def fit_prophet(dtf_train, dtf_test, lst_exog=None, model=None, freq="D", conf=0.95, figsize=(15,10)):
## setup prophet
if model is None:
model = Prophet(growth="linear", changepoints=None, n_changepoints=25, seasonality_mode="multiplicative",
yearly_seasonality="auto", weekly_seasonality="auto", daily_seasonality="auto",
holidays=None, interval_width=conf)
if lst_exog != None:
for regressor in lst_exog:
model.add_regressor(regressor)
## train
model.fit(dtf_train)
## test
dtf_prophet = model.make_future_dataframe(periods=len(dtf_test)+10, freq=freq, include_history=True)
if model.growth == "logistic":
dtf_prophet["cap"] = dtf_train["cap"].unique()[0]
if lst_exog != None:
dtf_prophet = dtf_prophet.merge(dtf_train[["ds"]+lst_exog], how="left")
dtf_prophet.iloc[-len(dtf_test):][lst_exog] = dtf_test[lst_exog].values
dtf_prophet = model.predict(dtf_prophet)
dtf_train = dtf_train.merge(dtf_prophet[["ds","yhat"]], how="left").rename(
columns={'yhat':'model', 'y':'ts'}).set_index("ds")
dtf_test = dtf_test.merge(dtf_prophet[["ds","yhat","yhat_lower","yhat_upper"]], how="left").rename(
columns={'yhat':'forecast', 'y':'ts', 'yhat_lower':'lower', 'yhat_upper':'upper'}).set_index("ds")
## evaluate
dtf = dtf_train.append(dtf_test)
dtf = utils_evaluate_ts_model(dtf, conf=conf, figsize=figsize, title="Prophet")
return dtf, model
def test_cross_validation_extra_regressors(self):
df = self.__df.copy()
df['extra'] = range(df.shape[0])
df['is_conditional_week'] = np.arange(df.shape[0]) // 7 % 2
m = Prophet()
m.add_seasonality(name='monthly', period=30.5, fourier_order=5)
m.add_seasonality(name='conditional_weekly',
period=7,
fourier_order=3,
prior_scale=2.,
condition_name='is_conditional_week')
m.add_regressor('extra')
m.fit(df)
df_cv = diagnostics.cross_validation(m,
horizon='4 days',
period='4 days',
initial='135 days')
self.assertEqual(len(np.unique(df_cv['cutoff'])), 2)
period = pd.Timedelta('4 days')
dc = df_cv['cutoff'].diff()
dc = dc[dc > pd.Timedelta(0)].min()
self.assertTrue(dc >= period)
self.assertTrue((df_cv['cutoff'] < df_cv['ds']).all())
df_merged = pd.merge(df_cv, self.__df, 'left', on='ds')
self.assertAlmostEqual(
np.sum((df_merged['y_x'] - df_merged['y_y'])**2), 0.0)
def test_seasonality_modes(self):
# Model with holidays, seasonalities, and extra regressors
holidays = pd.DataFrame({
'ds': pd.to_datetime(['2016-12-25']),
'holiday': ['xmas'],
'lower_window': [-1],
'upper_window': [0],
})
m = Prophet(seasonality_mode='multiplicative', holidays=holidays)
m.add_seasonality('monthly',
period=30,
mode='additive',
fourier_order=3)
m.add_regressor('binary_feature', mode='additive')
m.add_regressor('numeric_feature')
# Construct seasonal features
df = DATA.copy()
df['binary_feature'] = [0] * 255 + [1] * 255
df['numeric_feature'] = range(510)
df = m.setup_dataframe(df, initialize_scales=True)
m.history = df.copy()
m.set_auto_seasonalities()
seasonal_features, prior_scales, component_cols, modes = (
m.make_all_seasonality_features(df))
self.assertEqual(sum(component_cols['additive_terms']), 7)
self.assertEqual(sum(component_cols['multiplicative_terms']), 29)
self.assertEqual(
set(modes['additive']),
{
'monthly', 'binary_feature', 'additive_terms',
'extra_regressors_additive'
},
)
self.assertEqual(
set(modes['multiplicative']),
{
'weekly',
'yearly',
'xmas',
'numeric_feature',
'multiplicative_terms',
'extra_regressors_multiplicative',
'holidays',
},
)
def prophet_fit_and_predict_with_exog_and_advance_vars(
y: [[float]],
k: int,
t: [float],
a: [[float]],
model_params: dict = None) -> Tuple[List, List, Any, Any]:
