Time Series Forecasting with Facebook Prophet

Posted on Edited on In

Today, we are going to test out Facebook Prophet by following this DigitalOcean Tutorial.

Preparation

Required Python Packages

We FIRST make sure 2 Python packages - Prophet and PyStan have been successfully installed.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
➜  ~ pip show prophet
Name: prophet
Version: 0.1.1
Summary: Microframework for analyzing financial markets.
Home-page: http://prophet.michaelsu.io/
Author: Michael Su
Author-email: mdasu1@gmail.com
License: BSD
Location: /home/jiapei/.local/lib/python3.6/site-packages
Requires: six, pytz, pandas
Required-by: 
➜  ~ pip show pystan
Name: pystan
Version: 2.18.1.0
Summary: Python interface to Stan, a package for Bayesian inference
Home-page: https://github.com/stan-dev/pystan
Author: None
Author-email: None
License: GPLv3
Location: /home/jiapei/.local/lib/python3.6/site-packages
Requires: Cython, numpy
Required-by: fbprophet

Download the Time Series Data

We just need to download the CSV file to some directory:

1
2
3
4
➜  facebookprophet curl -O https://assets.digitalocean.com/articles/eng_python/prophet/AirPassengers.csv
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  1748  100  1748    0     0   2281      0 --:--:-- --:--:-- --:--:--  2279

Test

The Code

Trivial modifications have been done upon the code on this DigitalOcean Tutorial, as follows:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
import pandas as pd
from fbprophet import Prophet

import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')

df = pd.read_csv('AirPassengers.csv')
df.head(5)
df.dtypes

df['Month'] = pd.DatetimeIndex(df['Month'])
df.dtypes

df = df.rename(columns={'Month': 'ds', 'AirPassengers': 'y'})
df.head(5)

ax = df.set_index('ds').plot(figsize=(12, 8))
ax.set_ylabel('Monthly Number of Airline Passengers')
ax.set_xlabel('Date')

plt.show()

# set the uncertainty interval to 95% (the Prophet default is 80%)
my_model = Prophet(interval_width=0.95)
my_model.fit(df)
future_dates = my_model.make_future_dataframe(periods=36, freq='MS')
future_dates.tail()
forecast = my_model.predict(future_dates)
forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail()
fig1 = my_model.plot(forecast, uncertainty=True)
fig1.show()

my_model.plot_components(forecast).savefig('prophet_forcast.png');

Outcome

Original Data
Forcast
Forcast Component