Простой способ сделать это - сжать размеры индексов, поэлементно умножить на соответствующий вектор с одной горячей точкой, а затем расширить размеры позже.
import tensorflow as tf
weights = tf.constant([[0.1, 0.2], [0.3, 0.4]])
indices = tf.constant([[1], [0]])
# Reduce from 2d (2, 1) to 1d (2,)
indices1d = tf.squeeze(indices)
# One-hot vector corresponding to the indices. shape (2, 2)
action_one_hot = tf.one_hot(indices=indices1d, depth=weights.shape[1])
# Element-wise multiplication and sum across axis 1 to pick the weight. Shape (2,)
action_taken_weight = tf.reduce_sum(action_one_hot * weights, axis=1)
# Expand the dimension back to have a 2d. Shape (2, 1)
action_taken_weight2d = tf.expand_dims(action_taken_weight, axis=1)
sess = tf.InteractiveSession()
print("weights\n", sess.run(weights))
print("indices\n", sess.run(indices))
print("indices1d\n", sess.run(indices1d))
print("action_one_hot\n", sess.run(action_one_hot))
print("action_taken_weight\n", sess.run(action_taken_weight))
print("action_taken_weight2d\n", sess.run(action_taken_weight2d))
Должен дать вам следующий вывод:
weights
[[0.1 0.2]
[0.3 0.4]]
indices
[[1]
[0]]
indices1d
[1 0]
action_one_hot
[[0. 1.]
[1. 0.]]
action_taken_weight
[0.2 0.3]
action_taken_weight2d
[[0.2]
[0.3]]
Примечание: вы также можете сделать action_taken_weight = tf.reshape(action_taken_weight, tf.shape(indices))
вместо expand_dims.