""" Simpler wrapper for testing - univariate w/ advance vars w/ supplied times and future times """
assert len(t) == len(y) + k
assert len(a) == len(y) + k
assert isinstance(y[0], List)
a_cols = ['a' + str(i) for i in range(len(a[0]))]
df = pd.DataFrame(columns=a_cols, data=a[:-k])
Y = transpose(y)
df['y'] = Y[0]
n_exog = len(y[0]) - 1
y_cols = ['y'] + ['y' + str(i) for i in range(1, len(y[0]))]
for i in range(1, n_exog + 1):
df['y' + str(i)] = Y[i][:len(y)]
dt = epoch_to_naive_datetime(t)
df['ds'] = dt[:len(y)]
kwargs_used = dict([(k, v) for k, v in PROPHET_MODEL.items()])
if model_params:
kwargs_used.update(model_params)
m = Prophet(**kwargs_used)
for a_col in a_cols:
m.add_regressor(name=a_col)
for y_col in y_cols[1:]:
m.add_regressor(name=y_col)
with no_stdout_stderr():
m.fit(df)
freq = infer_freq_from_epoch(t)
future = m.make_future_dataframe(periods=k, freq=freq)
future['ds'] = dt
full_a_data = transpose(a)
for a_col, a_vals in zip(a_cols, full_a_data):
future[a_col] = a_vals
for i in range(1, n_exog + 1): # Just bring forward
future['y' + str(i)] = Y[i] + [Y[i][-1]] * k
forecast = m.predict(future)
x = forecast['yhat'].values[-k:] # Use m.plot(forecast) to take a peak
x_std = [
u - l for u, l in zip(forecast['yhat_upper'].values,
forecast['yhat_lower'].values)
]
return x, x_std, forecast, m
def estimate(file_path):
try:
df = pd.read_csv(file_path)
except:
raise FileNotFoundError('파일을 찾을 수 없습니다')
date_range = pd.date_range(start=START_DATE, end=END_DATE, freq='1H')[:-1]
df['ds'] = df['dt'].map(str) + " " + df['dhour'].map(str) + ":00:00"
df['ds'] = pd.to_datetime(df['ds'])
df = df.set_index('ds')
df = pd.merge(date_range.to_frame(), df, left_index=True, right_index=True, how='left')
df['ds'] = df.index
df = df.rename(columns={'sales' : 'y'})
missing_fill_val = {'avg_prc' : df.avg_prc.median(), 'y' : 0}
df.fillna(missing_fill_val, inplace=True)
q3 = df['y'].quantile(q=0.75)
cap = q3 * 1.5
df['y'] = df['y'].apply(lambda x : cap if x >= cap else x)
df = df[['ds', 'y', 'avg_prc']]
scaler = MinMaxScaler()
scaled_value = scaler.fit_transform(df[['avg_prc', 'y']].values)
df[['avg_prc', 'y']] = scaled_value
df['floor'] = 0
df['cap'] = 1.2
train = df[:-PRED_DAYS]
test = df[-PRED_DAYS:]
m = Prophet(growth='logistic', holidays=holidays, holidays_prior_scale=1)
m.add_regressor('avg_prc')
m.fit(train)
future = m.make_future_dataframe(periods=PRED_DAYS, freq='H')
future = pd.merge(future, train, left_on='ds', right_on='ds', how='left')
future = future[['ds', 'floor', 'cap', 'avg_prc']]
future_fill_missing = {'avg_prc' : df.iloc[len(df)-1]['avg_prc'], 'cap' : 1.2, 'floor' : 0}
future.fillna(future_fill_missing, inplace=True)
forecast = m.predict(future)
pred = forecast[['ds', 'yhat']][-PRED_DAYS:]
pred['yhat'] = np.where(pred['yhat'] < 0, 0, pred['yhat'])
rmse = math.sqrt(mean_squared_error(test['y'], pred['yhat']))
r2score = r2_score(test['y'], pred['yhat'])
return rmse, r2score
def test_regressor_coefficients(self):
m = Prophet()
N = DATA.shape[0]
df = DATA.copy()
np.random.seed(123)
df['regr1'] = np.random.normal(size=N)
df['regr2'] = np.random.normal(size=N)
m.add_regressor('regr1', mode='additive')
m.add_regressor('regr2', mode='multiplicative')
m.fit(df)
coefs = regressor_coefficients(m)
self.assertTrue(coefs.shape == (2, 6))
# No MCMC sampling, so lower and upper should be the same as mean
self.assertTrue(
np.array_equal(coefs['coef_lower'].values, coefs['coef'].values))
self.assertTrue(
np.array_equal(coefs['coef_upper'].values, coefs['coef'].values))
def get_model_forecast(info):
ds = ast.literal_eval(info['train_ds'])
