saymrwulf/prophet
1
0
Fork 0
mirror of https://github.com/saymrwulf/prophet.git synced 2026-07-13 18:08:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge in bugfixes from master (#349)

Browse source 
* Update memory requirement description per #326

* Fix R warning with extra regressor; disallow constant extra regressors.

* Fix unit test broken in new pandas

* Fix diagnostics unit tests for new pandas

* Fix copy with extra seasonalities / regressors Py

* Fix copy with extra seasonalities / regressors R

* Fix weekly_start and yearly_start in R plot_components

* Fix plotting in pandas 0.21 by using pydatetime instead of numpy
This commit is contained in:
Ben Letham 2017-11-08 10:09:08 -08:00 committed by GitHub
parent 1e30462779
commit e78f583f90
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
9 changed files with 139 additions and 70 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

36
R/R/prophet.R
View file

@ -392,9 +392,13 @@ initialize_scales_fn <- function(m, initialize_scales, df) {
m$start <- min(df$ds)
m$t.scale <- time_diff(max(df$ds), m$start, "secs")
for (name in names(m$extra_regressors)) {
n.vals <- length(unique(df[[name]]))
if (n.vals < 2) {
stop('Regressor ', name, ' is constant.')
}
standardize <- m$extra_regressors[[name]]$standardize
if (standardize == 'auto') {
if (all(sort(unique(df[[name]])) == c(0, 1))) {
if (n.vals == 2 && all(sort(unique(df[[name]])) == c(0, 1))) {
# Don't standardize binary variables
standardize <- FALSE
} else {
@ -404,9 +408,6 @@ initialize_scales_fn <- function(m, initialize_scales, df) {
if (standardize) {
mu <- mean(df[[name]])
std <- stats::sd(df[[name]])
if (std == 0) {
std <- mu
}
m$extra_regressors[[name]]$mu <- mu
m$extra_regressors[[name]]$std <- std
}
@ -1586,7 +1587,8 @@ seasonality_plot_df <- function(m, ds) {
#' @keywords internal
plot_weekly <- function(m, uncertainty = TRUE, weekly_start = 0) {
# Compute weekly seasonality for a Sun-Sat sequence of dates.
days <- seq(set_date('2017-01-01'), by='d', length.out=7) + weekly_start
days <- seq(set_date('2017-01-01'), by='d', length.out=7) + as.difftime(
weekly_start, units = "days")
df.w <- seasonality_plot_df(m, days)
seas <- predict_seasonal_components(m, df.w)
seas$dow <- factor(weekdays(df.w$ds), levels=weekdays(df.w$ds))
@ -1619,7 +1621,8 @@ plot_weekly <- function(m, uncertainty = TRUE, weekly_start = 0) {
#' @keywords internal
plot_yearly <- function(m, uncertainty = TRUE, yearly_start = 0) {
# Compute yearly seasonality for a Jan 1 - Dec 31 sequence of dates.
days <- seq(set_date('2017-01-01'), by='d', length.out=365) + yearly_start
days <- seq(set_date('2017-01-01'), by='d', length.out=365) + as.difftime(
yearly_start, units = "days")
df.y <- seasonality_plot_df(m, days)
seas <- predict_seasonal_components(m, df.y)
seas$ds <- df.y$ds
@ -1695,6 +1698,10 @@ plot_seasonality <- function(m, name, uncertainty = TRUE) {
#'
#' @keywords internal
prophet_copy <- function(m, cutoff = NULL) {
if (is.null(m$history)) {
stop("This is for copying a fitted Prophet object.")
}
if (m$specified.changepoints) {
changepoints <- m$changepoints
if (!is.null(cutoff)) {
@ -1704,13 +1711,15 @@ prophet_copy <- function(m, cutoff = NULL) {
} else {
changepoints <- NULL
}
return(prophet(
# Auto seasonalities are set to FALSE because they are already set in
# m$seasonalities.
m2 <- prophet(
growth = m$growth,
changepoints = changepoints,
n.changepoints = m$n.changepoints,
yearly.seasonality = m$yearly.seasonality,
weekly.seasonality = m$weekly.seasonality,
daily.seasonality = m$daily.seasonality,
yearly.seasonality = FALSE,
weekly.seasonality = FALSE,
daily.seasonality = FALSE,
holidays = m$holidays,
seasonality.prior.scale = m$seasonality.prior.scale,
changepoint.prior.scale = m$changepoint.prior.scale,
@ -1718,8 +1727,11 @@ prophet_copy <- function(m, cutoff = NULL) {
mcmc.samples = m$mcmc.samples,
interval.width = m$interval.width,
uncertainty.samples = m$uncertainty.samples,
fit = FALSE,
))
fit = FALSE
)
m2$extra_regressors <- m$extra_regressors
m2$seasonalities <- m$seasonalities
return(m2)
}
# fb-block 3

