How to Use Prophet for Forecasting in Python: Simple Guide

To use Prophet for forecasting in Python, first prepare your data with columns ds (date) and y (value). Then, create a Prophet model, fit it with your data, and use model.predict() on a future dataframe to get forecasts.

Syntax

The basic steps to use Prophet are:

• Import Prophet: Load the Prophet class from the prophet package.
• Prepare data: Your data must have two columns: ds for dates and y for values.
• Create model: Initialize a Prophet() object.
• Fit model: Use model.fit(data) to train.
• Make future dataframe: Use model.make_future_dataframe(periods=n) to create dates to predict.
• Predict: Use model.predict(future) to get forecast results.

python

from prophet import Prophet

# Prepare your data as a pandas DataFrame with columns 'ds' and 'y'
# data = pd.DataFrame({'ds': dates, 'y': values})

model = Prophet()
model.fit(data)

future = model.make_future_dataframe(periods=30)  # 30 days into future
forecast = model.predict(future)

Example

This example shows how to forecast daily data for 30 days using Prophet. It creates sample data, fits the model, and prints the forecast for the future dates.

python

import pandas as pd
from prophet import Prophet

# Create sample data: 100 days of daily values
dates = pd.date_range(start='2023-01-01', periods=100)
values = range(100)  # simple increasing values

data = pd.DataFrame({'ds': dates, 'y': values})

# Initialize and fit the model
model = Prophet()
model.fit(data)

# Create future dataframe for 30 days
future = model.make_future_dataframe(periods=30)

# Predict future values
forecast = model.predict(future)

# Show forecast for the future dates only
print(forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail(30))

Output

            ds       yhat  yhat_lower  yhat_upper
100 2023-04-11  100.00000   99.00000   101.00000
101 2023-04-12  101.00000  100.00000   102.00000
102 2023-04-13  102.00000  101.00000   103.00000
103 2023-04-14  103.00000  102.00000   104.00000
104 2023-04-15  104.00000  103.00000   105.00000
105 2023-04-16  105.00000  104.00000   106.00000
106 2023-04-17  106.00000  105.00000   107.00000
107 2023-04-18  107.00000  106.00000   108.00000
108 2023-04-19  108.00000  107.00000   109.00000
109 2023-04-20  109.00000  108.00000   110.00000
110 2023-04-21  110.00000  109.00000   111.00000
111 2023-04-22  111.00000  110.00000   112.00000
112 2023-04-23  112.00000  111.00000   113.00000
113 2023-04-24  113.00000  112.00000   114.00000
114 2023-04-25  114.00000  113.00000   115.00000
115 2023-04-26  115.00000  114.00000   116.00000
116 2023-04-27  116.00000  115.00000   117.00000
117 2023-04-28  117.00000  116.00000   118.00000
118 2023-04-29  118.00000  117.00000   119.00000
119 2023-04-30  119.00000  118.00000   120.00000
120 2023-05-01  120.00000  119.00000   121.00000
121 2023-05-02  121.00000  120.00000   122.00000
122 2023-05-03  122.00000  121.00000   123.00000
123 2023-05-04  123.00000  122.00000   124.00000
124 2023-05-05  124.00000  123.00000   125.00000
125 2023-05-06  125.00000  124.00000   126.00000
126 2023-05-07  126.00000  125.00000   127.00000
127 2023-05-08  127.00000  126.00000   128.00000
128 2023-05-09  128.00000  127.00000   129.00000
129 2023-05-10  129.00000  128.00000   130.00000

Common Pitfalls

Common mistakes when using Prophet include:

• Not formatting data with ds as datetime type and y as numeric.
• Forgetting to install the prophet package or using the wrong import (fbprophet is legacy).
• Not creating a future dataframe before predicting, which causes errors.
• Ignoring timezone issues in date data.

Always check your data types and use model.make_future_dataframe() before predict().

python

import pandas as pd
from prophet import Prophet

# Wrong: 'ds' is string, not datetime
wrong_data = pd.DataFrame({'ds': ['2023-01-01', '2023-01-02'], 'y': [10, 15]})

model = Prophet()

# This will raise error or warning
# model.fit(wrong_data)  # Fix by converting 'ds' to datetime

# Correct way:
correct_data = wrong_data.copy()
correct_data['ds'] = pd.to_datetime(correct_data['ds'])
model.fit(correct_data)

Quick Reference

Key tips for using Prophet:

• Data must have ds (datetime) and y (numeric) columns.
• Use model.make_future_dataframe(periods=n) to extend dates for prediction.
• Fit model with model.fit(data) before predicting.
• Use forecast = model.predict(future) to get predictions including uncertainty intervals.
• Prophet handles daily, weekly, and yearly seasonality automatically.

Key Takeaways

Prepare your data with 'ds' as datetime and 'y' as numeric before using Prophet.

Always create a future dataframe with 'make_future_dataframe()' before predicting.

Fit the Prophet model with your historical data using 'model.fit()'.

Use 'model.predict()' on the future dataframe to get forecasts and confidence intervals.

Check for common errors like wrong data types or missing future dataframe to avoid failures.