Generic Time Series Forecasting

What's in the box?

A Python script containing different time series forecasting methods. It takes as input set of time series in a tabular format (EXCEL) and output forecast time series.

Current available methods are:

• Seasonal ARIMA
• Facebook Prophet
• Custom method including a seasonal decomposition and a trend forecast using a gaussian process

Limitations

As these methods work by analyzing the macro trend and the seasonal patterns, there should be enough cycles included in the data. A minimum of 3 years of monthly data is required for the Seasonal ARIMA model.

Installation Tips

1. Running the code presumes that you have a distribution of Python (>=3.5) installed on your machine with a certain numbers of libraries.
2. The library containing the seasonal Arima can easily be installed with the command: pip install https://github.com/statsmodels/statsmodels/archive/master.zip
3. The bin folder contains the binary source code required to run the models along with the compiler and instructions (For Mac). For Mac users, here is a tip to successfully compile the executable.

User Guide

This repository contains a single forecasting scripts. It can be run from within an IDE (ex: Pycharm, Spyder etc.) or from a command line.
Note:

• All input files must be placed in the folder /data/hist_data/ and all output files will be saved in the folder /data/fcst_data/.
• The spreadsheet containing the input data should be named 'Sheet1' by default (if excel format).

Run from command line

The script takes as an input an excel file containing the time series to be forecasted and outputs an excel file containing the forecasted time series.

Script name: Monthly_Forecasting.py

$ python Monthly_Forecasting.py fcst_method start_date end_date input_file_name output_file_name\
$ python Monthly_Forecasting.py sarima 2017-03-01 2017-12-01 input_file_name.xlsx output_file_name.xlsx

Forecasting methods can be one of the following: 'sarima', 'sarimax', 'gaussian', 'facebook'

Note: It is possible to use the Seasonal ARIMA with exogenous regressors (method 'sarimax'). In that case the name of the exogenous file must be modified in the script.