/ TIMESERIES

Time Series Analysis - fbprophet

fbprophet

import pandas as pd
from fbprophet import Prophet

df = pd.read_csv('Data/BeerWineLiquor.csv')

df.head()
date beer
0 1/1/1992 1509
1 2/1/1992 1541
2 3/1/1992 1597
3 4/1/1992 1675
4 5/1/1992 1822 
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 324 entries, 0 to 323
Data columns (total 2 columns):
 #   Column  Non-Null Count  Dtype 
---  ------  --------------  ----- 
 0   date    324 non-null    object
 1   beer    324 non-null    int64 
dtypes: int64(1), object(1)
memory usage: 5.2+ KB

df.columns = ['ds','y']

df.head()
ds y
0 1/1/1992 1509
1 2/1/1992 1541
2 3/1/1992 1597
3 4/1/1992 1675
4 5/1/1992 1822 
df['ds'] = pd.to_datetime(df['ds'])

df.head()
ds y
0 1992-01-01 1509
1 1992-02-01 1541
2 1992-03-01 1597
3 1992-04-01 1675
4 1992-05-01 1822 
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 324 entries, 0 to 323
Data columns (total 2 columns):
 #   Column  Non-Null Count  Dtype         
---  ------  --------------  -----         
 0   ds      324 non-null    datetime64[ns]
 1   y       324 non-null    int64         
dtypes: datetime64[ns](1), int64(1)
memory usage: 5.2 KB

m = Prophet()

m.fit(df)

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 0x195b664a048>

future = m.make_future_dataframe(periods=24, freq='MS')

future
ds
0 1992-01-01
1 1992-02-01
2 1992-03-01
3 1992-04-01
4 1992-05-01
... ...
343 2020-08-01
344 2020-09-01
345 2020-10-01
346 2020-11-01
347 2020-12-01

348 rows × 1 columns

df.tail()
ds y
319 2018-08-01 4898
320 2018-09-01 4598
321 2018-10-01 4737
322 2018-11-01 5130
323 2018-12-01 6370 
future.tail()
ds
343 2020-08-01
344 2020-09-01
345 2020-10-01
346 2020-11-01
347 2020-12-01 
print(len(df))
print(len(future))

324
348

forecast = m.predict(future)

forecast
ds trend yhat_lower yhat_upper trend_lower trend_upper additive_terms additive_terms_lower additive_terms_upper yearly yearly_lower yearly_upper multiplicative_terms multiplicative_terms_lower multiplicative_terms_upper yhat
0 1992-01-01 1778.389678 1162.489880 1477.941694 1778.389678 1778.389678 -461.901459 -461.901459 -461.901459 -461.901459 -461.901459 -461.901459 0.0 0.0 0.0 1316.488218
1 1992-02-01 1780.987184 1190.076650 1515.528138 1780.987184 1780.987184 -427.924955 -427.924955 -427.924955 -427.924955 -427.924955 -427.924955 0.0 0.0 0.0 1353.062229
2 1992-03-01 1783.417109 1437.570650 1768.918182 1783.417109 1783.417109 -179.677739 -179.677739 -179.677739 -179.677739 -179.677739 -179.677739 0.0 0.0 0.0 1603.739371
3 1992-04-01 1786.014616 1420.623061 1746.514284 1786.014616 1786.014616 -196.091481 -196.091481 -196.091481 -196.091481 -196.091481 -196.091481 0.0 0.0 0.0 1589.923135
4 1992-05-01 1788.528331 1683.898940 2005.729855 1788.528331 1788.528331 45.713009 45.713009 45.713009 45.713009 45.713009 45.713009 0.0 0.0 0.0 1834.241340
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
343 2020-08-01 5204.920380 5084.057969 5410.213286 5192.805710 5217.161608 40.765082 40.765082 40.765082 40.765082 40.765082 40.765082 0.0 0.0 0.0 5245.685462
344 2020-09-01 5221.075787 4928.327820 5254.378883 5207.949767 5234.472907 -132.510619 -132.510619 -132.510619 -132.510619 -132.510619 -132.510619 0.0 0.0 0.0 5088.565168
345 2020-10-01 5236.710051 5031.756676 5349.758431 5222.760534 5250.974553 -43.280888 -43.280888 -43.280888 -43.280888 -43.280888 -43.280888 0.0 0.0 0.0 5193.429163
346 2020-11-01 5252.865458 5193.204708 5505.269976 5237.580236 5268.257137 87.236362 87.236362 87.236362 87.236362 87.236362 87.236362 0.0 0.0 0.0 5340.101820
347 2020-12-01 5268.499722 6196.217158 6508.591160 5251.861675 5285.213636 1071.594904 1071.594904 1071.594904 1071.594904 1071.594904 1071.594904 0.0 0.0 0.0 6340.094626

