Properly Load Microsoft.ML model in Xamarin App? - machine-learning

Question

I was learning a bit about how Machine Learning works and eventually built a small C# application because of a tutorial I was watching where a model determines what species of Bee a particular bee IS. It works well in the C# application, but I was also developing an app project that had picture taking capabilities in mind so I thought "No reason the model I saved in the first project can't work in the second, right?". Apparently there is a reason why it can't work because here is the necessary code to view in the app project:
MLContext m_mlObj;
DataViewSchema m_modelSchema;
ITransformer m_loadedTrainedModel;
var folderPath = DependencyService.Get<IFileSystem>().GetExternalStorage();
var fileDir = Path.Combine(folderPath, "trainedModel.zip");
bool testValue = File.Exists(fileDir);
if(testValue)
{
Console.WriteLine("File in fact exists.");
try
{
m_loadedTrainedModel = m_mlObj.Model.Load(fileDir, out m_modelSchema);
}
catch (Exception e)
{
Console.WriteLine("\n\n\nInner exception: " + e.InnerException);
}
}
else
{
Console.WriteLine("File does not exist");
}
I have a breakpoint on the "m_loadedTrainedModel = m_mlObj.Model.Load(fileDir, out m_modelSchema);" line of course and it always triggers an exception.
The exception states:
---> System.TypeLoadException: Could not load type of field 'Microsoft.ML.Transforms.DnnRetrainTransformer:_tfInputShapes' (9) due to: Could not resolve type with token 01000060 from typeref (expected class 'Tensorflow.TensorShape' in assembly 'TensorFlow.NET, Version=0.20.1.0, Culture=neutral, PublicKeyToken=cc7b13ffcd2ddd51') assembly:TensorFlow.NET, Version=0.20.1.0, Culture=neutral, PublicKeyToken=cc7b13ffcd2ddd5 1 type:Tensorflow.TensorShape member:(null)
What I THINK this means is it's checking for Tensorflow.TensorShape inside of the package TensorFlow.NET and isn't finding it. If that's the case, I believe I don't have the correct package installed so my Xamarin project will be able to load the model correctly, maybe? Or if not, then what may be the real underlying issue? The code seems straight forward so I'm a bit perplexed as to how this is failing.
If the solution is blatantly going over my head, my apologies and thank you to anyone who is willing to help.

Answer 1

When working with ML.NET on ARM, TensorFlow is one of the limitations at the time of this writing.
https://devblogs.microsoft.com/dotnet/ml-net-june-updates-model-builder/#ml-net-on-arm
What you might want to consider in the meantime is deploying your model as a Web API and making requests to that API from your mobile app.
https://learn.microsoft.com/dotnet/machine-learning/how-to-guides/serve-model-web-api-ml-net