Hi All,
I’m trying to transform an input tensor via a key-value mapping (which is defined via 2 tensor representing the keys and values separately) to a desired output. I did think of using a simple for-loop but I fear that that approach would potentially overwrite previous mapping if I had a larger key_tensor and value_tensor.
Is there an existing function in PyTorch can do the following efficiently?
input_tensor = torch.tensor([[ 9, 36, 36],
[18, 36, 36],
[18, 36, 36],
[18, 36, 36]])
key_tensor = torch.tensor([36, 18, 9])
value_tensor = torch.tensor([0, 1, 2])
def expected_func(input, keys, values):
"""
some amazing code here...
"""
return expected_output
expected_output = tensor([[ 2, 0, 0],
[ 1, 0, 0],
[ 1, 0, 0],
[ 1, 0, 0]])