matmul not working for rank 3 tensors (batch_matmul)

[wilocw]

When trying to apply matmul on rank 3 tensors, I get an error that the tensors must be rank 2. Using TensorFlow in python allows me to perform tensor multiplication, but I am unable to do it in Julia. This function is possible with batch_matmul but it throws the deprecation warning.

The following MWE is based on the tf.matmul documentation.

Python version

>>> import sys; print(sys.executable)
/home/wil/anaconda3/envs/PyCall.jl/bin/python
>>> import numpy as np
>>> import tensorflow as tf; print(tf.__version__)
1.10.0
>>> a = tf.constant(np.arange(1,13,dtype=np.int32), shape=[2,2,3])
>>> a
<tf.Tensor 'Const:0' shape=(2, 2, 3) dtype=int32>
>>> b = tf.constant(np.arange(13,25,dtype=np.int32),shape=[2,3,2])
>>> b
<tf.Tensor 'Const_1:0' shape=(2, 3, 2) dtype=int32>
>>> c = a@b # equiv to c = tf.matmul(a,b)
>>> c
<tf.Tensor 'matmul:0' shape=(2, 2, 2) dtype=int32>

Julia version

julia> ENV["PYTHON"]
"/home/wil/anaconda3/envs/PyCall.jl/bin/python"

julia> using TensorFlow

julia> a = constant(reshape(Int32.(1:12), 2,2,3))
<Tensor Const:1 shape=(2, 2, 3) dtype=Int32>

julia> b = constant(reshape(Int32.(13:24), 2,3,2))
<Tensor Const_2:1 shape=(2, 3, 2) dtype=Int32>

julia> a*b
ERROR: Tensorflow error: Status: Shape must be rank 2 but is rank 3 for 'MatMul' (op: 'MatMul') with input shapes: [2,2,3], [2,3,2].

Stacktrace:
 [1] check_status at /home/wil/.julia/packages/TensorFlow/A6TdG/src/core.jl:374 [inlined]
 [2] Operation(::TensorFlow.NodeDescription) at /home/wil/.julia/packages/TensorFlow/A6TdG/src/core.jl:1043
 [3] #mat_mul#96(::Nothing, ::Nothing, ::Nothing, ::Function, ::Tensor{Int32}, ::Tensor{Int32}) at /home/wil/.julia/packages/TensorFlow/A6TdG/src/ops/imported_ops.jl:683
 [4] mat_mul(::Tensor{Int32}, ::Tensor{Int32}) at /home/wil/.julia/packages/TensorFlow/A6TdG/src/ops/imported_ops.jl:668
 [5] #*#91(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}, ::Function, ::Tensor{Int32}, ::Tensor{Int32}) at /home/wil/.julia/packages/TensorFlow/A6TdG/src/ops.jl:34
 [6] *(::Tensor{Int32}, ::Tensor{Int32}) at /home/wil/.julia/packages/TensorFlow/A6TdG/src/ops.jl:34
 [7] top-level scope at none:0

julia> batch_matmul(a,b)
┌ Warning: batch_matmul is deprecated. Its functionality is now subsumed by matmul.
│   caller = top-level scope at none:0
└ @ Core none:0
<Tensor BatchMatMul_6:1 shape=(2, 2, 2) dtype=Int32>

Version Info

> tf_versioninfo()
----------------
Library Versions
----------------
Trying to evaluate ENV["TF_USE_GPU"] but got error: KeyError("TF_USE_GPU")
Trying to evaluate ENV["LIBTENSORFLOW"] but got error: KeyError("LIBTENSORFLOW")

tf_version(kind=:backend) = 1.10.0
Trying to evaluate tf_version(kind=:python) but got error: RemoteException(2, CapturedException(UndefVarError(:py_tf), Any[(#95 at TensorFlow.jl:183, 1), (#116 at process_messages.jl:276, 1), (run_work_thunk at process_messages.jl:56, 1), (run_work_thunk at process_messages.jl:65, 1), (#102 at task.jl:259, 1)]))
tf_version(kind=:julia) = 0.10.2

-------------
Python Status
-------------
PyCall.conda = true
ENV["PYTHON"] = /home/wil/anaconda3/envs/PyCall.jl/bin/python
PyCall.PYTHONHOME = /home/wil/anaconda3/envs/PyCall.jl:/home/wil/anaconda3/envs/PyCall.jl
String(read(#= /home/wil/.julia/packages/TensorFlow/A6TdG/src/version.jl:104 =# @cmd("pip --version"))) = pip 18.1 from /home/wil/anaconda3/lib/python3.6/site-packages/pip (python 3.6)

String(read(#= /home/wil/.julia/packages/TensorFlow/A6TdG/src/version.jl:105 =# @cmd("pip3 --version"))) = pip 18.1 from /home/wil/anaconda3/lib/python3.6/site-packages/pip (python 3.6)


------------
Julia Status
------------
Julia Version 1.0.1
Commit 0d713926f8 (2018-09-29 19:05 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i5-7600 CPU @ 3.50GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, skylake)
Environment:
  JULIA_EDITOR = atom -a
  JULIA_NUM_THREADS = 4