61
R/tests/testthat/test_prophet.R
View file

@ -511,24 +511,24 @@ test_that("added_regressors", {
expect_equal(fcst$seasonal[1],
fcst$seasonalities[1] + fcst$extra_regressors[1])
expect_equal(fcst$yhat[1], fcst$trend[1] + fcst$seasonal[1])
# Check fails if constant extra regressor
df$constant_feature <- 5
m <- prophet()
m <- add_regressor(m, 'constant_feature')
expect_error(fit.prophet(m, df))
})
test_that("copy", {
skip_if_not(Sys.getenv('R_ARCH') != '/i386')
df <- DATA
df$cap <- 200.
df$binary_feature <- c(rep(0, 255), rep(1, 255))
inputs <- list(
growth = c('linear', 'logistic'),
changepoints = c(NULL, c('2016-12-25')),
n.changepoints = c(3),
yearly.seasonality = c(TRUE, FALSE),
weekly.seasonality = c(TRUE, FALSE),
daily.seasonality = c(TRUE, FALSE),
holidays = c(NULL, 'insert_dataframe'),
seasonality.prior.scale = c(1.1),
holidays.prior.scale = c(1.1),
changepoints.prior.scale = c(0.1),
mcmc.samples = c(100),
interval.width = c(0.9),
uncertainty.samples = c(200)
holidays = c('null', 'insert_dataframe')
)
products <- expand.grid(inputs)
for (i in 1:length(products)) {
@ -538,32 +538,51 @@ test_that("copy", {
holidays <- NULL
}
m1 <- prophet(
growth = products$growth[i],
changepoints = products$changepoints[i],
n.changepoints = products$n.changepoints[i],
growth = as.character(products$growth[i]),
changepoints = NULL,
n.changepoints = 3,
yearly.seasonality = products$yearly.seasonality[i],
weekly.seasonality = products$weekly.seasonality[i],
daily.seasonality = products$daily.seasonality[i],
holidays = holidays,
seasonality.prior.scale = products$seasonality.prior.scale[i],
holidays.prior.scale = products$holidays.prior.scale[i],
changepoints.prior.scale = products$changepoints.prior.scale[i],
mcmc.samples = products$mcmc.samples[i],
interval.width = products$interval.width[i],
uncertainty.samples = products$uncertainty.samples[i],
seasonality.prior.scale = 1.1,
holidays.prior.scale = 1.1,
changepoints.prior.scale = 0.1,
mcmc.samples = 100,
interval.width = 0.9,
uncertainty.samples = 200,
fit = FALSE
)
out <- prophet:::setup_dataframe(m1, df, initialize_scales = TRUE)
m1 <- out$m
m1$history <- out$df
m1 <- prophet:::set_auto_seasonalities(m1)
m2 <- prophet:::prophet_copy(m1)
# Values should be copied correctly
for (arg in names(inputs)) {
args <- c('growth', 'changepoints', 'n.changepoints', 'holidays',
'seasonality.prior.scale', 'holidays.prior.scale',
'changepoints.prior.scale', 'mcmc.samples', 'interval.width',
'uncertainty.samples')
for (arg in args) {
expect_equal(m1[[arg]], m2[[arg]])
}
expect_equal(FALSE, m2$yearly.seasonality)
expect_equal(FALSE, m2$weekly.seasonality)
expect_equal(FALSE, m2$daily.seasonality)
expect_equal(m1$yearly.seasonality, 'yearly' %in% names(m2$seasonalities))
expect_equal(m1$weekly.seasonality, 'weekly' %in% names(m2$seasonalities))
expect_equal(m1$daily.seasonality, 'daily' %in% names(m2$seasonalities))
}
# Check for cutoff
# Check for cutoff and custom seasonality and extra regressors
changepoints <- seq.Date(as.Date('2012-06-15'), as.Date('2012-09-15'), by='d')
cutoff <- as.Date('2012-07-25')
m1 <- prophet(DATA, changepoints = changepoints)
m1 <- prophet(changepoints = changepoints)
m1 <- add_seasonality(m1, 'custom', 10, 5)
m1 <- add_regressor(m1, 'binary_feature')
m1 <- fit.prophet(m1, df)
m2 <- prophet:::prophet_copy(m1, cutoff)
changepoints <- changepoints[changepoints <= cutoff]
expect_equal(prophet:::set_date(changepoints), m2$changepoints)
expect_true('custom' %in% names(m2$seasonalities))
expect_true('binary_feature' %in% names(m2$extra_regressors))
})

