Question

我正在尝试将tf.confusion_matrix与tf.assign_add一起使用，以便在每个全局步骤中更新一个混淆矩阵。

y_true = tf.placeholder(tf.int16,shape=[None,])
y_pred = tf.placeholder(tf.int16,shape=[None,])
cm = tf.confusion_matrix(labels=y_true,predictions=y_pred)
cm_inc = tf.assign_add(cm, cm) #error

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer()) 
    for i in range(5):
        feed_dict={y_true:[0,1,0,0],y_pred:[1,1,0,0]}
        print(sess.run(cm_inc,feed_dict=feed_dict))

但是我得到

AttributeError: 'Tensor' object has no attribute 'assign_add'

有没有一种方法可以访问混淆矩阵对象的实际矩阵，即cm来完成这项工作。

Answer 1

您得到

AttributeError：“ Tensor”对象没有属性“ assign_add”

因为assign_add仅对变量有意义（请参见my other answer）

如您所知，应该创建一个空变量（所有条目均为零），并将assign_add的结果import tensorflow as tf NUM_CLASSES = 4 y_true = tf.placeholder(tf.int16, shape=[None, ]) y_pred = tf.placeholder(tf.int16, shape=[None, ]) cm_diff = tf.confusion_matrix(labels=y_true, predictions=y_pred, num_classes=NUM_CLASSES) cm = tf.get_variable('confusion_matrix', [NUM_CLASSES, NUM_CLASSES], dtype=tf.int32, initializer=tf.zeros_initializer()) cm_inc = tf.assign_add(cm, cm_diff) with tf.Session() as sess: sess.run(tf.global_variables_initializer()) for i in range(5): feed_dict = {y_true: [0, 1, 0, 0], y_pred: [1, 1, 0, 0]} print(sess.run(cm_inc, feed_dict=feed_dict))赋给该变量，例如

tf.confusion_matrix与tf.assign_add

1 个答案: