class XorFunctionIntegrationTest(unittest.TestCase): def setUp(self): self.networkFile = 'network_XOR_SIG.xml' self.network = NetParser(COMMON_DIR_PREFIX + self.networkFile).parse() def tearDown(self): pass def testWithSigmoidActivationFunction(self): input_vector = InputVectorParser(COMMON_DIR_PREFIX + 'input_0_0.xml').parse() value = self.network.calculte_answer(input_vector) print('f(0,0) = ', str(self.network.calculte_answer(input_vector))) self.assertAlmostEqual(0, value[0], 1) input_vector = InputVectorParser(COMMON_DIR_PREFIX + 'input_1_0.xml').parse() value = self.network.calculte_answer(input_vector) print('f(1,0) = ', str(self.network.calculte_answer(input_vector))) self.assertAlmostEqual(1, value[0], 1) input_vector = InputVectorParser(COMMON_DIR_PREFIX + 'input_0_1.xml').parse() value = self.network.calculte_answer(input_vector) print('f(0,1) = ', str(self.network.calculte_answer(input_vector))) self.assertAlmostEqual(1, value[0], 1) input_vector = InputVectorParser(COMMON_DIR_PREFIX + 'input_1_1.xml').parse() value = self.network.calculte_answer(input_vector) print('f(1,1) = ', str(self.network.calculte_answer(input_vector))) self.assertAlmostEqual(0, value[0], 1) def testWithThresholdActivationFunction(self): pass
def test_backpropagation(self): input_vectors = [InputVectorParser(COMMON_DIR_PREFIX + LEARNING_VECTOR_FILE_PREFIX + str(i) + '.xml').parse() for i in range(1, 5)] for input_vec in input_vectors: print(input_vec) #testVectors = [InputVectorParser(COMMON_DIR_PREFIX + TEST_VECTOR_FILE_PREFIX + str(i) + '.xml').parse() for i in range(1, 4)] network = NetParser(COMMON_DIR_PREFIX + NETWORK_FILE).parse()#lambda : random.uniform(-1.0, 1.0)) learning_rate = 0.7 momentum = 0.1 iterations = 2000 network.backpropagation_learn(input_vectors, learning_rate, iterations, momentum) print(network) error = 0.0 for inputVector in input_vectors : inputVector.printVector(3) outputDict = [] answer = network.calculte_answer(inputVector, False, outputDict) print(answer) for id, val in outputDict: error = error + 0.5*(inputVector.expected_value_dict[id]-val)**2 print(error) print(network.calculte_answer(InputVector({'input1' : 0.5, 'input2' : 0.5})))
weight_function = None if args.random: print( "Weight of links between nodes will be set to random values from range [{}; {}]".format( args.random[0], args.random[1] ) ) weight_function = lambda: random.uniform(args.random[0], args.random[1]) elif args.zeros: print("Weight of links between nodes will be set to 0.") weight_function = lambda: 0.0 network = NetParser(args.network).parse(weight_function) if args.vector_limits: input_vector = RandomInputVectorFactory.create_new( args.vector_limits[0], args.vector_limits[1], network.layers[0].get_nodes_ids() ) print( "Generated vector with values from range [{}; {}]:\n{}".format( args.vector_limits[0], args.vector_limits[1], input_vector ) ) else: input_vectors = [InputVectorParser(vector).parse() for vector in args.vector_file] if args.kohonen: network.learning_process(input_vectors, args.coefficient, args.coefficient_half_life, args.turns) if args.backpropagation: network.backpropagation_learn(input_vectors, args.coefficient, args.turns) else: print("Network response is:") print(network.calculte_answer(input_vectors[0]))