y = ast.literal_eval(info['train_y'])
avg_prc = ast.literal_eval(info['train_avg_prc'])
test_y = ast.literal_eval(info['test_y'])
test_avg_prc = ast.literal_eval(info['test_avg_prc'])
dic = {'ds': ds, 'y': y, 'avg_prc': avg_prc}
data = pd.DataFrame(dic)
holidays = pd.read_json(info['holidays'])
## feature engineering
if data['avg_prc'].max() > 0:
data['avg_prc'] = data['avg_prc'] / data['avg_prc'].max() * 100
else:
data['avg_prc'] = data['avg_prc'] / (data['avg_prc'].max() + 1) * 100
data['cap'] = 100.0
data['floor'] = 0.0
## run prophet
model = Prophet(growth='logistic', holidays=holidays)
model.add_country_holidays(country_name='KR')
model.add_seasonality(name='monthly', period=30.5, fourier_order=5)
if data['avg_prc'].isna().sum() == 0:
model.add_regressor('avg_prc')
model.fit(data)
## get estimation
future = model.make_future_dataframe(periods=PRED_DAYS)
if data['avg_prc'].isna().sum() == 0:
future['avg_prc'] = pd.concat(
[pd.Series(avg_prc), pd.Series(test_avg_prc)], ignore_index=True)
future['cap'] = 100
future['floor'] = 0.0
forecast = model.predict(future)
return model, forecast
def get_model_forecast_pred(self):
train, test = self.data[:-self.PRED_DAYS], self.data[-self.PRED_DAYS:]
model = Prophet(growth='logistic',
holidays=self.holidays,
holidays_prior_scale=self.holiday_weight,
seasonality_prior_scale=self.seasonality_weight,
changepoint_prior_scale=self.changepoint_weight,
changepoint_range=self.changepoint_range,
changepoints=self.changepoints if self.changepoints else None,
)
if self.ADD_COUNTRY_HOLIDAY:
model.add_country_holidays(country_name='KR')
if self.ADD_MONTHLY_SEASONALITY:
model.add_seasonality(name='montly_seasonality', period=30.5, fourier_order=5)
if self.PRC:
model.add_regressor('avg_prc', prior_scale=self.price_weight, standardize=False)
model.fit(train)
future = model.make_future_dataframe(periods=self.PRED_DAYS)
future = pd.merge(future, train, left_on='ds', right_on='ds', how='left')
future = future[['ds', 'floor', 'cap', 'avg_prc']]
future['avg_prc'] = self.data.avg_prc.values
future_fill_missing = {'cap' : 100, 'floor' : 0.0}
future.fillna(future_fill_missing, inplace=True)
else:
model.fit(train)
future = model.make_future_dataframe(periods=self.PRED_DAYS)
future['cap'] = 100
future['floor'] = 0.0
forecast = model.predict(future)
pred = forecast[['ds', 'yhat']][-self.PRED_DAYS:]
pred['yhat'] = np.where(pred['yhat'] < 0, 0, pred['yhat'])
return model, forecast, pred
def test_added_regressors(self):
m = Prophet()
m.add_regressor('binary_feature', prior_scale=0.2)
m.add_regressor('numeric_feature', prior_scale=0.5)
m.add_regressor('numeric_feature2',
prior_scale=0.5,
mode='multiplicative')
m.add_regressor('binary_feature2', standardize=True)
df = DATA.copy()
df['binary_feature'] = ['0'] * 255 + ['1'] * 255
df['numeric_feature'] = range(510)
df['numeric_feature2'] = range(510)
with self.assertRaises(ValueError):
# Require all regressors in df
m.fit(df)
df['binary_feature2'] = [1] * 100 + [0] * 410
m.fit(df)
# Check that standardizations are correctly set
self.assertEqual(
m.extra_regressors['binary_feature'],
{
'prior_scale': 0.2,
'mu': 0,
'std': 1,
'standardize': 'auto',
'mode': 'additive',
},
)
self.assertEqual(m.extra_regressors['numeric_feature']['prior_scale'],
0.5)
self.assertEqual(m.extra_regressors['numeric_feature']['mu'], 254.5)
self.assertAlmostEqual(m.extra_regressors['numeric_feature']['std'],
147.368585,
places=5)
self.assertEqual(m.extra_regressors['numeric_feature2']['mode'],
'multiplicative')
self.assertEqual(m.extra_regressors['binary_feature2']['prior_scale'],
10.)
