如何像 Pytorch 一樣在 Tensorflow 中分配張量？

Question

我正在嘗試將 Pytorch 腳本轉換為 Tensorflow 腳本。 但我無法像 pytorch 一樣在 tensorflow 中分配張量。

代碼：

import torch
import tensorflow as tf


def true_positive(pred, target, num_classes): #number of classes
    out = []
    for i in range(num_classes):
        out.append(((pred == i) & (target == i)).sum())

    return torch.tensor(out)

Pytorch 實現：工作中

p = torch.tensor([1]) 
t = torch.tensor([2])
n = torch.tensor([2])
y = true_positive(p,t,n)

TensorFlow 實現：不工作！

p = tf.constant([1]) #c = tf.constant([[1.0, 2.0], [3.0, 4.0]])
t = tf.constant([2])
n = tf.constant([2])
y = true_positive(p,t,n)

錯誤 ：

-------------------------------------------------- ------------------------- TypeError Traceback (最近一次調用最后一次) Input In [18], in <cell line: 22>() 20 t = tf.constant([2]) 21 n = tf.constant([2]) ---> 22 y = true_positive(p,t,n)

Input In [18], in true_positive(pred, target, num_classes) 5 def true_positive(pred, target, num_classes): #number of classes 6 out = [] ----> 7 for i in range(num_classes): 8 out.append(((pred == i) & (target == i)).sum()) 10 return torch.tensor(out)

文件 ~/opt/anaconda3/lib/python3.9/site-packages/tensorflow/python/framework/ops.py:1131，在 _EagerTensorBase 中。 index (self) 1130 def index (self): -> 1131 return self._numpy()。 索引()

TypeError：只有整數標量數組可以轉換為標量索引

Answer 1

也許是這樣的：

import tensorflow as tf


def true_positive(pred, target, num_classes): #number of classes
    out = []
    for i in tf.range(num_classes):
      out.append(tf.reduce_sum(tf.cast((pred == i) & (target == i), dtype=tf.int32)))
    return tf.stack(out)

p = tf.constant([4]) #c = tf.constant([[1.0, 2.0], [3.0, 4.0]])
t = tf.constant([4])
n = tf.constant([4])
y = true_positive(p,t,n)
y
# <tf.Tensor: shape=(4,), dtype=int32, numpy=array([0, 0, 0, 0], dtype=int32)>