2
README.md
View file

@ -52,7 +52,7 @@ On Windows, PyStan requires a compiler so you'll need to [follow the instruction
### Linux
Make sure compilers (gcc, g++) and Python development tools (python-dev) are installed. If you are using a VM, be aware that you will need at least 2GB of memory to run PyStan.
Make sure compilers (gcc, g++) and Python development tools (python-dev) are installed. If you are using a VM, be aware that you will need at least 4GB of memory to install fbprophet, and at least 2GB of memory to use fbprophet.
### Anaconda

2
docs/_docs/installation.md
View file

@ -43,7 +43,7 @@ On Windows, PyStan requires a compiler so you'll need to [follow the instruction
### Linux
Make sure compilers (gcc, g++) and Python development tools (python-dev) are installed. If you are using a VM, be aware that you will need at least 2GB of memory to run PyStan.
Make sure compilers (gcc, g++) and Python development tools (python-dev) are installed. If you are using a VM, be aware that you will need at least 4GB of memory to install fbprophet, and at least 2GB of memory to use fbprophet.
### Anaconda

2
docs/_docs/seasonality_and_holiday_effects.md
View file

@ -362,6 +362,6 @@ m.plot_components(forecast);
![png](/prophet/static/seasonality_and_holiday_effects_files/seasonality_and_holiday_effects_26_0.png)
NFL Sundays could also have been handled using the "holidays" interface described above, by creating a list of past and future NFL Sundays. The `add_regressor` function provides a more general interface for defining extra linear regressors, and in particular does not require that the regressor be a binary indicator. Another time series could be used as a regressor, although its future values would have to be known.
NFL Sundays could also have been handled using the "holidays" interface described above, by creating a list of past and future NFL Sundays. The `add_regressor` function provides a more general interface for defining extra linear regressors, and in particular does not require that the regressor be a binary indicator. Another time series could be used as a regressor, although its future values would have to be known. The regressor cannot be constant in the training data; fitting will exit with an error if it is.
The `add_regressor` function has optional arguments for specifying the prior scale (holiday prior scale is used by default) and whether or not the regressor is standardized - see the docstring with `help(Prophet.add_regressor)` in Python and `?add_regressor` in R.

2
notebooks/seasonality_and_holiday_effects.ipynb
View file

@ -728,7 +728,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"NFL Sundays could also have been handled using the \"holidays\" interface described above, by creating a list of past and future NFL Sundays. The `add_regressor` function provides a more general interface for defining extra linear regressors, and in particular does not require that the regressor be a binary indicator. Another time series could be used as a regressor, although its future values would have to be known.\n",
"NFL Sundays could also have been handled using the \"holidays\" interface described above, by creating a list of past and future NFL Sundays. The `add_regressor` function provides a more general interface for defining extra linear regressors, and in particular does not require that the regressor be a binary indicator. Another time series could be used as a regressor, although its future values would have to be known. The regressor cannot be constant in the training data; fitting will exit with an error if it is.\n",
"\n",
"The `add_regressor` function has optional arguments for specifying the prior scale (holiday prior scale is used by default) and whether or not the regressor is standardized - see the docstring with `help(Prophet.add_regressor)` in Python and `?add_regressor` in R."
]

