

             precision    recall  f1-score   support

    class 2       1.00      0.67      0.80         3
    class 0       0.50      1.00      0.67         1
    class 1       0.00      0.00      0.00         1

avg / total       0.70      0.60      0.61         5

Edit (see comments):

Applied the function on the example of the documentation on which you are working, substituting classification_report for our function:

print(sorted_classification_report(y_test, y_pred, target_names=target_names, sorted_by= 'support' , reverse = True))

We get the output (sorting from highest to lowest according to "support"):

                   precision    recall  f1-score   support

    George W Bush       0.84      0.98      0.90       146
     Colin Powell       0.78      0.87      0.82        60
       Tony Blair       0.96      0.75      0.84        36
  Donald Rumsfeld       0.81      0.63      0.71        27
Gerhard Schroeder       0.95      0.76      0.84        25
      Hugo Chavez       1.00      0.47      0.64        15
     Ariel Sharon       0.75      0.46      0.57        13

      avg / total       0.85      0.84      0.83       322