Merge pull request #9 from Techtonique/dov080

align with R version 0.8.1

Author: thierrymoudiki

CHANGES.md

@@ -1,3 +1,9 @@
+# version 0.8.1
+
+- Align with R version (see [https://github.com/Techtonique/ahead/blob/main/NEWS.md#version-070](https://github.com/Techtonique/ahead/blob/main/NEWS.md#version-070) and [https://github.com/Techtonique/ahead/blob/main/NEWS.md#version-080](https://github.com/Techtonique/ahead/blob/main/NEWS.md#version-080)) as much as possible
+- moving block bootstrap in `ridge2f`, `basicf`, in addition to circular block bootstrap from 0.6.2
+- adjust R-Vine copulas on residuals for `ridge2f` simulation (with empirical and Gaussian marginals)
+
 # version 0.6.2
 
 - Add Block Bootstrap to `ridge2f` and `basicf`

ahead/Basic/BasicForecaster.py

@@ -20,21 +20,23 @@ class BasicForecaster():
 
         type_pi: a string;
             Type of prediction interval (currently "gaussian",
-            or "bootstrap")
+            "bootstrap" (independent), "blockbootstrap" (circular), 
+            "movingblockbootstrap")
 
         block_length: an integer
-            length of block for multivariate circular block bootstrap (`type_pi == blockbootstrap`)            
+            length of block for multivariate block bootstrap (`type_pi == blockbootstrap`
+            or `type_pi == movingblockbootstrap`)            
             
         B: an integer;
-            Number of bootstrap replications for `type_pi == bootstrap`
+            Number of replications
 
         date_formatting: a string;
             Currently:
             - "original": yyyy-mm-dd
             - "ms": milliseconds
 
         seed: an integer;
-            reproducibility seed for type_pi == 'bootstrap'
+            reproducibility seed 
 
     Attributes:
 
@@ -151,8 +153,8 @@ def forecast(self, df):
 
         y = mv.compute_y_mts(self.input_df, frequency)
 
-        if self.type_pi is "blockbootstrap":            
-            assert self.block_length is not None, "For `type_pi == 'blockbootstrap'`, `block_length` must be not None"
+        if self.type_pi in ("blockbootstrap", "movingblockbootstrap"):            
+            assert self.block_length is not None, "For `type_pi in ('blockbootstrap', 'movingblockbootstrap')`, `block_length` must be not None"
 
         self.fcast_ = config.AHEAD_PACKAGE.basicf(
             y,
@@ -188,7 +190,8 @@ def forecast(self, df):
             for i in range(n_series)
         )
 
-        if self.type_pi == "bootstrap" or self.type_pi == "blockbootstrap":
+        if self.type_pi in ("bootstrap", "blockbootstrap", 
+                            "movingblockbootstrap"):
             self.sims_ = tuple(
                 np.asarray(self.fcast_.rx2["sims"][i]) for i in range(self.B)
             )

ahead/Ridge2/Ridge2Regressor.py

@@ -44,13 +44,19 @@ class Ridge2Regressor():
 
         type_pi: a string;
             Type of prediction interval (currently "gaussian",
-            or "bootstrap" or "blockbootstrap")
+            "bootstrap", (circular) "blockbootstrap", "movingblockbootstrap", or "rvinecopula")
         
         block_length: an integer
-            length of block for multivariate circular block bootstrap (`type_pi == blockbootstrap`)
+            length of block for multivariate block bootstrap (`type_pi == blockbootstrap` or 
+            `type_pi == movingblockbootstrap`)
+        
+        margins: a string;
+            distribution of residuals' marginals for `type_pi == rvinecopula`: "empirical" (default), 
+            "gaussian"
 
