From ba1bac834e864ccf736397194847b3ac95cff8cf Mon Sep 17 00:00:00 2001
From: Marc Ferradou <marc.ferradou@buzzfeed.com>
Date: Fri, 25 May 2018 18:53:19 -0400
Subject: [PATCH] Adding changepoint threshold (#299)

---
 R/R/prophet.R                  | 15 ++++++++++++---
 python/fbprophet/forecaster.py | 17 ++++++++++++++---
 2 files changed, 26 insertions(+), 6 deletions(-)

diff --git a/R/R/prophet.R b/R/R/prophet.R
index 92ed985..2c48d90 100644
--- a/R/R/prophet.R
+++ b/R/R/prophet.R
@@ -28,7 +28,9 @@ utils::globalVariables(c(
 #' @param n.changepoints Number of potential changepoints to include. Not used
 #'  if input `changepoints` is supplied. If `changepoints` is not supplied,
 #'  then n.changepoints potential changepoints are selected uniformly from the
-#'  first 80 percent of df$ds.
+#'  first `changepoint.threshold` percent of df$ds.
+#' @param changepoint.threshold Parameter controling where to select the changepoints.
+#'  Not used if input `changepoints` is supplied.
 #' @param yearly.seasonality Fit yearly seasonality. Can be 'auto', TRUE,
 #'  FALSE, or a number of Fourier terms to generate.
 #' @param weekly.seasonality Fit weekly seasonality. Can be 'auto', TRUE,
@@ -79,6 +81,7 @@ prophet <- function(df = NULL,
                     growth = 'linear',
                     changepoints = NULL,
                     n.changepoints = 25,
+                    changepoint.threshold = 0.8,
                     yearly.seasonality = 'auto',
                     weekly.seasonality = 'auto',
                     daily.seasonality = 'auto',
@@ -103,6 +106,7 @@ prophet <- function(df = NULL,
     growth = growth,
     changepoints = changepoints,
     n.changepoints = n.changepoints,
+    changepoint.threshold = changepoint.threshold,
     yearly.seasonality = yearly.seasonality,
     weekly.seasonality = weekly.seasonality,
     daily.seasonality = daily.seasonality,
@@ -451,9 +455,14 @@ set_changepoints <- function(m) {
       }
     }
   } else {
-    # Place potential changepoints evenly through the first 80 pcnt of
+    # Place potential changepoints evenly through the first changepoint.threshold pcnt of
     # the history.
-    hist.size <- floor(nrow(m$history) * .8)
+    if (m$changepoint.threshold > 1 || m$changepoint.threshold <= 0){
+        m$changepoint.threshold <- .8
+        message('changepoint.threshold greater than 1 or less than equal to 0. Using ',
+                m$changepoint.threshold)
+    }
+    hist.size <- floor(nrow(m$history) * m$changepoint.threshold)
     if (m$n.changepoints + 1 > hist.size) {
       m$n.changepoints <- hist.size - 1
       message('n.changepoints greater than number of observations. Using ',
diff --git a/python/fbprophet/forecaster.py b/python/fbprophet/forecaster.py
index f9a7984..bd9aa9a 100644
--- a/python/fbprophet/forecaster.py
+++ b/python/fbprophet/forecaster.py
@@ -57,7 +57,9 @@ class Prophet(object):
     n_changepoints: Number of potential changepoints to include. Not used
         if input `changepoints` is supplied. If `changepoints` is not supplied,
         then n_changepoints potential changepoints are selected uniformly from
-        the first 80 percent of the history.
+        the first `changepoint_threshold` percent of the history.
+    changepoint_threshold: Parameter controling where to select the changepoints.
+    Not used if input `changepoints` is supplied.
     yearly_seasonality: Fit yearly seasonality.
         Can be 'auto', True, False, or a number of Fourier terms to generate.
     weekly_seasonality: Fit weekly seasonality.
@@ -97,6 +99,7 @@ class Prophet(object):
             growth='linear',
             changepoints=None,
             n_changepoints=25,
+            changepoint_threshold=0.8,
             yearly_seasonality='auto',
             weekly_seasonality='auto',
             daily_seasonality='auto',
@@ -119,6 +122,7 @@ class Prophet(object):
             self.n_changepoints = n_changepoints
             self.specified_changepoints = False
 
+        self.changepoint_threshold = changepoint_threshold
         self.yearly_seasonality = yearly_seasonality
         self.weekly_seasonality = weekly_seasonality
         self.daily_seasonality = daily_seasonality
@@ -332,8 +336,15 @@ class Prophet(object):
                     raise ValueError(
                         'Changepoints must fall within training data.')
         else:
-            # Place potential changepoints evenly through first 80% of history
-            hist_size = np.floor(self.history.shape[0] * 0.8)
+            # Place potential changepoints evenly through first changepoint_threshold
+            # of history
+            if (self.changepoint_threshold > 1 or self.changepoint_threshold <= 0):
+                self.changepoint_threshold = 0.8
+                logger.info(
+                    'changepoint_threshold greater than 1 or less than equal to 0.'
+                    'Using {}.'.format(self.changepoint_threshold)
+                )
+            hist_size = np.floor(self.history.shape[0] * self.changepoint_threshold)
             if self.n_changepoints + 1 > hist_size:
                 self.n_changepoints = hist_size - 1
                 logger.info(