FailedPreconditionError使用tensorlfow.metrics.recall時

Question

我想使用tensorlfow.metrics.recall和tensorlfow.metrics.precision，但tensorflow引起FailedPreconditionError。請注意，我用sess.run(tf.global_variables_initializer())

的代碼如下：

import numpy as np 
import tensorflow as tf 

y_true = np.array([[1, 1, 0], [0, 1, 1], [0, 0, 1]]) 
y_pred = np.array([[1, 0, 0], [1, 0, 1], [0, 0, 1]]) 

predictions = tf.placeholder(tf.int64, shape=[3, 3]) 
labels = tf.placeholder(tf.int64, shape=[3, 3]) 
prec = tf.metrics.precision(labels, predictions) 
rec = tf.metrics.recall(labels, predictions) 
with tf.Session() as sess: 
    sess.run(tf.global_variables_initializer()) 
    p, r = sess.run([prec, rec], feed_dict={predictions: y_pred, labels: y_true}) 
    print("precision: {}, recall: {}".format(p, r)) 

• tensorflow版本：1.2.1
• Python版本：3.5.2

Answer 1

tf.metrics.precision和tf.metrics.recall內部創建local variable（使用collections=[tf.GraphKeys.LOCAL_VARIABLES]創建的變量）。所以你需要做：sess.run(tf.local_variables_initializer())。