NodeDef missing attr 'shape' from Op

[yeesian]

I'm getting the error message

julia> Operation(node)
ERROR: Tensorflow error: Status: NodeDef missing attr 'shape' from Op<name=Variable; signature= -> ref:Ref(dtype); attr=shape:shape; attr=dtype:type; attr=container:string,default=""; attr=shared_name:string,default=""; is_stateful=true>; NodeDef: hidden1_1/weights = Variable[_class=[], container="", dtype=DT_FLOAT, shared_name=""]()

 in check_status(::TensorFlow.Status) at /Users/yeesian/.julia/v0.5/TensorFlow/src/core.jl:95
 in TensorFlow.Operation(::TensorFlow.NodeDescription) at /Users/yeesian/.julia/v0.5/TensorFlow/src/core.jl:482
 in TensorFlow.Operation(::TensorFlow.tensorflow.NodeDef) at /Users/yeesian/.julia/v0.5/TensorFlow/src/core.jl:672

on the following object:

julia> dump(node)
TensorFlow.tensorflow.NodeDef
  name: String "hidden1_1/weights"
  op: String "Variable"
  input: Array{AbstractString}((0,))
  device: String ""
  attr: Dict{AbstractString,TensorFlow.tensorflow.AttrValue}
    slots: Array{UInt8}((16,)) UInt8[0x01,0x00,0x01,0x00,0x00,0x01,0x01,0x00,0x00,0x01,0x00,0x00,0x00,0x00,0x00,0x00]
    keys: Array{AbstractString}((16,))
      1: String "dtype"
      2: #undef
      3: String "shape"
      4: #undef
      5: #undef
      ...
      12: #undef
      13: #undef
      14: #undef
      15: #undef
      16: #undef
    vals: Array{TensorFlow.tensorflow.AttrValue}((16,))
      1: TensorFlow.tensorflow.AttrValue
        s: #undef
        i: Int64 0
        f: Float32 0.0
        b: Bool false
        _type: Int32 1
        shape: #undef
        tensor: #undef
        list: #undef
        placeholder: #undef
      2: #undef
      3: TensorFlow.tensorflow.AttrValue
        s: #undef
        i: Int64 0
        f: Float32 0.0
        b: Bool false
        _type: Int32 0
        shape: TensorFlow.tensorflow.TensorShapeProto
          dim: Array{TensorFlow.tensorflow.TensorShapeProto_Dim}((2,))
            1: TensorFlow.tensorflow.TensorShapeProto_Dim
              size: Int64 28
              name: String ""
            2: TensorFlow.tensorflow.TensorShapeProto_Dim
              size: Int64 128
              name: String ""
          unknown_rank: Bool false
        tensor: #undef
        list: #undef
        placeholder: #undef
      4: #undef
      5: #undef
      ...
      12: #undef
      13: #undef
      14: #undef
      15: #undef
      16: #undef
    ndel: Int64 0
    count: Int64 5
    age: UInt64 5
    idxfloor: Int64 1
    maxprobe: Int64 0

[malmaud]

Hmm, where did that NodeDef come from? Are you sure it's actually valid? Can you link me to to the raw protobytes for it so I can examine it?

[yeesian]

I'm not sure anymore, since I can't reproduce it anymore, sorry. Closing this issue for now, unless it comes up again.

The closest I've come to reproducing the same NodeDef is in this file:

  | | |_| | | | (_| |  |  Version 0.5.0 (2016-09-19 18:14 UTC)
 _/ |\__'_|_|_|\__'_|  |  Official http://julialang.org/ release
|__/                   |  x86_64-apple-darwin13.4.0

julia> using TensorFlow; const TF = TensorFlow
TensorFlow

julia> using ProtoBuf

julia> nodedef = open("nodedef.proto") do f
           readproto(f, TF.tensorflow.NodeDef())
       end
name: hidden1_1/weights
op: Variable
attr {
  key: dtype
  value {
    type: Float32
  }
}
attr {
  key: shape
  value {
    shape: [28,128]
  }
}
attr {
  key: _output_shapes
  value {
      }
}
attr {
  key: shared_name
  value {
    string:
  }
}
attr {
  key: container
  value {
    string:
  }
}


julia> dump(nodedef)
TensorFlow.tensorflow.NodeDef
  name: String "hidden1_1/weights"
  op: String "Variable"
  input: Array{AbstractString}((0,))
  device: String ""
  attr: Dict{AbstractString,TensorFlow.tensorflow.AttrValue}
    slots: Array{UInt8}((16,)) UInt8[0x01,0x00,0x01,0x00,0x00,0x01,0x01,0x00,0x00,0x01,0x00,0x00,0x00,0x00,0x00,0x00]
    keys: Array{AbstractString}((16,))
      1: String "dtype"
      2: #undef
      3: String "shape"
      4: #undef
      5: #undef
      ...
      12: #undef
      13: #undef
      14: #undef
      15: #undef
      16: #undef
    vals: Array{TensorFlow.tensorflow.AttrValue}((16,))
      1: TensorFlow.tensorflow.AttrValue
        s: #undef
        i: Int64 0
        f: Float32 0.0
        b: Bool false
        _type: Int32 1
        shape: #undef
        tensor: #undef
        list: #undef
        placeholder: #undef
      2: #undef
      3: TensorFlow.tensorflow.AttrValue
        s: #undef
        i: Int64 0
        f: Float32 0.0
        b: Bool false
        _type: Int32 0
        shape: TensorFlow.tensorflow.TensorShapeProto
          dim: Array{TensorFlow.tensorflow.TensorShapeProto_Dim}((2,))
            1: TensorFlow.tensorflow.TensorShapeProto_Dim
              size: Int64 28
              name: String ""
            2: TensorFlow.tensorflow.TensorShapeProto_Dim
              size: Int64 128
              name: String ""
          unknown_rank: Bool false
        tensor: #undef
        list: #undef
        placeholder: #undef
      4: #undef
      5: #undef
      ...
      12: #undef
      13: #undef
      14: #undef
      15: #undef
      16: #undef
    ndel: Int64 0
    count: Int64 5
    age: UInt64 5
    idxfloor: Int64 1
    maxprobe: Int64 0

which now works fine for me:

julia> TF.Operation(nodedef)
<Operation 'hidden1_1/weights'>