def __init__(self, ni, nh, no): """Creates a new network with ni input nodes, nh hidden nodes, and no output nodes""" self.ni = ni + 1 # +1 for the bias node (to be replaced in the future) self.nh = nh self.no = no # create the node objects self.ai = np.ndarray(self.ni, object) self.ah = np.ndarray(self.nh, object) self.ao = np.ndarray(self.no, object) for x in xrange(self.ni): self.ai[x] = node() for x in xrange(self.nh): self.ah[x] = node() for x in xrange(self.no): self.ao[x] = node() # create the weight matrices and set them to random values self.wi = np.ndarray( (self.nh, self.ni), object ) self.wo = np.ndarray( (self.no, self.nh), object ) for x in xrange(self.nh): for y in xrange(self.ni): self.wi[x,y] = edge(random.rand(), self.ah[x], self.ai[y]) for z in xrange(self.no): self.wo[z,x] = edge(random.rand(), self.ao[z], self.ah[x]) # create hidden weight matrix, a symmetric matrix with all diagonal # entries equal to zero self.wh = np.ndarray( (self.nh, self.nh), object ) for i in xrange(self.nh): for j in xrange(i, self.nh): if i == j: self.wh[i,j] = edge(0, self.ah[i], self.ah[j]) else: self.wh[i,j] = self.wh[j,i] = edge(random.rand(), self.ah[i], self.ah[j])
def update(self, inputs): """Updates the value of the neuron activations based on the given input""" if len(inputs) != self.ni - 1: raise ValueError, "wrong number of inputs" # set input node activations for c, v in enumerate(inputs): self.ai[c] = node(v) # set hidden node activations for c, n in enumerate(self.ah): n.update(np.dot(self.wi[c], self.ai) + np.dot(self.wh[c], self.ah)) # set output node activations for c, n in enumerate(self.ao): n.update(np.dot(self.wo[c], self.ah)) return self.ao
