Assignment 5

directories /test , /app, /serve

Inside /test

1. The unit test in unit.js

   • helpers.getRandomInt(2) should return integer 0 or 1
   • helpers.getRandomArbitrary(-1,1) should return a number between -1 and 1
   • ann.neuronSigmoidActivation should return a number
   • ann.neuronSigmoidActivation should return 0.8808 for specific input values
   • ann.neuronSigmoidActivation should not throw when threshold is not a number.
   • ann.neuronSigmoidActivation should not throw when some input is not a number.
   • ann.neuronSigmoidActivation should not throw when some weight is not a number.
   • ann.initializeNetwork should not throw when passing non integer values
   • ann.initializeNetwork should return an array with specific elements lengths for given input
   • ann.activate should not throw when passing a state that is not an array
   • ann.activate should not throw when passing invalid network input
   • ann.activate should not throw when passing invalid network a non-array as desired output
   • ann.activate should return an object
   • ann.activate should return expected values for given input
   • ann.backPropagate should return an array with expected value for given input
   • ann.backPropagate should not throw when passing a non-array previous state
   • ann.backPropagate should not throw when arrays's lengths do not comply
   • ann.backPropagate should not throw when alpha is not a number
   • ann.cycle response should be well formed and have the right values
   • ann.epoch response should be well formed
   • ann.train response should be well formed

2. The api test in api.js Quite minimal here.

   • app.init should start without throwing
   • http /ping should respond with 200
   • http /ann should respond with 200
   • page not found should respond with 404

Inside /app

Some application related functionality. The application assumption is about training a feedforward artificial neural network to implement the XOR logic. Backpropagation is used as training algorithm. Neurons activation function used, is sigmoid.

Some artificial neural networks functionality in lib.js

• initializeNetwork(neurons). Generates an ANN with neurons per layer as specified in input array and random input weights and thresholds. Returns the network state.
• neuronSigmoidActivation((weights, inputs, threshold). Returns neuron output, after implementing the sigmoid activation function in a neuron with specified threshold, for some specified inputs with some specified input weights.
• activate(network,inputs,desiredOutput). Returns an array with all outputs (final and intermediate), along with the error, for the given network state, network inputs and desired output.
• backPropagate(network,outputs,err, a). For the given network state, network layers' outputs, error and learning rate, computes the corection values and adds them to the existing. Returns a new network state
• addStates(stateA, stateB). Helper function that adds two network states. Returns the new one.
• cycle(initialState, inputs, desiredOutputs, a,callback). Forward and backward propagation. Network adjustments
• epoch(initialState, dataset, a, callback). Iterates applying values from a dataset to the ANN
• train(initialState, dataset, epochs, a, callback). Repeats for a number of epochs

Some helper functions in helpers.js

• getRandomInt(max). Returns a random integer in range 0 to max-1.
• getRandomArbitrary(min, max). Returns a random number between min and max.

Inside /serve

The usual server functionality.