Full Model Retrain Sample

[sierralee51]

A sample using the full model retrain API with the MNIST model and data.

[sierralee51]

working on the hard-coded parts that I didn't notice before right now

[CESARDELATORRE]

As part of this PR, could you create a full-retrain sample using a more complex image dataset like the one we have in the sample below? Basically, photos of multiple classes, then measuring the real accuracy when performing predictions, etc.

You could evolve that transfer-learning sample bewlow into a new sample using full-retrain:

https://github.com/dotnet/machinelearning-samples/blob/migration/v1.3/samples/csharp/getting-started/DeepLearning_TensorFlow_TransferLearning/ImageClassification.Train/Program.cs

Basically, I'd like to know what's the accuracy (based on predictions from a test imageset) of a real image classification sample scenario like the datasets used in the above sample so we're sure about a minimum quality bar before we make this feature available to users and announce it publicly in a blog post, etc. - Can the dev team provide that sample and check the accuracy with those datasets, etc. please?

[codemzs]