From 6a8dc893f95ea3251ae788d2befef7a3ec81efad Mon Sep 17 00:00:00 2001
From: ValentinaHutter <85164505+ValentinaHutter@users.noreply.github.com>
Date: Wed, 13 Apr 2022 10:21:53 +0200
Subject: [PATCH 1/3] Use stack and unstack inside prediction

---
 src/openeo_processes/cubes.py | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/src/openeo_processes/cubes.py b/src/openeo_processes/cubes.py
index c08db242..778f3511 100644
--- a/src/openeo_processes/cubes.py
+++ b/src/openeo_processes/cubes.py
@@ -2067,6 +2067,7 @@ def exec_xar(data, model, dimension, predictors_vars = None, client = None):
         predictor_cols = list(data.dims)
         if dimension in predictor_cols:
             predictor_cols.remove(dimension)
+        unstack = False
         if len(predictor_cols) == 1:
             stacked = data
             I = stacked[predictor_cols[0]].values
@@ -2074,6 +2075,7 @@ def exec_xar(data, model, dimension, predictors_vars = None, client = None):
             stacked = data.stack(z=predictor_cols)
             I = stacked['z'].values
             predictor_cols = ['z']
+            unstack = True
 
         X_hat = stacked.to_dataset(dim=dimension).to_dask_dataframe().drop(columns=predictor_cols)
         if 'spatial_ref' in X_hat:
@@ -2095,6 +2097,8 @@ def exec_xar(data, model, dimension, predictors_vars = None, client = None):
         predictions_xr = xr.ones_like(p) * y_hat_da
         predictions_xr.attrs = data.attrs
         predictions_xr[dimension] = np.array(['prediction'])
+        if unstack:
+            predictions_xr = predictions_xr.unstack()
         return predictions_xr
 
 

From 8dd5da3d70cb1105253e79f51b5b6ef570f17e5d Mon Sep 17 00:00:00 2001
From: ValentinaHutter <85164505+ValentinaHutter@users.noreply.github.com>
Date: Tue, 19 Apr 2022 10:10:37 +0200
Subject: [PATCH 2/3] remove predictors_vars parameter

---
 src/openeo_processes/cubes.py | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/src/openeo_processes/cubes.py b/src/openeo_processes/cubes.py
index deb5c704..2ad95842 100644
--- a/src/openeo_processes/cubes.py
+++ b/src/openeo_processes/cubes.py
@@ -2087,7 +2087,7 @@ def predict_random_forest():
 class PredictRandomForest:
 
     @staticmethod
-    def exec_xar(data, model, dimension, predictors_vars = None, client = None):
+    def exec_xar(data, model, dimension, client = None):
         if dimension in ['time', 't', 'times']:  # time dimension must be converted into values
             dimension = get_time_dimension_from_data(data, dimension)
         predictor_cols = list(data.dims)
@@ -2104,12 +2104,11 @@ def exec_xar(data, model, dimension, predictors_vars = None, client = None):
             unstack = True
 
         X_hat = stacked.to_dataset(dim=dimension).to_dask_dataframe().drop(columns=predictor_cols)
-        if 'spatial_ref' in X_hat:
-            X_hat = X_hat.drop(columns=['spatial_ref'])
-        if predictors_vars is not None:
-            for c in X_hat.columns:
-                if c not in predictors_vars:
-                    X_hat = X_hat.drop(columns=c)
+
+        predictors_vars = data[dimension].values
+        for c in X_hat.columns:
+            if c not in predictors_vars:
+                X_hat = X_hat.drop(columns=c)
 
         # Make sure that feature names fit to those of the trained model
         X_hat = X_hat[model.feature_names]

From c46b5d31713c81ec48b0821f025e128e3f93661d Mon Sep 17 00:00:00 2001
From: ValentinaHutter <85164505+ValentinaHutter@users.noreply.github.com>
Date: Tue, 19 Apr 2022 16:15:44 +0200
Subject: [PATCH 3/3] load model in prediction

---
 src/openeo_processes/cubes.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/openeo_processes/cubes.py b/src/openeo_processes/cubes.py
index 2ad95842..22d5a001 100644
--- a/src/openeo_processes/cubes.py
+++ b/src/openeo_processes/cubes.py
@@ -2087,7 +2087,9 @@ def predict_random_forest():
 class PredictRandomForest:
 
     @staticmethod
-    def exec_xar(data, model, dimension, client = None):
+    def exec_xar(data, model, dimension, client = None, input_filepath = None):
+        if isinstance(model, str):
+            model = load_ml_model(model, input_filepath = input_filepath)
         if dimension in ['time', 't', 'times']:  # time dimension must be converted into values
             dimension = get_time_dimension_from_data(data, dimension)
         predictor_cols = list(data.dims)