def get_fitness_gc(dna, kernel, train_set_num=500, test_set_num=1000):
# Get Data
mnist = data_reader.read_data_sets("./scripts/data")
# Change net kernels
converter = CnnConverter(
param_file_path="./config/converter_config_default.json")
converter.ac_kernel(kernel)
# Create net
net = CnnNetwork(converter)
net.create_inputlayer(converter.pro_width * converter.pro_height *
converter.kernels_num)
net.create_outputlayer(10)
net.link_inputlayer_outputlayer()
# train
net.train_all(mnist.train.data[:train_set_num],
mnist.train.target[:train_set_num],
iter_max=2,
accuracy_rate=.95)
# predict
accuracy = net.predict_all(mnist.test.data[:test_set_num],
mnist.test.target[:test_set_num],
is_record_error_msg=False)
# log result
if net.synchronizer.rank == 0:
print("{ DNA: ", dna, ",accuracy: ", accuracy, "}")
del net
return accuracy
def get_fitness(net, kernels, test_set_num=1000):
# Get Data
mnist = data_reader.read_data_sets("./scripts/data")
# Change net kernels
net.converter.change_kernal(kernels)
# predict
accuracy = net.predict_all(mnist.test.data[:test_set_num],
mnist.test.target[:test_set_num])
return accuracy
def read_result(path):
re_scv = np.loadtxt(path, dtype=np.int)
return re_scv
if __name__ == "__main__":
disconnect_w = 1. # adjust parameter
training_set_num = 1000
testing_set_num = 10000
classifier_num = 1
mnist = data_reader.read_data_sets("./scripts/data")
# 参数 转换函数，图片的宽高
converter = CnnConverter(
param_file_path="./config/converter_config_default.json")
# 设置teacher信号的位置
converter.teacher = 75.
print(converter.pro_width)
print(converter.pro_height)
is_fail = False
all_result = []
rest_synapse = []
# train
for _ in range(classifier_num):
intermediate_result[key] = self.convolution_layer(intermediate_result[key], kernal=self.kernels[key])
elif op == 'pooling':
for key in self.kernels.keys():
intermediate_result[key] = self.max_pooling_layer(intermediate_result[key])
return intermediate_result
@staticmethod
def draw_data(ori_image, new_image=221):
plt.subplot(new_image)
plt.imshow(ori_image)
# Unit Test
if __name__ == "__main__":
mnist = data_read.read_data_sets("../../scripts/data")
converter = CnnConverter(param_file_path='../../config/converter_config_default.json')
image = mnist.train.data[1]
print("Target %s" % mnist.train.target[1])
print(converter.pro_width, converter.pro_height)
pre = converter.pre_pro_c(image)
num = 1
for value in pre.values():
converter.draw_data(value, 220 + num)
num += 1
if num > 4:
num = 1
plt.figure()