self.assertAlmostEqual(m.extra_regressors['binary_feature2']['mu'],
0.1960784,
places=5)
self.assertAlmostEqual(m.extra_regressors['binary_feature2']['std'],
0.3974183,
places=5)
# Check that standardization is done correctly
df2 = m.setup_dataframe(df.copy())
self.assertEqual(df2['binary_feature'][0], 0)
self.assertAlmostEqual(df2['numeric_feature'][0], -1.726962, places=4)
self.assertAlmostEqual(df2['binary_feature2'][0], 2.022859, places=4)
# Check that feature matrix and prior scales are correctly constructed
seasonal_features, prior_scales, component_cols, modes = (
m.make_all_seasonality_features(df2))
self.assertEqual(seasonal_features.shape[1], 30)
names = ['binary_feature', 'numeric_feature', 'binary_feature2']
true_priors = [0.2, 0.5, 10.]
for i, name in enumerate(names):
self.assertIn(name, seasonal_features)
self.assertEqual(sum(component_cols[name]), 1)
self.assertEqual(
sum(np.array(prior_scales) * component_cols[name]),
true_priors[i],
)
# Check that forecast components are reasonable
future = pd.DataFrame({
'ds': ['2014-06-01'],
'binary_feature': [0],
'numeric_feature': [10],
'numeric_feature2': [10],
})
with self.assertRaises(ValueError):
m.predict(future)
future['binary_feature2'] = 0
fcst = m.predict(future)
self.assertEqual(fcst.shape[1], 37)
self.assertEqual(fcst['binary_feature'][0], 0)
self.assertAlmostEqual(
fcst['extra_regressors_additive'][0],
fcst['numeric_feature'][0] + fcst['binary_feature2'][0],
)
self.assertAlmostEqual(
fcst['extra_regressors_multiplicative'][0],
fcst['numeric_feature2'][0],
)
self.assertAlmostEqual(
fcst['additive_terms'][0],
fcst['yearly'][0] + fcst['weekly'][0] +
fcst['extra_regressors_additive'][0],
)
self.assertAlmostEqual(
fcst['multiplicative_terms'][0],
fcst['extra_regressors_multiplicative'][0],
)
self.assertAlmostEqual(
fcst['yhat'][0],
fcst['trend'][0] * (1 + fcst['multiplicative_terms'][0]) +
fcst['additive_terms'][0],
)
# Check works if constant extra regressor at 0
df['constant_feature'] = 0
m = Prophet()
m.add_regressor('constant_feature')
m.fit(df)
self.assertEqual(m.extra_regressors['constant_feature']['std'], 1)
def prophet_iskater_factory(y: [[float]],
k: int,
a: List = None,
t: List = None,
e=None,
freq: str = None,
n_max=1000,
recursive: bool = False,
model_params: dict = None,
return_forecast=True):
"""
:param y: A list of observations, each a vector.
:param k: Number of steps ahead to predict
:param a: Known in advance observations - should be k more of these than y's
:param t: Epoch times of observations y. If len(t)=len(y)+k the last k are interpreted as future times.
