Kaggle-House Prices

前段时间尝试着做了一下kaggle中的House Prices，是一个回归问题，通过对给定的训练集进行分析，来预测测试集中的房屋价格，测试集中的数据主要是房屋特征（features），包活很多比如：卧室的数量，临街否等等共79个，在对其进行处理的过程中，要对数值型的特征进行归一化，而对字符型的特征进行pd.dummies()处理，及将字符用数字0,1,2等来代替，这样才能统筹兼顾对所有的特征进行处理，代码如下：

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
from  mxnet import autograd,gluon,init,nd
from mxnet.gluon import loss as gloss,data as gdata,nn
import gluonbook as gb 

train_data = pd.read_csv(r'D:\data\train.csv')
test_data = pd.read_csv(r'D:\data\test.csv')

#数据预处理
all_features = pd.concat([train_data.iloc[:,1:-1],test_data.iloc[:,1:]],axis = 0,ignore_index= True)
numeric_index = all_features.dtypes[all_features.dtypes != 'object']  #numeric_index 是Series格式
all_features[numeric_index.index] = all_features[numeric_index.index].apply(lambda x: (x - x.mean())/(x.std()))
all_features = all_features.fillna(all_features.mean())
#创建指示特征
all_features = pd.get_dummies(all_features,dummy_na = True) #这一步将(字符串类特征)用数字0和1来进行表示！增加了特征的数目从79到331


#提取训练数据和测试数据
n_train = train_data.shape[0]
train_features = nd.array(all_features[:n_train].values)
test_features = nd.array(all_features[n_train:].values)
train_labels = nd.array(train_data.iloc[:,-1]).reshape((-1,1)) #维度为（n，1）

#训练模型
loss = gloss.L2Loss()
def get_net():
    net = nn.Sequential()
    net.add(nn.Dense(256, activation='relu'),nn.Dense(1))
    net.initialize()
    return net
#定义竞赛所要求的的对数均方根误差
def log_rmse(net,train_features,train_labels):
    clipped_pred = nd.clip(net(train_features),1,float('inf')) #将数据控制在大于1的范围，小于1的数变为1
    rmse = nd.sqrt(2*loss(clipped_pred.log(),train_labels.log()).mean()) #乘以2是因为在gluon中的loss公式中为1/2
    return rmse.asscalar()
#定义训练函数
#这里可加上学习率逐渐衰减的方法，并在网络中加入丢弃层
def train(net,train_features,train_labels,test_features,test_labels,num_opochs,learning_rate,weight_decay,batch_size):
    train_l ,test_l = [],[]
    train_iter = gdata.DataLoader(gdata.ArrayDataset(train_features,train_labels),batch_size,shuffle = True)  #小样本读取数据
    #使用Adam优化
    trainer= gluon.Trainer(net.collect_params(),'adam',{'learning_rate':learning_rate,'wd': weight_decay})
    for epoch in range(num_opochs):
        for X,y in train_iter:
            with autograd.record():
                l = loss(net(X),y)
            l.backward()
            trainer.step(batch_size)
        train_l.append(log_rmse(net,train_features,train_labels))
        if test_labels is not None:
            test_l.append(log_rmse(net,test_features,test_labels))
    return train_l,test_l

#进行K折交叉验证(得到训练集)

def get_k_fold_data(k,i,x,y):
    assert(k>1)
    fold_size = x.shape[0] // k
    x_train,y_train = None,None
    for j in range(k):
        ix = slice(j * fold_size,(j+1) * fold_size)    #slice的这种用法学习了
        x_part,y_part = x[ix,:],y[ix]
        if j == i:
            x_valid,y_valid = x_part,y_part          
        elif x_train is None:                         
            x_train,y_train = x_part,y_part           #给x_train,y_train 第一次赋值
        else:
            x_train = nd.concat(x_train,x_part,dim = 0)  #将（k-1）个数据集合合并为训练集
            y_train = nd.concat(y_train,y_part,dim = 0) 
    return x_train,y_train,x_valid,y_valid


