Machine-Learning

compute numerical gradient

feature normalise

fmincg

optimise the cost function use [final_params, cost] = fmincg(@nn_cost_function, nn_params, options) options = optimset('MaxIter', 50)

• input: cost function
• output: optimised weights, cost
• functions called: none

nn cost function

computes the cost function and gradient of a set of paramaters and X and y, for multiclass classification with num_labels.

• input: nn_params, input_layer_size, hidden_layer_size, num_labels, X, y, lambda
• output: J, grad
• functions called: sigmoid(), sigmoid_gradient()

predict

predicts output for multiclass clasification three layer NN

• input: Theta_1, Theta_2, x
• output: p
• functions used: sigmoid()

rand initialise weights

Randomly initialises the weights for one layer

• input: l_in - input nodes for that layer, l_out - output nodes for that layer
• output: w - matrix of weights
• functions called: none

sigmoid

Computes the sigmoid function

• Input: z
• Output: g
• Functions called: none

sigmoid gradient

Computes the gradient of the sigmoid function

• Input: z
• Output: g
• Functions called: sigmoid()