pixie-price-forecast 0.3.0

Price forecast tools.

Pixie Price Forecast

This repository's main aim is to get an overview of the price tendency throughout the next few months.

Install Requirements

To begin with, please install all the python modules required to run the scripts.

pip install -r requirements.txt

Running ETL

In order to generate the aggregated dataset you can run the following code:

from price_forecast.etl import export_aggregated_data

export_aggregated_data(mode={mode})

Where {mode} can be either 'demand' or 'offer'. The former corresponds to average daily data per province for ads with at least 1 lead. The latter computes data using ads no matter the amount of leads.

Forecasting Prices

As of today, it is possible to forecast the property price per province using one of the following models:

• Arima
• Random Forest: Using leads, visits and 3-6 months lags as drivers.

You can use the following code to run the forecast.

path = [location of input]
provinces = [list of provinces you are interested in]
search_grid = {dictionary with model parameters you want to try in a search grid}
model_name = [either 'arima' or 'random_forest']
date_train_window = [date with the following format '%Y-%m-%d']
run_forecast(model_name=model_name, input_path=path, province_list=provinces, search_grid=search_grid,
             train_from=date_train_window[0], train_to=date_train_window[1])

Results

Results can be found in this Tableau Dashboard.

In case you do not have access, please contact Jose Mielgo