【计算机科学前沿】第四章答案 2022 - 人工神经网络

第4章

4.1 利用神经网络完成商区分类任务

4.1.1 认识数据集

x_train, y_train, l_train = load_dataset("data/train.npz")
fig() + scatter(x_train, y_train, c=l_train)

4.1.2 搭建人工神经网络

model = DeepLearning([2,4,4,1])

fig() + structure(model)

4.1.3 神经网络训练

f_train = merge_features([x_train, y_train])

model.demo_train('business')

model.show_learning_curve()

4.1.4 神经网络测试

x_test, y_test, l_test = load_dataset("data/test.npz")

f_test = merge_features([x_test, y_test])

pred = model.predict(f_test)

print(accuracy(pred > 0.5, l_test))

4.1.5 参数调整与重新训练

model.reset_weights()

model.loss = 'mse'

model.demo_train('business_mse')

model.show_learning_curve()

pred = model.predict(f_test)

print(accuracy(pred > 0.5, l_test))

4.2 基于身高预测体重

4.2.1 写入实验数据

height, weight = load('hw.train')

height, weight = load('hw.train')

4.2.2 构建神经网络

net = MLP([1, 4, 4, 1])
fig() + structure(net)

4.2.3 神经网络训练

net.train(height, weight)

4.2.4 神经网络测试

t_height, t_weight = load('hw.test') 
pred = net.predict(data=t_height) 
fig() + scatter(t_height, pred) + scatter(t_height, t_weight)

error = net.compute_error(pred, t_weight)
print("Test error = %f" % error)