System.MissingMethodException IntPtr Tensorflow.Status.op_Implicit(Tensorflow.Status)''

Question

I'm trying to load TensorFlow model(SavedModel) to ML.NET and create an estimator with ML.NET.

I am loading the model without errors, but tensorFlowModel.ScoreTensorFlowModel got Exception "System.MissingMethodException IntPtr Tensorflow.Status.op_Implicit(Tensorflow.Status)".

My code:

using Microsoft.ML;

var inputNodeName = "serving_default_dense_input";
var outputNodeName = "StatefulPartitionedCall_1";

var tensorFlowModel = mlContext.Model.LoadTensorFlowModel(@"model"); //model directory contains saved_model.pb file
var pipeline = tensorFlowModel.ScoreTensorFlowModel(new[] { outputNodeName },new[] { inputNodeName }, addBatchDimensionInput: false);

Installed NuGet packages:

• ML.NET
• SciSharp.TensorFlow.Redist
• TensorFlow.NET

These packages are updated to the latest version.

Answer 1

I've run into this same error a few times and it's always been a mismatch between ML.NET and TensorFlow versions. I noticed you've included SciSharp.TensorFlow.Redist package and while there are probably tutorials that still mention using that later versions of ML.NET have a dependency on TensorFlow.NET >= 0.11.8.1 that seems to include the necessary files so that's not necessary and a likely cause of your problem.

The other thing is while the dependency is >= 0.11.8.1 later versions of TensorFlow.NET can cause the the same problem as per this Github Issue so also check you haven't updated to a later version of TensorFlow.NET and if you have try reverting back to 0.11.8.1.