如何在tensorflow中使用索引数组？

Question

如果给定矩阵a与形状(5,3)和索引阵列b与形状(5,)，我们可以很容易地通过获得相应的载体c，

c = a[np.arange(5), b]

但是，我不能做同样的事情tensorflow，

a = tf.placeholder(tf.float32, shape=(5, 3)) 
b = tf.placeholder(tf.int32, [5,]) 
# this line throws error 
c = a[tf.range(5), b] 

Traceback (most recent call last): File "", line 1, in File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 513, in _SliceHelper name=name)

File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 671, in strided_slice shrink_axis_mask=shrink_axis_mask) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/gen_array_ops.py", line 3688, in strided_slice shrink_axis_mask=shrink_axis_mask, name=name) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 763, in apply_op op_def=op_def) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2397, in create_op set_shapes_for_outputs(ret) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1757, in set_shapes_for_outputs shapes = shape_func(op) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1707, in call_with_requiring return call_cpp_shape_fn(op, require_shape_fn=True) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 610, in call_cpp_shape_fn debug_python_shape_fn, require_shape_fn) File "~/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 675, in _call_cpp_shape_fn_impl raise ValueError(err.message) ValueError: Shape must be rank 1 but is rank 2 for 'strided_slice_14' (op: 'StridedSlice') with input shapes: [5,3], [2,5], [2,5], [2].

我的问题是，如果我不能产生预期的resul t在tensorflow中使用上述方法在numpy中，我应该怎么做？

Answer 1

此功能目前尚未在TensorFlow中实施。 GitHub issue #4638正在跟踪NumPy风格的“高级”索引的实现。但是，您可以使用tf.gather_nd()运营商来实现您的程序：

a = tf.placeholder(tf.float32, shape=(5, 3)) 
b = tf.placeholder(tf.int32, (5,)) 

row_indices = tf.range(5) 

# `indices` is a 5 x 2 matrix of coordinates into `a`. 
indices = tf.transpose([row_indices, b]) 

c = tf.gather_nd(a, indices)