Timeseries upd (#81)

* set autocorrelation of lag 0 to be 1 * updated fix for differencing in timeseries models * windows test files Co-authored-by: dmorgankx <[email protected]> Co-authored-by: Deanna Morgan <[email protected]>
KxSystems · Oct 7, 2020 · 0628728 · 0628728
1 parent 09ad749
commit 0628728
Show file tree

Hide file tree

Showing 11 changed files with 9 additions and 9 deletions.
diff --git a/fresh/extract.q b/fresh/extract.q
@@ -10,7 +10,7 @@ fresh.feat.agglintrend:{
  t:fresh.feat.lintrend each(max;min;var;avg)@/:\:y cut x;
  (`$"_"sv'string cols[t]cross`max`min`var`avg)!raze value flip t}
 fresh.feat.augfuller:{`teststat`pvalue`usedlag!3#"f"$@[{fresh.i.adfuller[x]`};x;0n]}
-fresh.feat.autocorr:{(avg(x-m)*xprev[y;x]-m:avg x)%var x}
+fresh.feat.autocorr:{$[y=0;1f;(avg(x-m)*xprev[y;x]-m:avg x)%var x]}
 fresh.feat.binnedentropy:{neg sum p*log p:(count each group(y-1)&floor y*x%max x-:min x)%count x}
 / t-series non-linearity - Schreiber, T. and Schmitz, A. (1997). PHYSICAL REVIEW E, VOLUME 55, NUMBER 5
 fresh.feat.c3:{avg x*/xprev\:[-1 -2*y]x}

diff --git a/timeseries/fit.q b/timeseries/fit.q
@@ -69,7 +69,7 @@ ts.ARMA.fit:{[endog;exog;lags;resid;trend]
 ts.ARIMA.fit:{[endog;exog;lags;diff;resid;trend]
   exog:ts.i.fitDataCheck[endog;exog];
   // Apply integration (non seasonal)
-  I:ts.i.differ[endog;diff;()!()];
+  I:ts.i.differ[endog;diff;()!()]`final;
   // Fit an ARMA model on the differenced time series
   mdl:ts.ARMA.fit[I;diff _exog;lags;resid;trend];
   // Retrieve the original data to be used when fitting on new data
@@ -107,9 +107,9 @@ ts.SARIMA.fit:{[endog;exog;lags;diff;resid;trend;seas]
   // add additional seasonal components
   dict[`seas_add_P`seas_add_Q]:(raze'){1+til[x]+/:y}'[(lags;resid);dict`P`Q];
   // Generate data for regenerate data following differencing
-  origDiffSeason:`origd`origs!(diff{deltas x}/neg[diff]#endog;neg[prd seas`D`m]#endog);
+  origDiffSeason:`origd`origs!(diff{deltas x}/neg[diff]#endog;neg[prd seas`D`m]#I`init);
   // Apply SARMA model and postpend differenced original data
-  ts.i.SARMA.model[I;exog;dict],origDiffSeason
+  ts.i.SARMA.model[I`final;exog;dict],origDiffSeason
   }
 
 // @kind function

diff --git a/timeseries/tests/data/linux/fit/SARIMA3 b/timeseries/tests/data/linux/fit/SARIMA3
diff --git a/timeseries/tests/data/linux/fit/SARIMA4 b/timeseries/tests/data/linux/fit/SARIMA4
diff --git a/timeseries/tests/data/linux/pred/predSARIMA3 b/timeseries/tests/data/linux/pred/predSARIMA3
diff --git a/timeseries/tests/data/linux/pred/predSARIMA4 b/timeseries/tests/data/linux/pred/predSARIMA4
diff --git a/timeseries/tests/data/windows/fit/SARIMA3 b/timeseries/tests/data/windows/fit/SARIMA3
diff --git a/timeseries/tests/data/windows/fit/SARIMA4 b/timeseries/tests/data/windows/fit/SARIMA4
diff --git a/timeseries/tests/data/windows/pred/predSARIMA3 b/timeseries/tests/data/windows/pred/predSARIMA3
diff --git a/timeseries/tests/data/windows/pred/predSARIMA4 b/timeseries/tests/data/windows/pred/predSARIMA4
diff --git a/timeseries/utils.q b/timeseries/utils.q
@@ -709,17 +709,17 @@ ts.i.predDataCheck:{[mdl;exog]
 // @param endog {num[]}   endogenous dataset
 // @param diff  {integer} non seasonal differencing component (integer)
 // @param sdict {dict}    dictionary containing relevant seasonal differencing components
-// @return {num[]} Seasonal and non-seasonally differenced stationary time-series
+// @return {dict} Seasonal and non-seasonally differenced stationary time-series
 ts.i.differ:{[endog;d;s]
   // Apply non seasonal differencing if appropriate (handling of AR/ARMA)
   if[s~()!();s[`D]:0b];
-  I:ts.i.diff[endog;d];
+  initDiff:ts.i.diff[endog;d];
   // Apply seasonal differencing if appropriate
-  if[s[`D];I:s[`D]ts.i.seasonDiff[s`m]/I];
+  finalDiff:$[s[`D];s[`D]ts.i.seasonDiff[s`m]/initDiff;initDiff];
   // Check stationarity
-  if[not ts.i.stationary[I];ts.i.err.stat[]];
+  if[not ts.i.stationary[finalDiff];ts.i.err.stat[]];
   // Return integrated data
-  I
+  `final`init!(finalDiff;initDiff)
   }