Python 如何使用Tensorflow创建预测和基本真相标签的混淆矩阵?
在TensorFlow的帮助下,我实现了一个用于分类的通用网络模型。但是,我不知道如何才能通过预测分数(准确度)绘制混淆矩阵。我不是TensorFlow的专家,还在学习阶段。在这里,我将我的代码粘贴在下面。请告诉我如何编写代码,以避免与以下代码混淆:Python 如何使用Tensorflow创建预测和基本真相标签的混淆矩阵?,python,tensorflow,confusion-matrix,Python,Tensorflow,Confusion Matrix,在TensorFlow的帮助下,我实现了一个用于分类的通用网络模型。但是,我不知道如何才能通过预测分数(准确度)绘制混淆矩阵。我不是TensorFlow的专家,还在学习阶段。在这里,我将我的代码粘贴在下面。请告诉我如何编写代码,以避免与以下代码混淆: # Launch the graph with tf.Session() as sess: sess.run(init) # Set logs writer into folder /tmp/tensorflow_logs #summary_wr
# Launch the graph
with tf.Session() as sess:
sess.run(init)
# Set logs writer into folder /tmp/tensorflow_logs
#summary_writer = tf.train.SummaryWriter('/tmp/tensorflow_logs', graph_def=sess.graph_def)
# Training cycle
for epoch in range(training_epochs):
avg_cost = 0.
total_batch = int(X_train.shape[0]/batch_size)
# Loop over total length of batches
for i in range(total_batch):
#picking up random batches from training set of specific size
batch_xs, batch_ys = w2v_utils.nextBatch(X_train, y_train, batch_size)
# Fit training using batch data
sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys})
# Compute average loss
avg_cost += sess.run(cost, feed_dict={x: batch_xs, y: batch_ys})/total_batch
# Write logs at every iteration
#summary_str = sess.run(merged_summary_op, feed_dict={x: batch_xs, y: batch_ys})
#summary_writer.add_summary(summary_str, epoch*total_batch + i)
#append loss
loss_history.append(avg_cost)
# Display logs per epoch step
if (epoch % display_step == 0):
correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
# Calculate training accuracy
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
trainAccuracy = accuracy.eval({x: X_train, y: y_train})
train_acc_history.append(trainAccuracy)
# Calculate validation accuracy
valAccuracy = accuracy.eval({x: X_val, y: y_val})
val_acc_history.append(valAccuracy)
print "Epoch:", '%04d' % (epoch+1), "cost=", "{:.9f}".format(avg_cost), "train=",trainAccuracy,"val=", valAccuracy
print "Optimization Finished!"
# Test model
correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
# Calculate accuracy
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print "Final Training Accuracy:", accuracy.eval({x: X_train, y: y_train})
print "Final Test Accuracy:", accuracy.eval({x: X_test, y: y_test})
print "Final Gold Accuracy:", accuracy.eval({x: X_gold, y: y_gold})
到目前为止,我能够打印预测分数,但未能实现混淆矩阵,请帮助。
注意:(我正在使用一个热向量来表示我的标签)这段代码适合我。我自己解决:) 下面是示例输出:
Accuracy of Gold Test set Results: 0.642608
Predicted False True All
Actual
0 20 46 66
1 3 1 4
2 21 1 22
3 8 4 12
4 16 7 23
5 54 259 313
6 41 14 55
7 11 2 13
8 48 94 142
9 29 4 33
10 17 4 21
11 39 116 155
All 307 552 859
Predicted 0 1 2 3 4 5 6 7 8 9 10 11 All
Actual
0 46 0 0 0 0 8 0 2 2 2 0 6 66
1 0 1 0 1 0 2 0 0 0 0 0 0 4
2 3 0 1 3 0 12 0 0 1 0 0 2 22
3 2 0 0 4 1 3 1 1 0 0 0 0 12
4 1 0 0 0 7 12 0 0 1 0 0 2 23
5 8 0 0 1 5 259 9 0 9 3 1 18 313
6 1 0 0 1 6 30 14 1 2 0 0 0 55
7 3 0 0 0 0 2 0 2 4 0 1 1 13
8 6 0 0 1 1 18 0 3 94 8 1 10 142
9 9 0 0 0 0 1 1 1 9 4 0 8 33
10 1 0 0 0 3 6 0 1 1 0 4 5 21
11 5 1 0 1 0 18 1 0 6 5 2 116 155
All 85 2 1 12 23 371 26 11 129 22 9 168 859
precision recall f1-score support
0 0.54 0.70 0.61 66
1 0.50 0.25 0.33 4
2 1.00 0.05 0.09 22
3 0.33 0.33 0.33 12
4 0.30 0.30 0.30 23
5 0.70 0.83 0.76 313
6 0.54 0.25 0.35 55
7 0.18 0.15 0.17 13
8 0.73 0.66 0.69 142
9 0.18 0.12 0.15 33
10 0.44 0.19 0.27 21
11 0.69 0.75 0.72 155
avg / total 0.64 0.64 0.62 859
如果你想产生一个混淆矩阵,然后是精确性和召回率,你首先需要得到你的真阳性、真阴性、假阳性和假阴性的计数。以下是如何: 为了更好的可读性,我编写了非常详细的代码
def evaluation(logits,labels):
"Returns correct predictions, and 4 values needed for precision, recall and F1 score"
# Step 1:
# Let's create 2 vectors that will contain boolean values, and will describe our labels
is_label_one = tf.cast(labels, dtype=tf.bool)
is_label_zero = tf.logical_not(is_label_one)