:param freq: 'D', '5T' etc, see https://github.com/pandas-dev/pandas/blob/master/pandas/tseries/frequencies.py
:param n_max: Maximum number of observations to use, should you wish to prevent prophet from slowing down
:param recursive If True, exogenous variables y[1], y[2],... will be predicted forward in time
(obviously this adds to computation time)
:returns: x k-vector of predictions
x_std k-vector of standard deviations
forecast full forecast dataframe, familiar to users of fbprophet
"""
if a:
assert len(a) == len(y) + k
if isinstance(y[0], float):
y = [wrap(yj) for yj in y]
# Conversion of epoch times to UTC datetime
# User must supply times, len(y) or len(y)+k, or a valid frequency str
if t is None:
if freq is None or not freq:
freq = PROPHET_META['freq'] # Just assume away ...
else:
assert is_valid_freq(
freq), 'Freq ' + str(freq) + ' is not a valid frequency'
dt = pd.date_range(start=EPOCH, periods=len(y), freq=freq) # UTC
else:
freq = infer_freq_from_epoch(t)
dt = epoch_to_naive_datetime(t)
if len(dt) == len(y) + k:
ta = dt
dt = dt[:len(y)]
else:
assert len(dt) == len(
y), 'Time vector t should be len(y) or len(y)+k'
ta = None
# Truncate history so that prophet doesn't take forever to fit
y_shorter = y[-n_max:]
a_shorter = a[-(n_max + k):] if a is not None else [] # may be empty
dt_shorter = dt[-n_max:]
# Massage data into Prophet friendly dataframe with columns y, y1, ..., yk, a0,...aj
y_cols = [
'y' + str(i) if i > 0 else 'y' for i in range(len(y_shorter[-1]))
]
if a:
a_cols = ['a' + str(i) for i in range(len(a_shorter[-1]))]
data = [
list(yi) + list(ai)
for yi, ai in zip(y_shorter, a_shorter[:-k])
]
df = pd.DataFrame(columns=y_cols + a_cols, data=data)
else:
data = [list(yi) for yi in y_shorter]
df = pd.DataFrame(columns=y_cols, data=data)
df['ds'] = dt_shorter
# Instantiate Prophet model, ensure defaults are what we think they are
kwargs_used = dict([(k, v) for k, v in PROPHET_MODEL.items()])
if model_params:
kwargs_used.update(model_params)
m = Prophet(**kwargs_used)
# Add regressors
for y_col in y_cols[1:]:
m.add_regressor(name=y_col)
if a:
for a_col in a_cols:
m.add_regressor(name=a_col)
# Fit the model every invocation ... there isn't any other way
with no_stdout_stderr():
m.fit(df)
# Make future dataframe, adding known-in-advance variables
future = m.make_future_dataframe(periods=k, freq=freq)
if a:
for j, a_col in enumerate(a_cols):
future[a_col] = [ai[j] for ai in a_shorter] # Known in advance
if ta is not None:
future['ds'] = ta # override with user supplied future times
# Next, we wish to add contemporaneously observed variables
#
# This is somewhat problematic, for how should we bring exogenously observed variables forward?
# The simplest answer is, don't use them - only supply 1-vector y observations
# prophet implicitly assumes all exogenous are known, which is a pretty big shortcoming.
#
# However, if we are trying to support y[1:], ...
# - It seems consistent to use prophet to predict these forward,
# - It also seems likely that this will lead to over-fitting.
# I'm open to ideas here. Perhaps perform some hackery could effect attenuation of the coefficients
# assigned to y[1],... such as jiggling past observations. For now we use prophet on each
# one individually, feeding them the known in advance 'a' variables.
n_exog = len(y[0]) - 1
if n_exog > 0:
for j, y_col in enumerate(y_cols):
if j > 0:
yj = [yi[j] for yi in y_shorter]
if recursive:
yj_hat, yj_hat_std, yj_forecast, yj_m = prophet_iskater_factory(
y=yj,
k=k,
a=a_shorter,
freq=freq,
n_max=n_max,
recursive=False)
else:
yj_hat = [yj[-1]] * k
future[y_col] = yj + list(yj_hat)
# Call the prediction function
forecast = m.predict(future)
x = list(forecast['yhat'].values[-k:]
) # Use m.plot(forecast) to take a peak
# Interpret confidence level difference as scale to be returned. TODO: set alpha properly so this really is 1-std
x_std = list([
u - l for u, l in zip(forecast['yhat_upper'].values[-k:],
forecast['yhat_lower'].values[-k:])
])
if return_forecast:
return x, x_std, forecast, m
else:
return x, x_std
def test_copy(self):
df = DATA_all.copy()
df['cap'] = 200.
