Confusion Matrix with Recall, Precision and F1-Score for Caffe

confusionMatrix_Recall_Precision_F1Scroe_Caffe.py

#!/usr/bin/python

# Author: SeyyedHossein Hasanpour copyright 2017, license GPLv3.
# Seyyed Hossein Hasan Pour:
# [email protected]
# Changelog:
# 2015:
# initial code to calculate confusionmatrix by Axel Angel
# 7/3/2016:(adding new features-by-hossein) 
# added mean subtraction so that, the accuracy can be reported accurately just like caffe when doing a mean subtraction
# 01/03/2017:
# removed old codes and Added Recall/Precision/F1-Score as well 
# 03/05/2017
# Added ConfusionMatrix which was mistakenly ommited before.
#info: 
#if on windows, one can use these command in a batch file and ease him/her self 
#REM Calculating Confusing Matrix 
#python confusionMatrix_convnet_test.py --proto cifar10_deploy_94_68.prototxt --model cifar10_deploy_94_68.caffemodel --mean mean.binaryproto --db_type lmdb --db_path cifar10_test_lmdb  
#


import sys
import caffe
import numpy as np
import lmdb
import argparse
from collections import defaultdict
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
import itertools

def flat_shape(x):
    "Returns x without singleton dimension, eg: (1,28,28) -> (28,28)"
    return x.reshape(filter(lambda s: s > 1, x.shape))

def db_reader(fpath, type='lmdb'):
    if type == 'lmdb':
	    return lmdb_reader(fpath)
    else:
	   return leveldb_reader(fpath)


def plot_confusion_matrix(cm #confusion matrix
                         ,classes 
                          ,normalize=False
                          ,title='Confusion matrix'
                          ,cmap=plt.cm.Blues):
    """
    This function prints and plots the confusion matrix.
    Normalization can be applied by setting `normalize=True`.
    """
    plt.imshow(cm, interpolation='nearest', cmap=cmap)
    plt.title(title)
    plt.colorbar()
    tick_marks = np.arange(len(classes))
    plt.xticks(tick_marks, classes, rotation=45)
    plt.yticks(tick_marks, classes)

    if normalize:
        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
        print("confusion matrix is normalized!")
    
    #print(cm)

    thresh = cm.max() / 2.
    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
        plt.text(j, i, cm[i, j],
                 horizontalalignment="center",
                 color="white" if cm[i, j] > thresh else "black")

    plt.tight_layout()
    plt.ylabel('True label')
    plt.xlabel('Predicted label')



	   
def lmdb_reader(fpath):
    import lmdb
    lmdb_env = lmdb.open(fpath)
    lmdb_txn = lmdb_env.begin()
    lmdb_cursor = lmdb_txn.cursor()

    for key, value in lmdb_cursor:
        datum = caffe.proto.caffe_pb2.Datum()
        datum.ParseFromString(value)
        label = int(datum.label)
        image = caffe.io.datum_to_array(datum).astype(np.uint8)
        yield (key, flat_shape(image), label)

def leveldb_reader(fpath):
    import leveldb
    db = leveldb.LevelDB(fpath)

    for key, value in db.RangeIter():
        datum = caffe.proto.caffe_pb2.Datum()
        datum.ParseFromString(value)
        label = int(datum.label)
        image = caffe.io.datum_to_array(datum).astype(np.uint8)
        yield (key, flat_shape(image), label)


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument('--proto', help='path to the network prototxt file(deploy)', type=str, required=True)
    parser.add_argument('--model', help='path to your caffemodel file', type=str, required=True)
    parser.add_argument('--mean', help='path to the mean file(.binaryproto)', type=str, required=True)
    #group = parser.add_mutually_exclusive_group(required=True)
    parser.add_argument('--db_type', help='lmdb or leveldb', type=str, required=True)
    parser.add_argument('--db_path', help='path to your lmdb/leveldb dataset', type=str, required=True)
    args = parser.parse_args()

   # Extract mean from the mean image file
    mean_blobproto_new = caffe.proto.caffe_pb2.BlobProto()
    f = open(args.mean, 'rb')
    mean_blobproto_new.ParseFromString(f.read())
    mean_image = caffe.io.blobproto_to_array(mean_blobproto_new)
    f.close()
	
   # CNN reconstruction and loading the trained weights	
    net = caffe.Net(args.proto, args.model, caffe.TEST)
   # You may also use set_mode_cpu() if you didnt compile caffe with gpu support
    caffe.set_mode_gpu() 
	
    print ("args", vars(args))
	
    reader = db_reader(args.db_path, args.db_type.lower())
    
    predicted_lables=[]
    true_labels = []
    class_names = ['airplane','automobile','bird','cat','deer','dog','frog','horse','ship','truck']
    for i, image, label in reader:
        image_caffe = image.reshape(1, *image.shape)
        #print 'image shape: ',image_caffe.shape
        out = net.forward_all(data=np.asarray([ image_caffe ])- mean_image)
        plabel = int(out['prob'][0].argmax(axis=0))
        predicted_lables.append(plabel)	
        true_labels.append(label)
        print(i,' processed!')
	
    print( classification_report(y_true=true_labels,
	                             y_pred=predicted_lables,
	                             target_names=class_names))
    cm = confusion_matrix(y_true=true_labels,
	                        y_pred=predicted_lables)	
    print(cm)							
    # Compute confusion matrix
    cnf_matrix = cm
    np.set_printoptions(precision=2)

	# Plot non-normalized confusion matrix
    plt.figure()
    plot_confusion_matrix(cnf_matrix, classes=class_names,
						  title='Confusion matrix, without normalization')

	# Plot normalized confusion matrix
    plt.figure()
    plot_confusion_matrix(cnf_matrix, classes=class_names, normalize=True,
						  title='Normalized confusion matrix')

    plt.show()
							

Hi,
Nice to find this script but I have a question. I am still wondering whether there are any other reasons that cause the difference in accuracies because I found that the accuracies (in caffe and in convnet_test ) differ after a model trained without mean_file was tested .
Have you thoroughly solved this problem currently ?
Thanks ahead!

@lyuli: the difference in accuracy when mean file is used and when it is not used is pretty obvious.
by the way the script is updated and now recall, precision and F1score are all shown as well. much better now!)

It seems that your command instructions are not correct.

python confusionMatrix_convnet_test.py --proto cifar10_deploy.prototxt --model cifar10_SimpleNet_xavier_95.26.caffemodel.h5 --mean mean.binaryproto --lmdb cifar10_test_lmdb

should be:

python confusionMatrix_convnet_test.py --proto cifar10_deploy.prototxt --model cifar10_SimpleNet_xavier_95.26.caffemodel.h5 --mean mean.binaryproto --db_type lmdb --db_path cifar10_test_lmdb

Also I'm getting a KeyError for 'prob' in out:

Traceback (most recent call last): File "confusionMatrix_Recall_Precision_F1Scroe_Caffe.py", line 137, in <module> plabel = int(out['prob'][0].argmax(axis=0)) KeyError: 'prob'

Maybe it has to be 'loss' and argmax()?

EDIT: Yes that seems to work fine
EDIT2: Ok got it. It depends on what you use in top:
layer { name: "loss" type: "Softmax" bottom: "ip2" top: "loss" }

@flrndttrch: thanks for the note, when I updated the script I forgot to edit that part .
However I see I added the needed information as a comment to the script.
anyway I update the first post now.