348 rows × 16 columns

forecast.columns

Index(['ds', 'trend', 'yhat_lower', 'yhat_upper', 'trend_lower', 'trend_upper',
       'additive_terms', 'additive_terms_lower', 'additive_terms_upper',
       'yearly', 'yearly_lower', 'yearly_upper', 'multiplicative_terms',
       'multiplicative_terms_lower', 'multiplicative_terms_upper', 'yhat'],
      dtype='object')

forecast[['ds','yhat_lower', 'yhat_upper','yhat']].tail(12)
ds yhat_lower yhat_upper yhat
336 2020-01-01 4476.520302 4784.982116 4632.015644
337 2020-02-01 4514.459845 4838.813515 4682.147555
338 2020-03-01 4789.785892 5098.339918 4945.507893
339 2020-04-01 4780.225878 5107.184054 4945.249558
340 2020-05-01 5042.596291 5358.611803 5202.688312
341 2020-06-01 5041.145483 5343.992580 5183.565324
342 2020-07-01 5149.825503 5478.871858 5316.228164
343 2020-08-01 5084.057969 5410.213286 5245.685462
344 2020-09-01 4928.327820 5254.378883 5088.565168
345 2020-10-01 5031.756676 5349.758431 5193.429163
346 2020-11-01 5193.204708 5505.269976 5340.101820
347 2020-12-01 6196.217158 6508.591160 6340.094626 
import matplotlib.pyplot as plt
%matplotlib inline

m.plot(forecast)
plt.xlim('2014-01-01', '2021-01-01')

(array(16071.), array(18628.))

png

forecast.plot(x='ds',y='yhat', figsize=(8, 10))

<AxesSubplot:xlabel='ds'>

png

m.plot_components(forecast);

png

Prophet Evaluation

df = pd.read_csv('Data/Miles_Traveled.csv')

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 588 entries, 0 to 587
Data columns (total 2 columns):
 #   Column            Non-Null Count  Dtype  
---  ------            --------------  -----  
 0   DATE              588 non-null    object 
 1   TRFVOLUSM227NFWA  588 non-null    float64
dtypes: float64(1), object(1)
memory usage: 9.3+ KB

df.columns = ['ds', 'y']
df['ds'] = pd.to_datetime(df['ds'])

df.head()
ds y
0 1970-01-01 80173.0
1 1970-02-01 77442.0
2 1970-03-01 90223.0
3 1970-04-01 89956.0
4 1970-05-01 97972.0 
df.plot(x='ds',y='y')

<AxesSubplot:xlabel='ds'>

png

len(df)

588

train = df.iloc[:576]
test = df.iloc[576:]

m = Prophet()
m.fit(train)
future = m.make_future_dataframe(periods=12,freq='MS')
forecast = m.predict(future)

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.

forecast.tail()
ds trend yhat_lower yhat_upper trend_lower trend_upper additive_terms additive_terms_lower additive_terms_upper yearly yearly_lower yearly_upper multiplicative_terms multiplicative_terms_lower multiplicative_terms_upper yhat
583 2018-08-01 263219.378669 274005.563021 285854.055723 263165.973306 263286.645515 16451.125033 16451.125033 16451.125033 16451.125033 16451.125033 16451.125033 0.0 0.0 0.0 279670.503702
584 2018-09-01 263357.665883 255845.729306 267358.811392 263268.467813 263466.096392 -1666.545411 -1666.545411 -1666.545411 -1666.545411 -1666.545411 -1666.545411 0.0 0.0 0.0 261691.120472
585 2018-10-01 263491.492218 262880.627108 274558.498098 263368.767781 263640.637785 5310.587893 5310.587893 5310.587893 5310.587893 5310.587893 5310.587893 0.0 0.0 0.0 268802.080112
586 2018-11-01 263629.779432 249738.034670 261701.249744 263456.755896 263809.346746 -8203.436940 -8203.436940 -8203.436940 -8203.436940 -8203.436940 -8203.436940 0.0 0.0 0.0 255426.342492
587 2018-12-01 263763.605767 251277.958828 262955.321543 263569.324677 264002.170664 -6916.514689 -6916.514689 -6916.514689 -6916.514689 -6916.514689 -6916.514689 0.0 0.0 0.0 256847.091078 
ax = forecast.plot(x='ds',y='yhat', label='Predictions',legend=True, figsize=(12,8))
test.plot(x='ds',y='y',label='True Test Data', legend=True, ax=ax,
         xlim=('2018-01-01','2019-01-01'))

<AxesSubplot:xlabel='ds'>

png

from statsmodels.tools.eval_measures import rmse

predictions = forecast.iloc[-12:]['yhat']
predictions

576    243683.663737
577    235311.776917
578    262511.200772
579    262708.863987
580    272429.495856
581    272679.913346
582    279136.474721
583    279670.503702
584    261691.120472
585    268802.080112
586    255426.342492
587    256847.091078
Name: yhat, dtype: float64

test['y']

576    245695.0
577    226660.0
578    268480.0
579    272475.0
580    286164.0
581    280877.0
582    288145.0
583    286608.0
584    260595.0
585    282174.0
586    258590.0
587    268413.0
Name: y, dtype: float64

rmse(predictions, test['y'])

8750.25052083061

test.mean()

