Prophet - Detecting Changes in Trend
Detecting Trend Changes
import pandas as pd
import numpy as np
from fbprophet import Prophet
Load Data
Load Hospitality dataset
df = pd.read_csv('HospitalityEmployees.csv')
# rename columns to Prophet specs
# Change date to time series
df.columns = ['ds', 'y']
df['ds'] = pd.to_datetime(df['ds'])
df.head()
| ds | y | |
|---|---|---|
| 0 | 1990-01-01 | 1064.5 |
| 1 | 1990-02-01 | 1074.5 |
| 2 | 1990-03-01 | 1090.0 |
| 3 | 1990-04-01 | 1097.4 |
| 4 | 1990-05-01 | 1108.7 |
df.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 348 entries, 0 to 347
Data columns (total 2 columns):
ds 348 non-null datetime64[ns]
y 348 non-null float64
dtypes: datetime64[ns](1), float64(1)
memory usage: 5.5 KB
df.describe()
| y | |
|---|---|
| count | 348.000000 |
| mean | 1452.506897 |
| std | 256.604914 |
| min | 1064.500000 |
| 25% | 1238.050000 |
| 50% | 1436.200000 |
| 75% | 1586.300000 |
| max | 2022.100000 |
# plot data
pd.plotting.register_matplotlib_converters()
df.plot(x='ds', y='y', figsize=(12,8))
Create Model
m = Prophet()
# fit entire df tomodel
m.fit(df)
# create future placeholder -- forecast 12 periods into future
future = m.make_future_dataframe(periods=12, freq='MS')
# forecast
forecast = m.predict(future)
INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this.
from fbprophet.plot import add_changepoints_to_plot
# take model and plot forecast
fig = m.plot(forecast)
a = add_changepoints_to_plot(fig.gca(), m, forecast) # gca = get current axis