df['binary_feature'] = [0] * 255 + [1] * 255
# These values are created except for its default values
holiday = pd.DataFrame({
'ds': pd.to_datetime(['2016-12-25']),
'holiday': ['x']
})
products = itertools.product(
['linear', 'logistic'], # growth
[None, pd.to_datetime(['2016-12-25'])], # changepoints
[3], # n_changepoints
[0.9], # changepoint_range
[True, False], # yearly_seasonality
[True, False], # weekly_seasonality
[True, False], # daily_seasonality
[None, holiday], # holidays
['additive', 'multiplicative'], # seasonality_mode
[1.1], # seasonality_prior_scale
[1.1], # holidays_prior_scale
[0.1], # changepoint_prior_scale
[100], # mcmc_samples
[0.9], # interval_width
[200] # uncertainty_samples
)
# Values should be copied correctly
for product in products:
m1 = Prophet(*product)
m1.country_holidays = 'US'
m1.history = m1.setup_dataframe(df.copy(), initialize_scales=True)
m1.set_auto_seasonalities()
m2 = diagnostics.prophet_copy(m1)
self.assertEqual(m1.growth, m2.growth)
self.assertEqual(m1.n_changepoints, m2.n_changepoints)
self.assertEqual(m1.changepoint_range, m2.changepoint_range)
if m1.changepoints is None:
self.assertEqual(m1.changepoints, m2.changepoints)
else:
self.assertTrue(m1.changepoints.equals(m2.changepoints))
self.assertEqual(False, m2.yearly_seasonality)
self.assertEqual(False, m2.weekly_seasonality)
self.assertEqual(False, m2.daily_seasonality)
self.assertEqual(m1.yearly_seasonality, 'yearly'
in m2.seasonalities)
self.assertEqual(m1.weekly_seasonality, 'weekly'
in m2.seasonalities)
self.assertEqual(m1.daily_seasonality, 'daily' in m2.seasonalities)
if m1.holidays is None:
self.assertEqual(m1.holidays, m2.holidays)
else:
self.assertTrue((m1.holidays == m2.holidays).values.all())
self.assertEqual(m1.country_holidays, m2.country_holidays)
self.assertEqual(m1.seasonality_mode, m2.seasonality_mode)
self.assertEqual(m1.seasonality_prior_scale,
m2.seasonality_prior_scale)
self.assertEqual(m1.changepoint_prior_scale,
m2.changepoint_prior_scale)
self.assertEqual(m1.holidays_prior_scale, m2.holidays_prior_scale)
self.assertEqual(m1.mcmc_samples, m2.mcmc_samples)
self.assertEqual(m1.interval_width, m2.interval_width)
self.assertEqual(m1.uncertainty_samples, m2.uncertainty_samples)
# Check for cutoff and custom seasonality and extra regressors
changepoints = pd.date_range('2012-06-15', '2012-09-15')
cutoff = pd.Timestamp('2012-07-25')
m1 = Prophet(changepoints=changepoints)
m1.add_seasonality('custom', 10, 5)
m1.add_regressor('binary_feature')
m1.fit(df)
m2 = diagnostics.prophet_copy(m1, cutoff=cutoff)
changepoints = changepoints[changepoints < cutoff]
self.assertTrue((changepoints == m2.changepoints).all())
self.assertTrue('custom' in m2.seasonalities)
self.assertTrue('binary_feature' in m2.extra_regressors)
class FBProphetModel(PredictionModel):
"""Facebook's Prophet prediction model."""
def __init__(self, params: dict, transformation: str = "none"):
super().__init__(params, name="FBProphet", transformation=transformation)
# Stuff needed to make Prophet shut up during training.
self.suppress_stdout_stderr = suppress_stdout_stderr
self.fbmodel = Prophet()
try:
self.fbprophet_parameters = params["model_parameters"]["fbprophet_parameters"]
except KeyError:
self.fbprophet_parameters = None
def train(self, input_data: DataFrame, extra_regressors: DataFrame = None):
"""Overrides PredictionModel.train()"""
if self.fbprophet_parameters is not None:
try:
timeseries_name = input_data.columns[0]
date_format = self.fbprophet_parameters["holidays_dataframes"]["date_format"]
holidays = pd.read_csv(self.fbprophet_parameters["holidays_dataframes"][timeseries_name])
holidays.loc[:, "ds"].apply(lambda x: pd.to_datetime(x, format=date_format))
self.fbmodel = Prophet(holidays=holidays)
log.debug(f"Using a dataframe for holidays...")
