Kaggle competition two Sigma connect: rental listing inquiries

Kaggle competition , Website links ：Two Sigma Connect: Rental Listing Inquiries

According to the data information on the rental website , Predict the popularity of the house .（ This is a question of classification , Contains the following data , Variable with category 、 Integer variable 、 Text variable ）.

Random forest model

Use sklearn Complete modeling and prediction . The data set can be downloaded from the official website of the competition .

import numpy as np
import pandas as pd
import zipfile  # The official website data set is zip type , Use zipfile open 
import os
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import log_loss
for dirname, _, filenames in os.walk(r'E:\Kaggle\Kaggle_dataset01\two_sigma'):   # Change your path 
    for filename in filenames:
        print(os.path.join(dirname, filename))

train_df = pd.read_json(zipfile.ZipFile(r'E:\Kaggle\Kaggle_dataset01\two_sigma\train.json.zip').open('train.json'))
test_df = pd.read_json(zipfile.ZipFile(r'E:\Kaggle\Kaggle_dataset01\two_sigma\test.json.zip').open('test.json'))

# Here is a customized data processing function .
def data_preprocessing(data):
    data['created_year'] = pd.to_datetime(data['created']).dt.year
    data['created_month'] = pd.to_datetime(data['created']).dt.month
    data['created_day'] = pd.to_datetime(data['created']).dt.day
    data['num_description_words'] = data['description'].apply(lambda x:len(x.split(' ')))
    data['num_features'] = data['features'].apply(len)
    data['num_photos'] = data['photos'].apply(len)
    New_data = data[['created_year','created_month','created_day','num_description_words','num_features','num_photos','bathrooms','bedrooms','latitude','longitude','price']]
    return New_data
train_x = data_preprocessing(train_df)
train_y = train_df['interest_level']
test_x = data_preprocessing(test_df)
X_train,X_val,y_train,y_val = train_test_split(train_x,train_y,test_size=0.33)  # Data segmentation 

clf = RandomForestClassifier(n_estimators=1000)   # Random forest model 
clf.fit(X_train,y_train)
y_val_pred = clf.predict_proba(X_val)
log_loss(y_val,y_val_pred)

y_test_predict = clf.predict_proba(test_x)
labels2idx = {
    label:i for i,label in enumerate(clf.classes_)}
sub = pd.DataFrame()
sub['listing_id'] = df['listing_id']
for label in labels2idx.keys():
    sub[label] = y[:,labels2idx[label]]
# Save the submission 
#sub.to_csv('submission.csv',index=False) # Competition submission ！

Run the above code , The effect of random forest is not very good . Some people will ask why there is no normalization preprocessing for data ？ In fact, there is no need to normalize the data when using random forest , So I didn't do . If you want to do it , Try to verify it yourself . If you want to use random forest to improve the robustness of the model , Consider improving the feature engineering part , Get better features ！