In following Lyndon White’s example of TensorFlow at http://white.ucc.asn.au/2017/01/24/JuliaML-and-TensorFlow-Tuitorial.html I was able to get it to run successfully after two changes:
- reduce_sum uses the now deprecated (and refused on my 1.4 dev TF) reduction_indices - changing this to axis worked fine
- “get_shape(Y_obs .* log(Y_pred))” was not able to evaluate the argument with message
ERROR: syntax: “.*(Y_obs,log(Y_pred))” is not a valid function argument name
However taking the argument and assigning it to a variable and then using the new variable name allowed it to work correctly. Just wondering what subtle change in Julia is responsible here.