63
python/fbprophet/forecaster.py
View file

@ -11,6 +11,7 @@ from __future__ import print_function
from __future__ import unicode_literals
from collections import defaultdict
from copy import deepcopy
from datetime import timedelta
import logging
@ -278,6 +279,9 @@ class Prophet(object):
self.t_scale = df['ds'].max() - self.start
for name, props in self.extra_regressors.items():
standardize = props['standardize']
n_vals = len(df[name].unique())
if n_vals < 2:
raise ValueError('Regressor {} is constant.'.format(name))
if standardize == 'auto':
if set(df[name].unique()) == set([1, 0]):
# Don't standardize binary variables.
@ -287,8 +291,6 @@ class Prophet(object):
if standardize:
mu = df[name].mean()
std = df[name].std()
if std == 0:
std = mu
self.extra_regressors[name]['mu'] = mu
self.extra_regressors[name]['std'] = std
@ -1248,16 +1250,16 @@ class Prophet(object):
ax = fig.add_subplot(111)
else:
fig = ax.get_figure()
ax.plot(self.history['ds'].values, self.history['y'], 'k.')
ax.plot(fcst['ds'].values, fcst['yhat'], ls='-', c='#0072B2')
fcst_t = fcst['ds'].dt.to_pydatetime()
ax.plot(self.history['ds'].dt.to_pydatetime(), self.history['y'], 'k.')
ax.plot(fcst_t, fcst['yhat'], ls='-', c='#0072B2')
if 'cap' in fcst and plot_cap:
ax.plot(fcst['ds'].values, fcst['cap'], ls='--', c='k')
ax.plot(fcst_t, fcst['cap'], ls='--', c='k')
if self.logistic_floor and 'floor' in fcst and plot_cap:
ax.plot(fcst['ds'].values, fcst['floor'], ls='--', c='k')
ax.plot(fcst_t, fcst['floor'], ls='--', c='k')
if uncertainty:
ax.fill_between(fcst['ds'].values, fcst['yhat_lower'],
fcst['yhat_upper'], color='#0072B2',
alpha=0.2)
ax.fill_between(fcst_t, fcst['yhat_lower'], fcst['yhat_upper'],
color='#0072B2', alpha=0.2)
ax.grid(True, which='major', c='gray', ls='-', lw=1, alpha=0.2)
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
@ -1345,15 +1347,16 @@ class Prophet(object):
if not ax:
fig = plt.figure(facecolor='w', figsize=(10, 6))
ax = fig.add_subplot(111)
artists += ax.plot(fcst['ds'].values, fcst[name], ls='-', c='#0072B2')
fcst_t = fcst['ds'].dt.to_pydatetime()
artists += ax.plot(fcst_t, fcst[name], ls='-', c='#0072B2')
if 'cap' in fcst and plot_cap:
artists += ax.plot(fcst['ds'].values, fcst['cap'], ls='--', c='k')
artists += ax.plot(fcst_t, fcst['cap'], ls='--', c='k')
if self.logistic_floor and 'floor' in fcst and plot_cap:
ax.plot(fcst['ds'].values, fcst['floor'], ls='--', c='k')
ax.plot(fcst_t, fcst['floor'], ls='--', c='k')
if uncertainty:
artists += [ax.fill_between(
fcst['ds'].values, fcst[name + '_lower'],
fcst[name + '_upper'], color='#0072B2', alpha=0.2)]
fcst_t, fcst[name + '_lower'], fcst[name + '_upper'],
color='#0072B2', alpha=0.2)]
ax.grid(True, which='major', c='gray', ls='-', lw=1, alpha=0.2)
ax.set_xlabel('ds')
ax.set_ylabel(name)
@ -1441,11 +1444,11 @@ class Prophet(object):
pd.Timedelta(days=yearly_start))
df_y = self.seasonality_plot_df(days)
seas = self.predict_seasonal_components(df_y)
artists += ax.plot(df_y['ds'], seas['yearly'], ls='-',
c='#0072B2')
artists += ax.plot(
df_y['ds'].dt.to_pydatetime(), seas['yearly'], ls='-', c='#0072B2')
if uncertainty:
artists += [ax.fill_between(
df_y['ds'].values, seas['yearly_lower'],
df_y['ds'].dt.to_pydatetime(), seas['yearly_lower'],
seas['yearly_upper'], color='#0072B2', alpha=0.2)]
ax.grid(True, which='major', c='gray', ls='-', lw=1, alpha=0.2)
months = MonthLocator(range(1, 13), bymonthday=1, interval=2)
@ -1481,14 +1484,16 @@ class Prophet(object):
days = pd.to_datetime(np.linspace(start.value, end.value, plot_points))
df_y = self.seasonality_plot_df(days)
seas = self.predict_seasonal_components(df_y)
artists += ax.plot(df_y['ds'], seas[name], ls='-',
artists += ax.plot(df_y['ds'].dt.to_pydatetime(), seas[name], ls='-',
c='#0072B2')
if uncertainty:
artists += [ax.fill_between(
df_y['ds'].values, seas[name + '_lower'],
df_y['ds'].dt.to_pydatetime(), seas[name + '_lower'],
seas[name + '_upper'], color='#0072B2', alpha=0.2)]
ax.grid(True, which='major', c='gray', ls='-', lw=1, alpha=0.2)
ax.set_xticks(pd.to_datetime(np.linspace(start.value, end.value, 7)))
xticks = pd.to_datetime(np.linspace(start.value, end.value, 7)
).to_pydatetime()
ax.set_xticks(xticks)
if period <= 2:
fmt_str = '{dt:%T}'
elif period < 14:
@ -1514,6 +1519,9 @@ class Prophet(object):
-------
Prophet class object with the same parameter with model variable
"""
if self.history is None:
raise Exception('This is for copying a fitted Prophet object.')
if self.specified_changepoints:
changepoints = self.changepoints
if cutoff is not None:
@ -1522,18 +1530,23 @@ class Prophet(object):
else:
changepoints = None
return Prophet(
# Auto seasonalities are set to False because they are already set in
# self.seasonalities.
m = Prophet(
growth=self.growth,
n_changepoints=self.n_changepoints,
changepoints=changepoints,
yearly_seasonality=self.yearly_seasonality,
weekly_seasonality=self.weekly_seasonality,
daily_seasonality=self.daily_seasonality,
yearly_seasonality=False,
weekly_seasonality=False,
daily_seasonality=False,
holidays=self.holidays,
seasonality_prior_scale=self.seasonality_prior_scale,
changepoint_prior_scale=self.changepoint_prior_scale,
holidays_prior_scale=self.holidays_prior_scale,
mcmc_samples=self.mcmc_samples,
interval_width=self.interval_width,
uncertainty_samples=self.uncertainty_samples
uncertainty_samples=self.uncertainty_samples,
)
m.extra_regressors = deepcopy(self.extra_regressors)
m.seasonalities = deepcopy(self.seasonalities)
return m

