Skip to content

Lazy data in PP save pairs #2452

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
cpelley merged 1 commit into from
Apr 25, 2017
Merged

Lazy data in PP save pairs #2452

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions lib/iris/etc/pp_save_rules.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -761,12 +761,12 @@ THEN

#MDI
IF
isinstance(cm.data, ma.core.MaskedArray)
cm.fill_value is not None
THEN
pp.bmdi = cm.data.fill_value
pp.bmdi = cm.fill_value

IF
not isinstance(cm.data, ma.core.MaskedArray)
cm.fill_value == None
THEN
pp.bmdi = -1e30

Expand Down
27 changes: 24 additions & 3 deletions lib/iris/fileformats/pp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
# (C) British Crown Copyright 2010 - 2016, Met Office
# (C) British Crown Copyright 2010 - 2017, Met Office
#
# This file is part of Iris.
#
Expand Down Expand Up @@ -2322,6 +2322,18 @@ def as_pairs(cube, field_coords=None, target=None):
target=target)


def _data_fill_value(cube):
# Function to deduce a fill_value for a cube's data.
# This should eventually be superceded by a cube 'fill_value' property.
if cube.has_lazy_data():
fill_value = cube.lazy_data().fill_value
elif isinstance(cube.data, ma.MaskedArray):
fill_value = cube.data.fill_value
else:
fill_value = None
return fill_value


def save_pairs_from_cube(cube, field_coords=None, target=None):
"""
Use the PP saving rules (and any user rules) to convert a cube or
Expand Down Expand Up @@ -2404,6 +2416,11 @@ def save_pairs_from_cube(cube, field_coords=None, target=None):

# Save each named or latlon slice2D in the cube
for slice2D in cube.slices(field_coords):
# Attach an extra cube "fill_value" property, allowing the save rules
# to deduce MDI more easily without realising the data.
# NOTE: it is done this way because this property may exist in future.
slice2D.fill_value = _data_fill_value(slice2D)
Copy link

@cpelley cpelley Apr 20, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Adding a fill_value attribute at run-time seems rather messy to me especially as it presumes that the dask branch ends up merged onto master as it is. Perhaps it would be better to point this PR to the dask branch instead?
Saying that I'm not going to push this, If your happy then I'm happy.

Copy link
Contributor

@djkirkham djkirkham Apr 21, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Perhaps this could be avoided by making the code in _data_fill_value() into a set of save rules to determine pp.bmdi.

Copy link
Member Author

@pp-mo pp-mo Apr 25, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a purely temporary measure, based on the idea that the forthcoming dask change will definitely be introducing the generic "cube.fill_value" property.
We will remove this hack when the dask branch is merged back.
Obviously, the method actually used here is 🤢 but it is only here so you guys can check that it fixes the current memory burn problem.


# Start with a blank PPField
pp_field = PPField3()

Expand All @@ -2426,8 +2443,12 @@ def save_pairs_from_cube(cube, field_coords=None, target=None):
# From UM doc F3: "Set to -99 if LBEGIN not known"
pp_field.lbuser[1] = -99

# Set the data
pp_field.data = slice2D.data
# Set the data, keeping it lazy where possible.
if slice2D.has_lazy_data():
slice_core_data = slice2D.lazy_data()
else:
slice_core_data = slice2D.data
Copy link

@cpelley cpelley Apr 20, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this really necessary? Can't you simply avoid the if-else and instead simply always set it to the lazy array i.e. I doubt there would be much overhead in re-instantiating a lazy array when it has been realised.

i.e. pp_field._data = slice2D.lazy_data()