C:\Users\ilvna\.conda\envs\tf-2.3\lib\site-packages\ipykernel_launcher.py:1: FutureWarning:

DataFrame.mean and DataFrame.median with numeric_only=None will include datetime64 and datetime64tz columns in a future version.






y    268739.666667
dtype: float64

from fbprophet.diagnostics import cross_validation, performance_metrics
from fbprophet.plot import plot_cross_validation_metric

# INITIAL
initial = 5 * 365
initial = str(initial) + ' days'
# PERIOD
period = 5 * 365
period = str(period) + ' days'
# HORIZION
horizon = 365
horizon = str(horizon) + ' days'

df_cv = cross_validation(m,initial=initial, period=period, horizon = horizon)

INFO:fbprophet:Making 9 forecasts with cutoffs between 1976-12-11 00:00:00 and 2016-12-01 00:00:00

df_cv.head()
ds yhat yhat_lower yhat_upper y cutoff
0 1977-01-01 108479.087389 106886.766224 109878.212671 102445.0 1976-12-11
1 1977-02-01 102996.111525 101535.125858 104359.275043 102416.0 1976-12-11
2 1977-03-01 118973.318113 117492.329065 120504.309675 119960.0 1976-12-11
3 1977-04-01 120612.923709 119128.403891 122021.648614 121513.0 1976-12-11
4 1977-05-01 127883.031830 126331.087726 129351.363487 128884.0 1976-12-11 
len(df_cv)

108

performance_metrics(df_cv)
horizon mse rmse mae mape mdape coverage
0 52 days 2.437359e+07 4936.961190 4528.637861 0.027750 0.023384 0.4
1 53 days 2.187095e+07 4676.638408 4262.628303 0.024999 0.023384 0.4
2 54 days 1.842045e+07 4291.904691 3732.310696 0.020061 0.022021 0.5
3 55 days 2.351247e+07 4848.965429 4271.125183 0.023241 0.023384 0.4
4 57 days 2.115130e+07 4599.053997 3989.900860 0.021416 0.022021 0.5
... ... ... ... ... ... ... ...
94 360 days 1.862165e+07 4315.281433 3775.153749 0.019707 0.019425 0.5
95 361 days 1.777184e+07 4215.666214 3507.111932 0.018367 0.018879 0.5
96 362 days 3.210247e+07 5665.904399 4435.117794 0.022175 0.024755 0.4
97 364 days 3.022906e+07 5498.095592 4257.907114 0.021507 0.021415 0.5
98 365 days 5.578676e+07 7469.053614 5689.503458 0.026806 0.025276 0.4

99 rows × 7 columns

plot_cross_validation_metric(df_cv, metric='rmse');

png

Facebook prophet Trend

import pandas as pd
from fbprophet import Prophet
%matplotlib inline

df = pd.read_csv('Data/HospitalityEmployees.csv')

df.columns = ['ds', 'y']
df['ds'] = pd.to_datetime(df['ds'])

df.plot(x='ds', y='y', figsize=(12,10))

<AxesSubplot:xlabel='ds'>

png

m = Prophet()
m.fit(df)
future = m.make_future_dataframe(periods=12,freq='MS')
forecast = m.predict(future)

C:\Users\ilvna\.conda\envs\tf-2.3\lib\importlib\_bootstrap.py:219: RuntimeWarning:

numpy.ufunc size changed, may indicate binary incompatibility. Expected 192 from C header, got 216 from PyObject

C:\Users\ilvna\.conda\envs\tf-2.3\lib\importlib\_bootstrap.py:219: RuntimeWarning:

numpy.ufunc size changed, may indicate binary incompatibility. Expected 192 from C header, got 216 from PyObject

C:\Users\ilvna\.conda\envs\tf-2.3\lib\importlib\_bootstrap.py:219: RuntimeWarning:

numpy.ufunc size changed, may indicate binary incompatibility. Expected 192 from C header, got 216 from PyObject

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.

from fbprophet.plot import add_changepoints_to_plot
fig = m.plot(forecast)
a = add_changepoints_to_plot(fig.gca(),m, forecast)

png

Facebook Prophet Seasonality

df = pd.read_csv('Data/airline_passengers.csv')

df.head()
Month Thousands of Passengers
0 1949-01 112
1 1949-02 118
2 1949-03 132
3 1949-04 129
4 1949-05 121 
df.columns = ['ds', 'y']
df['ds'] = pd.to_datetime(df['ds'])

m = Prophet()
m.fit(df)
future = m.make_future_dataframe(50, freq='MS')
forecast = m.predict(future)
fig = m.plot(forecast)

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.

png

fig = m.plot_components(forecast);

png

from fbprophet.plot import add_changepoints_to_plot
fig = m.plot(forecast)
a = add_changepoints_to_plot(fig.gca(),m, forecast)

png

m = Prophet(seasonality_mode='multiplicative')
m.fit(df)
future = m.make_future_dataframe(50, freq='MS')
forecast = m.predict(future)
fig = m.plot(forecast)
b = add_changepoints_to_plot(fig.gca(), m, forecast)

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.

png