tf.TensorBuffer.get returns unexpected value

[vabarbosa]

To get help from the community, we encourage using Stack Overflow and the tensorflow.js tag.

TensorFlow.js version

0.15.3

Browser version

Firefox 65.0.1
Chrome 72.0.3626.119

Describe the problem or feature request

i may be wrong in my understanding but my expectation is that TensorBuffer.get should return the value at a given index/location. if an out of bound index/location is provided i would expect it to fail. however, that does not appear to be the case. if you provide a location for a given dimension greater than its size tf.TensorBuffer.get may still return a value.

Code to reproduce the bug / link to feature request

comparing tf.TensorBuffer.get and indexing with tf.Tensor.arraySync:

// [ [ [1, 2], [3, 4] ],
//   [ [5, 6], [7, 8] ] ]
const t = tf.tensor([1, 2, 3, 4, 5, 6, 7, 8], [2, 2, 2])

const b = t.bufferSync()
const a = t.arraySync()

b.get(0, 0, 0) // 1
a[0][0][0]     // 1

b.get(0, 0, 1) // 2
a[0][0][1]     // 2

b.get(0, 0, 2) // 3
a[0][0][2]     // undefined

b.get(0, 2, 0) // 5
a[0][2][0]     // undefined

b.get(2, 0, 0) // undefined
a[2][0][0]     // undefined

b.get(2, 0, 0) correctly returns undefined. however, b.get(0, 0, 2) and b.get(0, 2, 0) incorrectly return a value (for an invalid location).

[bileschi]

tensorflow/tfjs-core#1630