tutorials/Tutorial-3-Validating-the-Accuracy-of-a-Model.md

+
+### Overview
+This tutorial will show you how to generate the `validation()` method in addition to the standard `eval()`
+method and how to use it to validate that the C++ implementation of your network model matches the 
+TensorFlow original. This will build upon the convolutional MNIST example used in tutorials 1 & 2, you'll
+need to add some code to the python and C++ sources.
+
+### Generating the Validate Method
+In order to generate the `validate()` method we need to have a set of example input data to be used for the validation
+case. These are passed to the `generate()` as a dictionary where the input placeholder names are keys
+and numpy.ndarrays values.
+
+In this example we'll use the first element of MNIST training data for validation:
+```python
+validation_input = mnist.test.images[:1]
+```
+Add this to the call to `generate()` as shown below, and set the validation_type parameter to `Full`.
+```python
+c_exporter.generate("tfmin_generated/mnist_model",
+                    "mnistModel",
+                    validation_inputs={"input/x-input": validation_input},
+                    validation_type='Full',
+                    timing=True,
+                    layout='RowMajor')
+``` 
+If you now run the python script again the output should the following lines at the bottom of its output:
+```
+Analysed flow-graph
+Optimised memory map.
+Generated constructor
+Generated inference method.
+Generated validation method.
+Generated timing method.
+Generated data header.
+Complete
+```
+
+### The Validate Method
+The declaration of the validate that has been generated by the modified python script is shown below:
+```
+template <typename Device>
+bool YourObjectName::validate(const Device &d);
+```
+Since the `validate()` method is working on the provided validation data it doesn't require any input or
+output buffers like the eval or timing methods. This method prints out the validation result of each
+operation of the model as it is being processed, if an operation fails validation then it stops executing
+the model and returns false. If all operations pass validation then the method returns true.
+
+### Using the Validate Method in your C++ Project
+It is simpler to use the `validate()` method than either the eval or timing methods, it's only parameter
+is the Eigen device object and it returns a Boolean indicating the result of the validation task. 
+
+```cpp
+std::cout << "about to verify model." << std::endl;
+if (mnist.validate(device))
+    std::cout << "Verification Passed." << std::endl;
+else
+    std::cout << "Error: Verification Failed." << std::endl;
+```
+Copy and pasted this code snippet after the timing call in the test_mnist.cpp source file of the mnist_conv
+example after the timing, and it should display a successful validation result as shown below:
+
+```
+Validating model computations against stored results from TensorFlow
+Eigen Version 3.3.90
+About to perform Reshape operation [Reshape]
+Passed
+About to perform Conv2D operation [Conv1/convolution]
+Passed
+About to perform Mul operation [Conv1/activations/mul]
+Passed
+About to perform Maximum operation [Conv1/activations/Maximum]
+Passed
+About to perform MaxPool operation [Conv1/pooling]
+Passed
+About to perform Reshape operation [Conv1/Reshape_4]
+Passed
+About to perform MatMul operation [Dense1/Wx_plus_b/MatMul]
+Passed
+About to perform Add operation [Dense1/Wx_plus_b/add]
+Passed
+About to perform Mul operation [Dense1/activation/mul]
+Passed
+About to perform Maximum operation [Dense1/activation/Maximum]
+Passed
+About to perform MatMul operation [layer2/Wx_plus_b/MatMul]
+Passed
+About to perform Add operation [layer2/Wx_plus_b/add]
+Passed
+Verification Passed.
+```
+
+### Summary
+This tutorial should have demonstrated how to generate use the validate() method of the MNIST inference model to
+validate that the C++ implementation produces the same results as the TensorFlow original.
+
+---
+[Previous Tutorial](/Tutorials/Tutorial-2-Evaluating-the-Runtime-of-a-Model) - Evaluating the Runtime of a Model
+
+[Next Tutorial](/Tutorials/Tutorial-4-Adding-Support-for-More-Operations) - Adding Support for More Operations
+