NumPy-like slicing in TensorFlow

user_7934786 0
Viewed 1405

I have a tensor object, and I want to slice part of it.

tf_a1 = tf.Variable([    [9.968594,  8.655439,  0.,        0.       ],
                         [0.,        8.3356,    0.,        8.8974   ],
                         [0.,        0.,        6.103182,  7.330564 ],
                         [6.609862,  0.,        3.0614321, 0.       ],
                         [9.497023,  0.,        3.8914037, 0.       ],
                         [0.,        8.457685,  8.602337,  0.       ],
                         [0.,        0.,        5.826657,  8.283971 ],
                         [0.,        0.,        0.,        0.       ]])

Also, I have this array:

tf_a2 = tf.constant([[1, 2, 5],
                    [1, 4, 6],
                    [0, 7, 7],
                    [2, 3, 6],
                    [2, 4, 7]])

I want to do this numpy like slicing:

 tf_a1[tf_a2]

The expected output from the numpy code will be like this:

[[[0.        8.3356    0.        8.8974   ]
  [0.        0.        6.103182  7.330564 ]
  [0.        8.457685  8.602337  0.       ]]

 [[0.        8.3356    0.        8.8974   ]
  [9.497023  0.        3.8914037 0.       ]
  [0.        0.        5.826657  8.283971 ]]

 [[9.968594  8.655439  0.        0.       ]
  [0.        0.        0.        0.       ]
  [0.        0.        0.        0.       ]]

 [[0.        0.        6.103182  7.330564 ]
  [6.609862  0.        3.0614321 0.       ]
  [0.        0.        5.826657  8.283971 ]]

 [[0.        0.        6.103182  7.330564 ]
  [9.497023  0.        3.8914037 0.       ]
  [0.        0.        0.        0.       ]]]

I thought I can do a similar operation in tensorflow using:

tf.gather_nd(tf_a1, tf_a2)

but it raises this error:

tensorflow.python.framework.errors_impl.InvalidArgumentError: index innermost dimension length must be <= params rank; saw: 3 vs. 2 [Op:GatherNd]

Any help is appreciated:)

tensor tensorflow slice python
1 Answers

I think you can use tf.gather:

tf.gather(tf_a1, tf_a2, axis=0)                                                                                        
# <tf.Tensor 'GatherV2_10:0' shape=(5, 3, 4) dtype=float32>

Reproducible Example on TensorFlow 2.0

tf.__version__
# '2.0.0-beta0'

tf.gather(tf_a1, tf_a2, axis=0)

<tf.Tensor: id=9, shape=(5, 3, 4), dtype=float32, numpy=
array([[[0.       , 8.3356   , 0.       , 8.8974   ],
        [0.       , 0.       , 6.103182 , 7.330564 ],
        [0.       , 8.457685 , 8.602337 , 0.       ]],

       [[0.       , 8.3356   , 0.       , 8.8974   ],
        [9.497023 , 0.       , 3.8914037, 0.       ],
        [0.       , 0.       , 5.826657 , 8.283971 ]],

       [[9.968594 , 8.655439 , 0.       , 0.       ],
        [0.       , 0.       , 0.       , 0.       ],
        [0.       , 0.       , 0.       , 0.       ]],

       [[0.       , 0.       , 6.103182 , 7.330564 ],
        [6.609862 , 0.       , 3.0614321, 0.       ],
        [0.       , 0.       , 5.826657 , 8.283971 ]],

       [[0.       , 0.       , 6.103182 , 7.330564 ],
        [9.497023 , 0.       , 3.8914037, 0.       ],
        [0.       , 0.       , 0.       , 0.       ]]], dtype=float32)>
user_49090870
Related