find better compression parameters

models/background.py

@@ -294,8 +294,9 @@ def save( self, filename=None, **kwargs ):
                         f"Variance shape {self.variance.shape} does not match image shape {self.image_shape}"
                     )
 
-                bggrp.create_dataset( 'counts', data=self.counts, compression='gzip', compression_opts=9 )
-                bggrp.create_dataset( 'variance', data=self.variance, compression='gzip', compression_opts=9 )
+                opts = dict(compression='gzip', compression_opts=1, chunks=(128, 128))
+                bggrp.create_dataset( 'counts', data=self.counts, **opts )
+                bggrp.create_dataset( 'variance', data=self.variance, **opts )
             elif self.format == 'polynomial':
                 raise NotImplementedError('Currently we do not support a polynomial background model. ')
                 bggrp.create_dataset( 'coeffs', data=self.counts )

tests/models/test_background.py

@@ -1,3 +1,8 @@
+import os
+
+import sep
+import time
+
 import pytest
 import numpy as np
 import h5py
@@ -72,10 +77,15 @@ def test_save_load_backgrounds(decam_raw_image, code_version):
         with pytest.raises(RuntimeError, match='Counts shape .* does not match image shape .*'):
             b2.save()
 
-        b2.counts = np.random.normal(bg_mean, 1, size=image.data.shape)
-        b2.variance = np.random.normal(bg_var, 1, size=image.data.shape)
-        b2.save()
+        # use actual background measurements so we can get a realistic estimate of the compression
+        back = sep.Background(image.data)
+        b2.counts = back.back()
+        b2.variance = back.rms() ** 2
 
+        t0 = time.perf_counter()
+        b2.save()
+        print(f'Background save time: {time.perf_counter() - t0:.3f} s')
+        print(f'Background file size: {os.path.getsize(b2.get_fullpath()) / 1024 ** 2:.3f} MB')
         # check the filename contains the provenance hash
         assert prov.id[:6] in b2.get_fullpath()