         B: an integer;
-            Number of bootstrap replications for `type_pi == bootstrap` or `type_pi == blockbootstrap` 
+            Number of bootstrap replications for `type_pi == bootstrap`, "blockbootstrap", 
+            "movingblockbootstrap", or "rvinecopula"
         
         type_aggregation: a string;
             Type of aggregation, ONLY for bootstrapping; either "mean" or "median"
@@ -140,6 +146,7 @@ def __init__(
         dropout = 0,
         type_pi = "gaussian",
         block_length = 3, # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
+        margins = "empirical",
         B = 100,
         type_aggregation = "mean",
         centers = 2,
@@ -148,6 +155,12 @@ def __init__(
         date_formatting ="original",
         seed =123,
     ):
+        
+        if not config.R_IS_INSTALLED:
+            raise ImportError("R is not installed! \n" + config.USAGE_MESSAGE)
+        
+        if not config.RPY2_IS_INSTALLED:
+            raise ImportError(config.RPY2_ERROR_MESSAGE + config.USAGE_MESSAGE)                
 
         self.h = h
         self.level = level
@@ -161,6 +174,7 @@ def __init__(
         self.dropout = dropout
         self.type_pi = type_pi
         self.block_length = block_length
+        self.margins = margins
         self.B = B
         self.type_aggregation = type_aggregation
         self.centers = centers # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
@@ -209,29 +223,28 @@ def forecast(self, df, xreg = None):
 
         if xreg is None: 
 
-            with cv.localconverter(default_converter + config.NONE_CONVERTER):
-                self.fcast_ = config.AHEAD_PACKAGE.ridge2f(
-                y,
-                xreg = xreg,
-                h=self.h,
-                level=self.level,
-                lags=self.lags,
-                nb_hidden=self.nb_hidden,
-                nodes_sim=self.nodes_sim,
-                activ=self.activation,
-                a=self.a,
-                lambda_1=self.lambda_1,
-                lambda_2=self.lambda_2,
-                dropout=self.dropout,
-                type_pi=self.type_pi,
-                block_length=self.block_length, # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
-                B=self.B,
-                type_aggregation = self.type_aggregation,
-                centers = self.centers,  # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
-                type_clustering = self.type_clustering,
-                cl=self.cl,
-                seed=self.seed,
-            ) 
+            self.fcast_ = config.AHEAD_PACKAGE.ridge2f(
+            y,
+            h=self.h,
+            level=self.level,
+            lags=self.lags,
+            nb_hidden=self.nb_hidden,
+            nodes_sim=self.nodes_sim,
+            activ=self.activation,
+            a=self.a,
+            lambda_1=self.lambda_1,
+            lambda_2=self.lambda_2,
+            dropout=self.dropout,
+            type_pi=self.type_pi,
+            margins=self.margins,
+            block_length=self.block_length, # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
+            B=self.B,
+            type_aggregation = self.type_aggregation,
+            centers = self.centers,  # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
+            type_clustering = self.type_clustering,
+            cl=self.cl,
+            seed=self.seed,
+        )         
 
         else: # xreg is not None:  
 
@@ -263,6 +276,7 @@ def forecast(self, df, xreg = None):
             lambda_2=self.lambda_2,
             dropout=self.dropout,
             type_pi=self.type_pi,
+            margins=self.margins,
             block_length=self.block_length, # can be NULL, but in R (use 0 in R instead of NULL for v0.7.0)
             B=self.B,
             type_aggregation = self.type_aggregation,
@@ -295,7 +309,8 @@ def forecast(self, df, xreg = None):
             for i in range(n_series)
         )
 
-        if self.type_pi == "bootstrap" or self.type_pi == "blockbootstrap":
+        if self.type_pi in ("bootstrap", "blockbootstrap", 
+                            "movingblockbootstrap", "rvinecopula"):
             self.sims_ = tuple(
                 np.asarray(self.fcast_.rx2["sims"][i]) for i in range(self.B)
             )

ahead/utils/univariate.py

@@ -10,6 +10,7 @@
 from rpy2.robjects.vectors import StrVector
 from .unimultivariate import get_frequency
 
+base = importr("base")
 stats = importr("stats")
 
 
@@ -20,10 +21,9 @@ def compute_y_ts(df, df_frequency):
     ts = stats.ts(
         FloatVector(input_series.flatten()),
         frequency=get_frequency(df_frequency),
+        names = StrVector(df.columns.tolist())
     )
 
