From 3bf034b39fab555c1cca04b17413c5c7bc7399f3 Mon Sep 17 00:00:00 2001
From: Gabriel Moldovan <gabriel.moldovan@ecmwf.int>
Date: Fri, 2 Aug 2024 10:38:07 +0000
Subject: [PATCH] added config for bounding

---
 src/anemoi/training/config/data/zarr.yaml        |  7 +++++++
 src/anemoi/training/config/training/default.yaml | 10 ++++++++++
 2 files changed, 17 insertions(+)

diff --git a/src/anemoi/training/config/data/zarr.yaml b/src/anemoi/training/config/data/zarr.yaml
index 27c17edb..f77fef76 100644
--- a/src/anemoi/training/config/data/zarr.yaml
+++ b/src/anemoi/training/config/data/zarr.yaml
@@ -30,6 +30,9 @@ diagnostic:
 normalizer:
   default: "mean-std"
   min-max:
+  # std:
+  # - "tp"
+  # - "cp"
   max:
   - "sdor"
   - "slor"
@@ -49,6 +52,10 @@ normalizer:
 imputer:
   default: "none"
 
+  # # For the fraction activation to be robust, the fraction variable should
+  # # be normalised with the standard deviation of the total variable
+  # fraction_normalisation: [["tp","cp"]] #first variable is the total variable
+
 # processors including imputers and normalizers are applied in order of definition
 processors:
   # example_imputer:
diff --git a/src/anemoi/training/config/training/default.yaml b/src/anemoi/training/config/training/default.yaml
index 870eeb7a..815a1d78 100644
--- a/src/anemoi/training/config/training/default.yaml
+++ b/src/anemoi/training/config/training/default.yaml
@@ -79,6 +79,16 @@ metrics:
 - u_850
 - v_850
 
+# Bounding configuration
+# bounding_strategies: #order in which these variables are defined matters!
+#   tp:
+#     _target_: anemoi.models.models.bounding.ReluBoundingStrategy #[0, infinity)
+#   cp:
+#     _target_: anemoi.models.models.bounding.FractionHardtanhBoundingStrategy # fraction of tp
+#     min_val: 0
+#     max_val: 1
+#     total_var: tp
+
 pressure_level_scaler:
   _target_: anemoi.training.data.scaling.ReluPressureLevelScaler
   minimum: 0.2