Join GitHub today
GitHub is home to over 40 million developers working together to host and review code, manage projects, and build software together.Sign up
mismatch in the description of BatchDot and TensorFlow's implementation. #30846
URL(s) with the issue:
Description of issue (what needs changing):
I raised an issue on the plaidML repo, and after some back and forth we determined the documentation for BatchDot doesn't quite match the actual implementation in the tensorflow code.
A BatchDot with x.shape=(1,2,6,2) and y.shape=(1,2,2,3) and axes = (3, 1)has an output shape of (1,2,6,3)) whereas by the TF definition for output shape "A tensor with shape equal to the concatenation of x's shape (less the dimension that was summed over) and y's shape (less the batch dimension and the dimension that was summed over). If the final rank is 1, we reshape it to (batch_size, 1)." sounds like it should have an output shape of (1,2,6,2,3).
Submit a pull request?
I am not planning to submit a PR at this time, but I may do it later
To further elaborate on this issue: there also seems to be output shape mismatches between the TensorFlow and Theano backend implementations of
Could we either change the documentation to more clearly describe TensorFlow's engineering decisions behind