8
python/fbprophet/tests/test_diagnostics.py
View file

@ -84,7 +84,9 @@ class TestDiagnostics(TestCase):
df_shf2 = diagnostics.simulated_historical_forecasts(
m, horizon='10 days', k=1, period='5 days')
self.assertAlmostEqual(
((df_shf1 - df_shf2)**2)[['y', 'yhat']].sum().sum(), 0.0)
((df_shf1['y'] - df_shf2['y']) ** 2).sum(), 0.0)
self.assertAlmostEqual(
((df_shf1['yhat'] - df_shf2['yhat']) ** 2).sum(), 0.0)
def test_cross_validation(self):
m = Prophet()
@ -111,4 +113,6 @@ class TestDiagnostics(TestCase):
df_cv2 = diagnostics.cross_validation(
m, horizon='32 days', period='10 days', initial='96 days')
self.assertAlmostEqual(
((df_cv1 - df_cv2)**2)[['y', 'yhat']].sum().sum(), 0.0)
((df_cv1['y'] - df_cv2['y']) ** 2).sum(), 0.0)
self.assertAlmostEqual(
((df_cv1['yhat'] - df_cv2['yhat']) ** 2).sum(), 0.0)

33
python/fbprophet/tests/test_prophet.py
View file

@ -62,7 +62,6 @@ class TestProphet(TestCase):
def test_fit_changepoint_not_in_history(self):
train = DATA[(DATA['ds'] < '2013-01-01') | (DATA['ds'] > '2014-01-01')]
train[(train['ds'] > '2014-01-01')] += 20
future = pd.DataFrame({'ds': DATA['ds']})
forecaster = Prophet(changepoints=['2013-06-06'])
forecaster.fit(train)
@ -548,8 +547,17 @@ class TestProphet(TestCase):
fcst['yhat'][0],
fcst['trend'][0] + fcst['seasonal'][0],
)
# Check fails if constant extra regressor
df['constant_feature'] = 5
m = Prophet()
m.add_regressor('constant_feature')
with self.assertRaises(ValueError):
m.fit(df.copy())
def test_copy(self):
df = DATA.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']})
@ -571,13 +579,22 @@ class TestProphet(TestCase):
# Values should be copied correctly
for product in products:
m1 = Prophet(*product)
m1.history = m1.setup_dataframe(
df.copy(), initialize_scales=True)
m1.set_auto_seasonalities()
m2 = m1.copy()
self.assertEqual(m1.growth, m2.growth)
self.assertEqual(m1.n_changepoints, m2.n_changepoints)
self.assertEqual(m1.changepoints, m2.changepoints)
self.assertEqual(m1.yearly_seasonality, m2.yearly_seasonality)
self.assertEqual(m1.weekly_seasonality, m2.weekly_seasonality)
self.assertEqual(m1.daily_seasonality, m2.daily_seasonality)
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:
@ -589,11 +606,15 @@ class TestProphet(TestCase):
self.assertEqual(m1.interval_width, m2.interval_width)
self.assertEqual(m1.uncertainty_samples, m2.uncertainty_samples)
# Check for cutoff
# 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.fit(DATA)
m1.add_seasonality('custom', 10, 5)
m1.add_regressor('binary_feature')
m1.fit(df)
m2 = m1.copy(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)