except KeyError:
self.fbmodel = Prophet()
try:
holiday_country = self.fbprophet_parameters["holiday_country"]
self.fbmodel.add_country_holidays(country_name=holiday_country)
log.debug(f"Set {holiday_country} as country for holiday calendar...")
except KeyError:
pass
else:
self.fbmodel = Prophet()
if extra_regressors is not None:
# We could apply self.transformation also on the extra regressors.
# From tests, it looks like it doesn't change much/it worsens the forecasts.
input_data = input_data.join(extra_regressors)
input_data.reset_index(inplace=True)
column_indices = [0, 1]
new_names = ['ds', 'y']
old_names = input_data.columns[column_indices]
input_data.rename(columns=dict(zip(old_names, new_names)), inplace=True)
[self.fbmodel.add_regressor(col) for col in extra_regressors.columns]
else:
input_data.reset_index(inplace=True)
input_data.columns = ['ds', 'y']
with self.suppress_stdout_stderr():
self.fbmodel.fit(input_data)
#######################
# param_grid = {
# 'changepoint_prior_scale': [0.001, 0.01, 0.1, 0.5],
# 'seasonality_prior_scale': [0.01, 0.1, 1.0, 10.0],
# }
# param_grid = {
# 'changepoint_prior_scale': [0.001, 0.01],
# 'seasonality_prior_scale': [0.01, 0.1],
# }
#
# if extra_regressors is not None:
# input_data = input_data.join(extra_regressors)
# input_data.reset_index(inplace=True)
# column_indices = [0, 1]
# new_names = ['ds', 'y']
# old_names = input_data.columns[column_indices]
# input_data.rename(columns=dict(zip(old_names, new_names)), inplace=True)
#
# else:
# input_data.reset_index(inplace=True)
# input_data.columns = ['ds', 'y']
#
# # Generate all combinations of parameters
# all_params = [dict(zip(param_grid.keys(), v)) for v in itertools.product(*param_grid.values())]
# rmses = [] # Store the RMSEs for each params here
#
# # Use cross validation to evaluate all parameters
# for params in all_params:
# m = Prophet(**params)
# [m.add_regressor(col) for col in extra_regressors.columns] if extra_regressors is not None else None
# with self.suppress_stdout_stderr():
# m.fit(input_data) # Fit model with given params
# df_cv = cross_validation(m, horizon=self.prediction_lags, parallel="processes")
# df_p = performance_metrics(df_cv, rolling_window=1)
# rmses.append(df_p['rmse'].values[0])
#
# # Find the best parameters
# tuning_results = pd.DataFrame(all_params)
# tuning_results['rmse'] = rmses
#
# best_params = all_params[np.argmin(rmses)]
# print(best_params)
#
# self.fbmodel = Prophet(**best_params)
# [self.fbmodel.add_regressor(col) for col in extra_regressors.columns] if extra_regressors is not None else None
# with self.suppress_stdout_stderr():
# self.fbmodel.fit(input_data)
def predict(self, future_dataframe: DataFrame, extra_regressors: DataFrame = None) -> DataFrame:
"""Overrides PredictionModel.predict()"""
future = future_dataframe.reset_index()
future.rename(columns={'index': 'ds'}, inplace=True)
if extra_regressors is not None:
future.set_index('ds', inplace=True)
future = future.join(extra_regressors.copy())
future.reset_index(inplace=True)
forecast = self.fbmodel.predict(future)
forecast.set_index('ds', inplace=True)
return forecast
import pandas as pd
from prophet import Prophet
# float_precision='high' required for pd.read_csv to match precision of Rover.read_csv
df = pd.read_csv('examples/example_wp_log_peyton_manning.csv',
float_precision='high')
def nfl_sunday(ds):
date = pd.to_datetime(ds)
if date.weekday() == 6 and (date.month > 8 or date.month < 2):
return 1
else:
return 0
df['nfl_sunday'] = df['ds'].apply(nfl_sunday)
m = Prophet()
m.add_regressor('nfl_sunday')
m.fit(df)
future = m.make_future_dataframe(periods=365)
future['nfl_sunday'] = future['ds'].apply(nfl_sunday)
forecast = m.predict(future)
m.plot(forecast).savefig('/tmp/py_regressors.png')
m.plot_components(forecast).savefig('/tmp/py_regressors2.png')
