In this assignment you will complete the skeleton code for an ANN. It is a multi-layer perceptron (MLP). The number of hidden layers is 1 consisting of 16 neurons.

1. Start with loading and visualization of data.
2. Flatten the pixels into a 1-D array and treat that as input. Remember this is a classification problem so create output layer accordingly. Figure out architecture accordingly.
3. As mentioned in the code. The trick is to think in terms of matrices and layer-wise computation instead of neuron-wise.
4. In the test function, write a code to measure the accuracy of your model. Use the accuracy function provided in utils.
5. Demo Loss functions, activation function etc. are provided in utils.py. In this project use the cross-entropy loss, sigmoid activation for hidden units and softmax for output units.