Copy link
Member Author

@pp-mo pp-mo Apr 25, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's best like this because creating a simple wrapper around real data is substantially more costly in dask than it was with biggus : The earliest dask branch retained existing cube property code to get cube.shape and cube.dtype from cube.lazy_data(), and then the tests took nearly twice as long!
Having said that, this code should be removed when we transition to dask anyway, as mentioned above

pp_field._data = slice_core_data

# Run the PP save rules on the slice2D, to fill the PPField,
# recording the rules that were used
Expand Down
39 changes: 20 additions & 19 deletions lib/iris/tests/integration/test_pp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,16 @@
import iris.util


def run_save_rules(cube, field):
# Add an extra "fill_value" property, as required by PP save rules, and
# now normally supplied by pp.save_pairs_from_cube.
cube.fill_value = iris.fileformats.pp._data_fill_value(cube)
# Ensure save rules are ready.
iris.fileformats.pp._ensure_save_rules_loaded()
# Run save rules to populate the field from the cube
iris.fileformats.pp._save_rules.verify(cube, field)


class TestVertical(tests.IrisTest):
def _test_coord(self, cube, point, bounds=None, **kwargs):
coords = cube.coords(**kwargs)
Expand Down Expand Up @@ -69,8 +79,7 @@ def test_soil_level_round_trip(self):
field.lbvc = 0
field.brsvd = [None] * 4
field.brlev = None
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)

# Check the vertical coordinate is as originally specified.
self.assertEqual(field.lbvc, 6)
Expand Down Expand Up @@ -103,8 +112,7 @@ def test_soil_depth_round_trip(self):
field.lbvc = 0
field.brlev = None
field.brsvd = [None] * 4
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)

# Check the vertical coordinate is as originally specified.
self.assertEqual(field.lbvc, 6)
Expand Down Expand Up @@ -132,8 +140,7 @@ def test_potential_temperature_level_round_trip(self):
field = iris.fileformats.pp.PPField3()
field.lbfc = 0
field.lbvc = 0
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)

# Check the vertical coordinate is as originally specified.
self.assertEqual(field.lbvc, 19)
Expand Down Expand Up @@ -197,15 +204,14 @@ def field_with_data(scale=1):
pressure_field.lbvc = 0
pressure_field.brsvd = [None, None]
pressure_field.lbuser = [None] * 7
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(pressure_cube, pressure_field)
run_save_rules(pressure_cube, pressure_field)

data_field = iris.fileformats.pp.PPField3()
data_field.lbfc = 0
data_field.lbvc = 0
data_field.brsvd = [None, None]
data_field.lbuser = [None] * 7
iris.fileformats.pp._save_rules.verify(data_cube, data_field)
run_save_rules(data_cube, data_field)

# The reference surface field should have STASH=409
self.assertArrayEqual(pressure_field.lbuser,
Expand Down Expand Up @@ -285,8 +291,7 @@ def test_hybrid_height_with_non_standard_coords(self):
field.lbvc = 0
field.brsvd = [None, None]
field.lbuser = [None] * 7
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)

self.assertEqual(field.blev, delta)
self.assertEqual(field.brlev, delta_lower)
Expand Down Expand Up @@ -322,8 +327,7 @@ def test_hybrid_pressure_with_non_standard_coords(self):
field.lbvc = 0
field.brsvd = [None, None]
field.lbuser = [None] * 7
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)

self.assertEqual(field.bhlev, delta)
self.assertEqual(field.bhrlev, delta_lower)
Expand Down Expand Up @@ -384,8 +388,7 @@ def field_with_data(scale=1):
data_field.lbvc = 0
data_field.brsvd = [None, None]
data_field.lbuser = [None] * 7
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(data_cube, data_field)
run_save_rules(data_cube, data_field)

# Check the data field has the vertical coordinate as originally
# specified.
Expand Down Expand Up @@ -421,8 +424,7 @@ def convert_cube_to_field(self, cube):
field.lbfc = 0
field.lbvc = 0
field.lbtim = 0
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)
return field

def test_time_mean_from_forecast_period(self):
Expand Down Expand Up @@ -608,8 +610,7 @@ def create_cube(self, longitude_coord='longitude'):
def convert_cube_to_field(self, cube):
field = iris.fileformats.pp.PPField3()
field.lbvc = 0
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, field)
run_save_rules(cube, field)
return field

