(pytorch – Deep Learning Series) simple implementation of kaggle house price forecast

Import the required modules ：

Reading data sets ：
（ The details shall be subject to the location of the data set stored by yourself ）

Basic information of data set ：

The first feature is Id, For machine learning ,id Can’t bring effective feature information , So we don’t use this attribute as a feature .

Preprocess the data ：

After standardization , The mean value of each numerical characteristic becomes 0, So you can use it directly 0 To replace the missing value
Use one_hot Coding decomposes discrete features into multiple features , The decomposed features can be used 0/1 To express , such , This transformation converts the number of features from 79 Increased to 331

Define the functions required by the model ：

Given predicted value y ^ 1 , … , y ^ n hat y_1, ldots, hat y_n y^​1​,…,y^​n​ And the corresponding real label y 1 , … , y n y_1,ldots, y_n y1​,…,yn​, The root mean square error of logarithm is defined as

1 n ∑ i = 1 n ( log ⁡ ( y i ) − log ⁡ ( y ^ i ) ) 2 . sqrt{frac{1}{n}sum_{i=1}^nleft(log(y_i)-log(hat y_i)right)^2}. n1​i=1∑n​(log(yi​)−log(y^​i​))2 ​.

Realization K Crossover verification ：

.
.
Training models ：

After the above code is executed, a submission.csv file , The document conforms to kaggle Submission format for , You can directly re kaggle Submit the game link .