tensorflow

Question

我尝试确定我的神经网络模型的 confusion_matrix 使用 google tensorflow 在 python 中编写。通过使用这段代码：

cm = tf.zeros(shape=[2,2], dtype=tf.int32)

for i in range(0, validation_data.shape[0], batch_size_validation):
    batched_val_data = np.array(validation_data[i:i+batch_size_validation, :, :], dtype='float')
    batched_val_labels = np.array(validation_labels[i:i+batch_size_validation, :], dtype='float')

    batched_val_data = batched_val_data.reshape((-1, n_chunks, chunk_size))            

    _acc, _c, _p = sess.run([accuracy, correct, pred], feed_dict=({x:batched_val_data, y:batched_val_labels}))

    #batched_val_labels.shape ==> (2048, 2)
    #_p.shape                 ==> (2048, 2)
    #this piece of code throws the error!
    cm = tf.confusion_matrix(labels=batched_val_labels, predictions=_p)

我收到以下错误： ValueError: 形状 (2, 2048, 2) 必须具有等级 2

至少你应该知道验证标签数组 batched_val_labels 是一个 one hot array。有人可以帮我吗？提前致谢！

Answer 1

问题是我使用的是 one hot array。按照此说明操作：Tensorflow confusion matrix using one-hot code

我修改了这段代码：

cm = tf.confusion_matrix(labels=batched_val_labels, predictions=_p)

进入：

cm = tf.confusion_matrix(labels=tf.argmax(batched_val_labels, 1), predictions=tf.argmax(_p, 1))

tensorflow - tf.confusion_matrix() throws error ValueError: Shape (2, 2048, 2) must have rank 2

tensorflow - tf.confusion_matrix() throws error ValueError: Shape (2, 2048, 2) must have rank 2

python

python-3.x

confusion-matrix