-    ts.colnames = StrVector(df.columns.tolist())
-
     return ts
 
 

examples/basicforecaster.py

@@ -122,3 +122,37 @@
 print(d4.mean_)
 print(d4.lower_)
 print(d4.upper_)
+
+print("\n")
+print("Example 5 -----")
+
+dataset = {
+ 'date' : ['2001-01-01', '2002-01-01', '2003-01-01', 
+           '2004-01-01', '2005-01-01', '2006-01-01', 
+           '2007-01-01'],
+ 'series1' : [34, 30, 35.6, 33.3, 38.1, 34.4, 33.9],    
+ 'series2' : [4, 5.5, 5.6, 6.3, 5.1, 4.9, 4.7],
+ 'series3' : [100, 100.5, 100.6, 100.2, 100.1, 99.9, 101.0]}
+df = pd.DataFrame(dataset).set_index('date')
+
+
+d5 = BasicForecaster(h = 5, date_formatting = "original", 
+type_pi="movingblockbootstrap", B=20, block_length=3)
+
+start = time()
+d5.forecast(df)
+print(f"Elapsed: {time()-start} \n")
+
+print("\n output_dates_: \n")
+print(d5.output_dates_)
+
+print("\n mean, lower, upper as numpy arrays: \n")
+print(d5.mean_)
+print(d5.lower_)
+print(d5.upper_)
+
+print("\n result_dfs_: \n")
+print(d5.result_dfs_)
+
+print("\n sims_: \n")
+print(d5.sims_)

examples/ridge2regressor.py

@@ -38,7 +38,7 @@
 print("\n")
 print(d1.ranges_[2])
 print("\n")
-plot(d1, selected_series="series1")
+#plot(d1, selected_series="series1")
 
 print("Example 2 -----")
 d2 = Ridge2Regressor(h = h, date_formatting = "original")
@@ -58,7 +58,7 @@
 print("\n")
 print(d2.ranges_[2])
 print("\n")
-plot(d2, selected_series="series1")
+#plot(d2, selected_series="series1")
 print("\n")
 
 print("Example 3 -----")
@@ -94,7 +94,7 @@
 print(d3.lower_)
 print(d3.upper_)
 print("\n")
-plot(d3, selected_series="series1")
+#plot(d3, selected_series="series1")
 print("\n")
 
 print("Example 4 -----")
@@ -141,7 +141,7 @@
 print(d4.lower_)
 print(d4.upper_)
 print("\n")
-plot(d4, selected_series="series1")
+#plot(d4, selected_series="series1")
 print("\n")
 
 # print("Example 5 -----")
@@ -191,3 +191,50 @@
 # plot(d5, selected_series="series1")
 # print("\n")
 
+
+print("Example 6 -----")
+
+d6 = Ridge2Regressor(h = 7, date_formatting = "original", 
+                     type_pi="rvinecopula", 
+                     margins="empirical",
+                     B=10, seed=1)
+
+start = time()
+d6.forecast(df, xreg = xreg)
+print(f"Elapsed: {time()-start} \n")
+
+print("\n mean: \n")
+print(d6.mean_)
+
+print("\n lower: \n")
+print(d6.lower_)
+
+print("\n upper: \n")
+print(d6.upper_)
+
+print("\n sims: \n")
+print(d6.sims_)
+print("\n")
+
+print("Example 7 -----")
+
+d7 = Ridge2Regressor(h = 7, date_formatting = "original", 
+                     type_pi="rvinecopula", 
+                     margins="gaussian",
+                     B=8, seed=1)
+
+start = time()
+d7.forecast(df)
+print(f"Elapsed: {time()-start} \n")
+
+print("\n mean: \n")
+print(d7.mean_)
+
+print("\n lower: \n")
+print(d7.lower_)
+
+print("\n upper: \n")
+print(d7.upper_)
+
+print("\n sims: \n")
+print(d6.sims_)

setup.py

@@ -2,7 +2,7 @@
 from codecs import open
 from os import path
 
-__version__ = "0.6.2"
+__version__ = "0.8.1"
 
 here = path.abspath(path.dirname(__file__))