Time Series Analysis with Facebook’s Prophet

The ability to predict and forecast future events and outcome is essential to any business and organization.

Fortunately there are several tools and procedure to enable us to do so. One of these procedures is time series analysis.  So what is time series analysis?

First of all let us define a time series and then what time series analysis is?

A time series is a series of data points indexed (or listed or graphed) in time order. Most commonly, a time series is a sequence taken at successive equally spaced points in time. Thus it is a sequence of discrete-time data.

Time series analysis refers to the analysis of  a time series to extract meaningful insight in order to identify trends and pattern and in order to forecast future events.

Time series forecasting is the use of a model to predict future values based on previously observed values.

Applications of Time Series Analysis

• To identify trends or patterns
• To identify seasonal changes and cycles
• To understand the past in order to predict the future
• For forecasting future events,outcomes
• Ex -useful for forecasting market for products that fluctuate seasonally, such as commodities and clothing retail businesses

Now let us see some packages we can use to do time series analysis in Python and in Julia.

For Python

• Pandas: for data manipulation using datetime features
• Statsmodels: A collection of tools and algorithms for doing statistical modeling and time series analysis
• FbProphet(Facebook Prophet): A simple and easy package made by Facebook for time series forecasting.
• PmProphet: A similar package to fbprophet for time series forecasting
• Pyflux

For Julia

• DataFrames.jl : for data manipulations
• StatsModels.jl
• TimeSeries.jl : a lightweight framework for working with time series data in Julia.

Working with Facebook Prophet

Facebook Prophet utilizes an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects for forecasting time series data.

To Install

pip install pystan
pip install fbprophet

Steps/Workflow For Using FB Prophet

• Initialize Model :: Prophet()
• Set columns as ds,y
• Fit dataset :: Prophet().fit()
• Create Dates To predict :: Prophet().make_future_dataframe(periods=365)
• Predict :: Prophet().predict(future_dates)
• Plot :: Prophet().plot(predictions)

Let us start .

# Load EDA Pkgs
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline

# Load FB Prophet
import fbprophet

dir(fbprophet)

['Prophet',
 '__builtins__',
 '__cached__',
 '__doc__',
 '__file__',
 '__loader__',
 '__name__',
 '__package__',
 '__path__',
 '__spec__',
 '__version__',
 'diagnostics',
 'forecaster',
 'hdays',
 'make_holidays',
 'models',
 'plot']

# Load our Dataset
df = pd.read_csv("flights_data.csv")

df.head()

In [60]:

df.plot()

Out[60]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f96e9a03d68>

#yt = yt -y(t-1)
df['no_of_flights'] = df['no_of_flights'] - df['no_of_flights'].shift(1)

In [63]:

df.plot()

Out[63]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f96e9492898>

from fbprophet import Prophet

# Features of Prophet
dir(Prophet)

['__class__',
 '__delattr__',
 '__dict__',
 '__dir__',
 '__doc__',
 '__eq__',
 '__format__',
 '__ge__',
 '__getattribute__',
 '__gt__',
 '__hash__',
 '__init__',
 '__init_subclass__',
 '__le__',
 '__lt__',
 '__module__',
 '__ne__',
 '__new__',
 '__reduce__',
 '__reduce_ex__',
 '__repr__',
 '__setattr__',
 '__sizeof__',
 '__str__',
 '__subclasshook__',
 '__weakref__',
 'add_country_holidays',
 'add_group_component',
 'add_regressor',
 'add_seasonality',
 'construct_holiday_dataframe',
 'fit',
 'fourier_series',
 'initialize_scales',
 'linear_growth_init',
 'logistic_growth_init',
 'make_all_seasonality_features',
 'make_future_dataframe',
 'make_holiday_features',
 'make_seasonality_features',
 'parse_seasonality_args',
 'piecewise_linear',
 'piecewise_logistic',
 'plot',
 'plot_components',
 'predict',
 'predict_seasonal_components',
 'predict_trend',
 'predict_uncertainty',
 'predictive_samples',
 'regressor_column_matrix',
 'sample_model',
 'sample_posterior_predictive',
 'sample_predictive_trend',
 'set_auto_seasonalities',
 'set_changepoints',
 'setup_dataframe',
 'validate_column_name',
 'validate_inputs']

# Initialize the Model
model = Prophet()

df.columns

Index(['Dates', 'no_of_flights'], dtype='object')

# Works with a ds and y column names
df.rename(columns={'Dates':'ds','no_of_flights':'y'},inplace=True)

df.head()

df = df[1:]

df.head()

# Fit our Model to our Data
model.fit(df)

/usr/local/lib/python3.6/dist-packages/fbprophet/forecaster.py:880: FutureWarning: Series.nonzero() is deprecated and will be removed in a future version.Use Series.to_numpy().nonzero() instead
  min_dt = dt.iloc[dt.nonzero()[0]].min()
INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this.
INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.

<fbprophet.forecaster.Prophet at 0x7f96e980c240>

# Shape of Dataset
df.shape

(35, 2)

# Create Future Dates of 365 days
future_dates = model.make_future_dataframe(periods=365)

# Shape after adding 365 days
future_dates.shape

(400, 1)

future_dates.head()

# Make Prediction with our Model
prediction = model.predict(future_dates)

prediction.head()

In [79]:

# Plot Our Predictions
model.plot(prediction)

Out[79]:

In [80]:

# Visualize Each Component [Trends,Weekly]
model.plot_components(prediction)

Out[80]:

# Load Pkgs
from fbprophet.diagnostics import cross_validation

df.shape

(35, 2)

cv = cross_validation(model,initial='35 days', period='180 days', horizon = '365 days')

INFO:fbprophet:Making 4 forecasts with cutoffs between 2005-06-09 00:00:00 and 2006-12-01 00:00:00
/usr/local/lib/python3.6/dist-packages/fbprophet/forecaster.py:880: FutureWarning: Series.nonzero() is deprecated and will be removed in a future version.Use Series.to_numpy().nonzero() instead
  min_dt = dt.iloc[dt.nonzero()[0]].min()
INFO:fbprophet:n_changepoints greater than number of observations.Using 3.0.
INFO:fbprophet:n_changepoints greater than number of observations.Using 7.0.
INFO:fbprophet:n_changepoints greater than number of observations.Using 12.0.
INFO:fbprophet:n_changepoints greater than number of observations.Using 17.0.

cv.head()

from fbprophet.diagnostics import performance_metrics

df_pm = performance_metrics(cv)

df_pm

from fbprophet.plot import plot_cross_validation_metric

In [91]:

plot_cross_validation_metric(cv,metric='rmse')

Out[91]:

You can also get the code and the dataset here.

To get a video tutorial you can check the video below on youtube.

Thanks for reading.

Below are some interesting books for mastering Data Science and Machine Learning in python

Python Cookbook
Python For Data Analysis :
Python Data Science HandBook
Python Machine Learning by Sebastian Raschka
Hands On Machine Learning with Scikit-Learn & TensorFlow
Mastering ML with Scikitlearn
Monetizing ML

Thanks For Your Attention

Jesus Saves

By Jesse E. Agbe(JCharis)