From f3f1c003c312bdbba3ebffc6e57353da7bc54e06 Mon Sep 17 00:00:00 2001
From: "Daniel.Frisinghelli" <daniel.frisinghelli@eurac.edu>
Date: Tue, 26 Oct 2021 15:22:25 +0200
Subject: [PATCH] Implemented generic computation of anomalies on arbitrary
time-scale.
---
climax/core/dataset.py | 47 ++++++++++++++++++++++++++++++++++++++++++
1 file changed, 47 insertions(+)
diff --git a/climax/core/dataset.py b/climax/core/dataset.py
index fce18fe..332442b 100644
--- a/climax/core/dataset.py
+++ b/climax/core/dataset.py
@@ -6,6 +6,7 @@
# builtins
import logging
import pathlib
+import warnings
from datetime import date
# externals
@@ -232,6 +233,52 @@ class EoDataset(torch.utils.data.Dataset):
return dem_features
+ @staticmethod
+ def anomalies(ds, timescale='time.dayofyear', standard=False):
+ # group dataset by day of the year
+ LOGGER.info('Computing standardized anomalies ...')
+ groups = ds.groupby('time.dayofyear').groups
+
+ # compute standardized anomalies for each day of the year over time
+ anomalies = {}
+ for time, time_scale in groups.items():
+ with warnings.catch_warnings():
+ warnings.simplefilter('ignore', category=RuntimeWarning)
+ # anomaly = (x(t) - mean(x, t))
+ anomalies[time] = (ds.isel(time=time_scale) -
+ ds.isel(time=time_scale).mean(dim='time'))
+
+ # standardized anomaly = (x(t) - mean(x, t)) / std(x, t)
+ if standard:
+ anomalies[time] = (anomalies[time] /
+ ds.isel(time=time_scale).std(dim='time')
+ )
+
+ # concatenate anomalies and sort chronologically
+ anomalies = xr.concat(anomalies.values(), dim='time')
+ anomalies = anomalies.sortby(anomalies.time)
+
+ return anomalies
+
+ @staticmethod
+ def normalize(ds, dim=('time', 'y', 'x'), period=None):
+ # normalize predictors to [0, 1]
+ LOGGER.info('Normalizing data to [0, 1] ...')
+ with warnings.catch_warnings():
+ warnings.simplefilter('ignore', category=RuntimeWarning)
+
+ # whether to normalize using statistics for a specific period
+ # NOTE: this can result in values that are not in [0, 1]
+ if period is not None:
+ ds -= ds.sel(time=period).min(dim=dim)
+ ds /= ds.sel(time=period).max(dim=dim)
+ # normalize using entire period: [0, 1]
+ else:
+ ds -= ds.min(dim=dim)
+ ds /= ds.max(dim=dim)
+
+ return ds
+
class NetCDFDataset(EoDataset):
--
GitLab