[PYTHON] Easy time series prediction with Prophet

1.First of all

Recently, I tried using Prophet for easy time series prediction. Since it's a big deal, I decided to share everything from introduction to operation, so I wrote an article.

• Operating environment
• OS : Windows10
• Python : 3.8.3 // Miniconda 4.9.1
• Visual Studio Community : 2019

2. What is Prophet?

Prophet is a time-series forecast package released by facebook as OSS (MIT license). The detailed theory is not touched upon here, but predictions are made using a generalized additive model.

Official page github

3. Create a virtual environment for verification and switch to the virtual environment

python

#prophet_A virtual environment called env in python3.Created with 7 series(3.8 series fbprophet 0.Not compatible with 6)
conda create -n prophet_env python=3.7

#prophet created from base environment_Switch to env virtual environment
conda activate prophet_env

4. Installation-operation check

First, install the Visual Stuido Community.

• C ++ Compiler is required Visual Studio Code Site

Then install the following libraries

python

#Prophet introduced because it depends on PyStan
conda install plotly -y

#Introduced fbprophet(2020/11/Version at 1 is 0.7.It was 1, but I got an error, so 0.6)
conda install -c conda-forge fbprophet==0.6

#Because the environment is miniconda ...(I didn't put numpy and pandas individually because they happened to be introduced together in ↑)
conda install matplotlib

If no error occurs with the following command, installation is successful

python

from fbprophet import Prophet

5. Actually use

This time, we will use a simple data set called "Air Passengers" as a sample data set. The Month column contains the time-series part, and the Passengers column contains the changes in the number of passengers. Dataset link

5-1: Until file reading

python

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from fbprophet import Prophet

#Read csv of AirPassengers
df = pd.read_csv('AirPassengers.csv')

5-2: Learn Prophet

python

#Prophet needs to change the column names to ds and y, so change them
df.columns = ['ds', 'y']

#This time, learn from the back of the data frame to the 10th in Prophet(fit)Let
m = Prophet()
m.fit(df[:-10])

5-3: Predicted by Prophet

python

'''
periods=Predict 10:10 squares
freq='M': Unit is "Month"
→ In other words, it means to predict for 10 months
'''
future = m.make_future_dataframe(periods=10, freq='M')

forecast = m.predict(future)

#Illustrate the result of forecast
fig = m.plot(forecast)

It should be displayed as ↓.

• Black dots are learning data

6. Finally

How was it? The program itself works easily, so it may be more difficult to install. Different versions may cause an error somewhere, and in fact I stumbled there at first. (PyStan, libpython, etc. were introduced together with fbprophet == 0.6, so they were not introduced individually in the end) It's easy, so please give it a try.