# Imagine that labels = [0,1]
# Then
# is_label_one = [False,True]
# is_label_zero = [True,False]
# Step 2:
# get the prediction and false prediction vectors. correct_prediction is something that you choose within your model.
correct_prediction = tf.nn.in_top_k(logits, labels, 1, name="correct_answers")
false_prediction = tf.logical_not(correct_prediction)
# Step 3:
# get the 4 metrics by comparing boolean vectors
# TRUE POSITIVES
true_positives = tf.reduce_sum(tf.to_int32(tf.logical_and(correct_prediction,is_label_one)))
# FALSE POSITIVES
false_positives = tf.reduce_sum(tf.to_int32(tf.logical_and(false_prediction, is_label_zero)))
# TRUE NEGATIVES
true_negatives = tf.reduce_sum(tf.to_int32(tf.logical_and(correct_prediction, is_label_zero)))
# FALSE NEGATIVES
false_negatives = tf.reduce_sum(tf.to_int32(tf.logical_and(false_prediction, is_label_one)))
return true_positives, false_positives, true_negatives, false_negatives
# Now you can do something like this in your session:
true_positives, \
false_positives, \
true_negatives, \
false_negatives = sess.run(evaluation(logits,labels), feed_dict=feed_dict)
# you can print the confusion matrix using the 4 values from above, or get precision and recall:
precision = float(true_positives) / float(true_positives+false_positives)
recall = float(true_positives) / float(true_positives+false_negatives)
# or F1 score:
F1_score = 2 * ( precision * recall ) / ( precision+recall )
目前,我使用此解决方案获得混淆矩阵:
# load the data
(train_x, train_y), (dev_x, dev_y), (test_x, test_y) = dataLoader.load()
# build the classifier
classifier = tf.estimator.DNNClassifier(...)
# train the classifier
classifier.train(input_fn=lambda:train_input_fn(), steps=1000)
# evaluate and prediction on the test set
test_evaluate = classifier.evaluate(input_fn=lambda:eval_input_fn())
test_predict = classifier.predict(input_fn = lambda:eval_input_fn())
# parse the prediction to retrieve the predicted labels
predictions = []
for i in list(test_predict):
predictions.append(i['class_ids'][0])
# build the prediction matrix
matrix = tf.confusion_matrix(test_y, predictions)
#display the prediction matrix
with tf.Session():
print(str(tf.Tensor.eval(matrix)))
但我的循环无法说服我检索预测的标签…应该有更好的Python方法来实现这一点…(或TensorFlow方法…这是什么是打印矩阵?抱歉,我没有编写该函数。实际上,它只是使用pandas交叉表函数来创建混淆矩阵。你的意思是如果我说直接打印pandas.crosstab(y_pred,y_actu),那么它就工作了!!不,请检查更新的代码,我添加了一个函数print_conflusion_matrix(),除了x_gold和y_gold之外,我做了所有的工作…无论我分配什么,例如x_批和y_批,我都会得到错误,这应该是由于np.array不匹配…TypeError:unhable type:'numpy.ndarray'…我的y批正在工作,但是我的x_批如下:[[1 118 16…,0 0 0][3 213 27…,0 0 0][1856 1786 0…,0 0 0],[2284 19 7…,0 0 0][1 1094 1028…,0 0 0][3 736 206…,0 0 0]]你认为这应该是问题吗?
# load the data
(train_x, train_y), (dev_x, dev_y), (test_x, test_y) = dataLoader.load()
# build the classifier
classifier = tf.estimator.DNNClassifier(...)
# train the classifier
classifier.train(input_fn=lambda:train_input_fn(), steps=1000)
# evaluate and prediction on the test set
test_evaluate = classifier.evaluate(input_fn=lambda:eval_input_fn())
test_predict = classifier.predict(input_fn = lambda:eval_input_fn())
# parse the prediction to retrieve the predicted labels
predictions = []
for i in list(test_predict):
predictions.append(i['class_ids'][0])
# build the prediction matrix
matrix = tf.confusion_matrix(test_y, predictions)
#display the prediction matrix
with tf.Session():
print(str(tf.Tensor.eval(matrix)))