def test_time_mean_only(self):
Expand Down
4 changes: 2 additions & 2 deletions lib/iris/tests/results/cube_to_pp/append_single.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@
bdy: -2.5
bzx: -3.75
bdx: 3.75
bmdi: -1e+30
bmdi: 9999.0
bmks: 1.0
data: [[ 254.64399719 254.64399719 254.64399719 ..., 254.64399719
254.64399719 254.64399719]
Expand Down Expand Up @@ -115,7 +115,7 @@
bdy: -2.5
bzx: -3.75
bdx: 3.75
bmdi: -1e+30
bmdi: 9999.0
bmks: 1.0
data: [[ 254.64399719 254.64399719 254.64399719 ..., 254.64399719
254.64399719 254.64399719]
Expand Down
2 changes: 1 addition & 1 deletion lib/iris/tests/results/cube_to_pp/replace_single.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@
bdy: -2.5
bzx: -3.75
bdx: 3.75
bmdi: -1e+30
bmdi: 9999.0
bmks: 1.0
data: [[ 254.64399719 254.64399719 254.64399719 ..., 254.64399719
254.64399719 254.64399719]
Expand Down
2 changes: 1 addition & 1 deletion lib/iris/tests/results/cube_to_pp/simple.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@
bdy: -2.5
bzx: -3.75
bdx: 3.75
bmdi: -1e+30
bmdi: 9999.0
bmks: 1.0
data: [[ 254.64399719 254.64399719 254.64399719 ..., 254.64399719
254.64399719 254.64399719]
Expand Down
2 changes: 1 addition & 1 deletion lib/iris/tests/results/cube_to_pp/user_rules.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@
bdy: -2.5
bzx: -3.75
bdx: 3.75
bmdi: -1e+30
bmdi: 9999.0
bmks: 1.0
data: [[ 254.64399719 254.64399719 254.64399719 ..., 254.64399719
254.64399719 254.64399719]
Expand Down
12 changes: 6 additions & 6 deletions lib/iris/tests/test_cube_to_pp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
# (C) British Crown Copyright 2010 - 2016, Met Office
# (C) British Crown Copyright 2010 - 2017, Met Office
#
# This file is part of Iris.
#
Expand Down Expand Up @@ -28,12 +28,10 @@
import cf_units
import numpy as np

import iris
import iris.coords
import iris.coord_systems
import iris.fileformats.pp as ff_pp
import iris.fileformats.pp
from iris.fileformats.pp import PPField3
import iris.io
from iris.tests import mock
import iris.tests.pp as pp
import iris.util
Expand Down Expand Up @@ -207,12 +205,14 @@ def test_365_calendar_export(self):
new_unit = cf_units.Unit('hours since 1970-01-01 00:00:00',
calendar=cf_units.CALENDAR_365_DAY)
cube.coord('time').units = new_unit
# Add an extra "fill_value" property, as used by the save rules.
cube.fill_value = None
pp_field = mock.MagicMock(spec=PPField3)
iris.fileformats.pp._ensure_save_rules_loaded()
iris.fileformats.pp._save_rules.verify(cube, pp_field)
self.assertEqual(pp_field.lbtim.ic, 4)


class FakePPEnvironment(object):
''' fake a minimal PP environment for use in cross-section coords, as in PP save rules '''
y = [1, 2, 3, 4]
Expand Down Expand Up @@ -340,7 +340,7 @@ def fields_from_cube(cubes):
fh.seek(0)

# load in the saved pp fields and check the appropriate metadata
for field in ff_pp.load(tmp_file.name):
for field in iris.fileformats.pp.load(tmp_file.name):
yield field


Expand Down
11 changes: 10 additions & 1 deletion lib/iris/tests/unit/fileformats/pp/test_as_pairs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
# (C) British Crown Copyright 2015, Met Office
# (C) British Crown Copyright 2015 - 2017, Met Office
#
# This file is part of Iris.
#
Expand Down Expand Up @@ -48,6 +48,15 @@ def test_field_coords(self):
self.assertEqual(aslice.shape, (11, 9))
self.assertEqual(field.lbcode, 101)

def test_lazy_data(self):
cube = self.cube.copy()
# "Rebase" the cube onto a lazy version of its data.
cube.lazy_data(cube.lazy_data())
# Check that lazy data is preserved in save-pairs generation.
slices_and_fields = pp.as_pairs(cube)
for aslice, _ in slices_and_fields:
self.assertTrue(aslice.has_lazy_data())


if __name__ == "__main__":
tests.main()