#返回训练和验证的平均误差
def k_fold(k,x_train,y_train,num_epochs,learning_rate,weight_decay,batch_size):
    train_l_sum,valid_l_sum = 0,0
    for i in range(k):
        data = get_k_fold_data(k,i,x_train,y_train)
        net = get_net()
        train_ls , valid_ls = train(net,*data,num_epochs,learning_rate,weight_decay,batch_size)
        train_l_sum += train_ls[-1]  #这里我们只需要在最后一个迭代周期中的损失，因为此时的数值是最优化参数w,b之后得到的最小损失
        valid_l_sum += valid_ls[-1] 
        if i == 0:
            #gb.semilogy具体参数可以查看模型选择、欠拟合和过拟合那一章节的内容
            gb.semilogy(range(1,num_epochs+1),train_ls,'epochs','rmse',range(1, num_epochs + 1), valid_ls,['train', 'valid'])
        print('fold %d,train rmse:%f,valid rmse: %f'%(i,train_ls[-1],valid_ls[-1]))
    return train_l_sum/k,valid_l_sum/k
k, num_epochs, lr, weight_decay, batch_size = 3, 30, 0.1, 20, 64 

train_l, valid_l = k_fold(k, train_features, train_labels, num_epochs,lr,weight_decay, batch_size)
print('%d-fold validation: avg train rmse: %f, avg valid rmse: %f'% (k, train_l, valid_l))


   

进行K折交叉验证之后，可以观察在迭代过程中，训练集合的误差和测试集合的误差，从而进行超参数的调节：

下一步，对整个训练集合进行预测，并生成竞赛所需要提交的.csv文件：

#定义预测函数(这里没有使用上面的K折交叉，而是重新使用完整的训练数据集来重新训练模型，并保存csv格式)
def train_and_pred(train_features,test_features,train_labels,test_data,num_epochs,lr,weight_decay,batch_size):
    net = get_net()
    train_ls ,_ = train(net,train_features,train_labels,None,None,num_epochs,lr,weight_decay,batch_size)
    gb.semilogy(range(1,num_epochs+1),train_ls,'epochs','rmse')
    print('train rmse %f'%(train_ls[-1]))
    preds = net(test_features).asnumpy()
    test_data['SalePrice'] = pd.Series(preds.reshape((-1,1))[:,0])
    submission = pd.concat([test_data['Id'],test_data['SalePrice']],axis  =1)
    submission.to_csv(r'D:\data\submission.csv',index = False)
        
train_and_pred(train_features,test_features,train_labels,test_data,num_epochs,lr,weight_decay,batch_size)

到此基本框架完成，下一步就是调节参数的过程，可以适当增加神经网络的层数、节点数、降低learning rate,增大weight_decay等方法，也可以尝试增加丢弃层等。这个入门级的比赛旨在了解实际情况的建模经验，对数据处理模型建立和参数调节过程有个大概的感觉。

 

小计：

1)np.clip():

numpy.clip(a, a_min, a_max, out=None)，具体用法可参考：https://blog.youkuaiyun.com/HHTNAN/article/details/79799612

2）函数中的*，例如：

train_ls, valid_ls = train(net, *data, num_epochs, learning_rate, weight_decay, batch_size)

* 函数接收参数为元组，** 表示函数接收参数为一个字典。

 

除此方法外，还可以使用Ridge Regression和Random Forest这两个模型来共同预测，然后求取平均值的方法来的到最后的值。大概过程如下：

房价预测案例

Step 1: 检视源数据集

In [5]:

import numpy as np
import pandas as pd

读入数据

• 一般来说源数据的index那一栏没什么用，我们可以用来作为我们pandas dataframe的index。这样之后要是检索起来也省事儿。

In [6]:

train_df = pd.read_csv('../input/train.csv', index_col=0)
test_df = pd.read_csv('../input/test.csv', index_col=0)

检视源数据

In [7]:

train_df.head()

Out[7]:

<

	MSSubClass	MSZoning	LotFrontage	LotArea	Street	Alley	LotShape	LandContour	Utilities	LotConfig	...	PoolArea	PoolQC	Fence	MiscFeature	MiscVal	MoSold	YrSold	SaleType	SaleCondition	SalePrice
Id
1	60	RL	65.0	8450	Pave	NaN	Reg	Lvl	AllPub	Inside	...	0	NaN	NaN	NaN	0	2	2008	WD	Normal	208500
2	20	RL	80.0	9600	Pave	NaN	Reg	Lvl	AllPub	FR2	...	0	NaN	NaN	NaN	0	5	2007	WD	Normal	181500
3	60	RL	68.0	11250	Pave	NaN	IR1	Lvl	AllPub	Inside	...	0	NaN	NaN	NaN	0	9	2008	WD	Normal	223500
4	70	RL	60.0	9550	Pave	NaN	IR1	Lvl	AllPub	Corner	...	0	NaN	NaN	NaN	0	2	2006	WD	Abnorml	140000