如何从张量中获取前 3 个最大数字

Question

我如何从y_classe = tf.argmax(preds, axis=1, output_type=tf.int32)获得前 3 个最大数字？

Answer 1

You can sort and take the first 3:

import tensorflow as tf

a = [1, 10, 26.9, 2.8, 166.32, 62.3]
sorted_a = tf.sort(a,direction='DESCENDING')
max_3 = tf.gather(sorted_a, [0,1,2])
print(max_3)

Answer 2

You can use tf.math.top_k :

import tensorflow as tf

y_pred = [[-18.6, 0.51, 2.94, -12.8]]

max_entries = 3

values, indices = tf.math.top_k(y_pred, k=max_entries)
print(values)
print(indices)

tf.Tensor([[  2.94   0.51 -12.8 ]], shape=(1, 3), dtype=float32)
tf.Tensor([[2 1 3]], shape=(1, 3), dtype=int32)