Using tensorflow to forecast the rental price of airbnb in New York City

author |TIMOTHY102 compile |VK source |Analytics Vidhya

Introduce

Airbnb It's an online market , Allow people to rent their property or spare rooms to guests . Per booking 3 Guests , collect 12% and 6% Commission .

The company started from 2009 Since its establishment in , Has been helped from every year 2.1 Ten thousand guests found accommodation , To help every year 600 Ten thousand people on holiday , Currently in 90 From different countries 34000 Cities list amazing 80 Ten thousand houses .

In this paper , I will use Kaggle-newyorkcityairbnb Open data set , Try to use TensorFlow Build a neural network model to predict .

The goal is to build a suitable machine learning model , Be able to predict the price of future accommodation data .

In this paper , I'm going to show you what I've created Jupyter Notebook. You can GitHub Find it on the ：https://github.com/Timothy102/Tensorflow-for-Airbnb-Prices

Load data

First , Let's see how to load data . We use it wget Directly from Kaggle Get data on the website . Be careful -o The flag indicates the file name .

The dataset should look like this . share 48895 That's ok 16 Column .

Data analysis and preprocessing

Seaborn There's a very simple API, You can draw all kinds of graphs for all kinds of data . If you are not familiar with grammar , Check out this article ：https://www.analyticsvidhya.com/blog/2019/09/comprehensive-data-visualization-guide-seaborn-python/

stay pandas Use on data frame corr after , We pass it on to a heatmap function . give the result as follows ：

Since we have longitude and longitude and neighborhood data , Let's create a scatter plot ：

Besides , I've deleted duplicate items and some unnecessary Columns , And filled in “reviews_per_month”, Because it has too many missing values . The data looks like this . It has 10 Column , There is no zero value ：

very good , Right ？

First , Computers do numbers . That's why we need to convert a sort column into a one-hot Encoded vector . This is the use of pandas Of factorize Method . You can use a lot of other tools :

In order to keep the loss function in a stable range , Let's normalize some data , Let the average be 0, The standard deviation is 1.

Feature crossover

We have to make a change , This is an essential change . To correlate longitude and latitude with model output , We have to create a feature crossover . The following links should provide you with sufficient background knowledge , So that you can feel the cross of features correctly ：

• https://developers.google.com/machine-learning/crash-course/feature-crosses/video-lecture
• https://www.kaggle.com/vikramtiwari/feature-crosses-tensorflow-mlcc

Our goal is to introduce latitude longitude crossing , This is one of the oldest techniques in the book . If we just put these two columns in the model as values , It will assume that these values are gradually related to the output .

contrary , We're going to use feature crossover , That means we're going to put longitude * The longitude map is divided into a grid . Fortunately, ,TensorFlow Make it easy .

I go through iteration （max-min）/100, So as to generate a frame grid with uniform distribution .

I use it 100×100 grid ：

Essentially , What we're doing here , Is to define a bucked Columns and the boundaries defined earlier , And create a DenseFeatures layer , Then pass it to Sequential API.

If you're not familiar with it Tensorflow grammar , Please check the documentation ：https://www.tensorflow.org/api_docs/python/tf/feature_column/

Now? , finally , We are ready for model training . Apart from splitting the data part , in other words .

obviously , We have to create two datasets , One contains all the data , The other contains the predicted score . Due to data size mismatch , This may cause problems for our model , So I decided to truncate data that was too long .

Creating models

Last , Established Keras Sequence model .

We use Adam Optimizer 、 Mean square error loss and two metrics to compile the model .

Besides , We use two callbacks ：

• Stop early , This is self-evident

• Reduce the learning rate at high altitude .

after 50 individual epoch Training for ,batch The size is 64, Our model is quite successful .

ending

We use New York City AirBnB The data builds a fully connected neural network to predict future prices .Pandas and seaborn It makes it very easy to visualize and examine data . We introduce the idea of latitude longitude crossing as a feature in the model . And thanks to that Kaggle Open data set of , We have a fully operational machine learning model .

Link to the original text ：https://www.analyticsvidhya.com/blog/2020/10/predicting-nyc-airbnb-rental-prices-tensorflow/

Welcome to join us AI Blog station ： http://panchuang.net/

sklearn Machine learning Chinese official documents ： http://sklearn123.com/

Welcome to pay attention to pan Chuang blog resource summary station ： http://docs.panchuang.net/