def test_emcam_lowest_bin_non_monotonic(reliability_table_slice):
"""Test expected values are returned where the lowest observation
count bin is non-monotonic."""
expected_data = np.array([[1000, 500, 500, 750], [250, 500, 750, 1000],
[2000, 1000, 1000, 1000]])
expected_bin_coord_points = np.array([0.2, 0.5, 0.7, 0.9],
dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.4], [0.4, 0.6], [0.6, 0.8], [0.8, 1.0]],
dtype=np.float32,
)
reliability_table_slice.data = np.array([
[1000, 0, 250, 500, 750], # Observation count
[0, 250, 500, 750, 1000], # Sum of forecast probability
[1000, 1000, 1000, 1000, 1000], # Forecast count
])
result = Plugin()._enforce_min_count_and_montonicity(
reliability_table_slice.copy())
assert_array_equal(result.data, expected_data)
assert_allclose(
result.coord("probability_bin").points, expected_bin_coord_points)
assert_allclose(
result.coord("probability_bin").bounds, expected_bin_coord_bounds)
def test_emcam_combine_undersampled_bins_non_monotonic(
reliability_table_slice):
"""Test expected values are returned when a bin is below the minimum
forecast count when the observed frequency is non-monotonic."""
expected_data = np.array([[1000, 425, 1000], [1000, 425, 1000],
[2000, 600, 1000]])
expected_bin_coord_points = np.array([0.2, 0.6, 0.9], dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.4], [0.4, 0.8], [0.8, 1.0]],
dtype=np.float32,
)
reliability_table_slice.data = np.array(
[
[750, 250, 50, 375, 1000], # Observation count
[750, 250, 50, 375, 1000], # Sum of forecast probability
[1000, 1000, 100, 500, 1000], # Forecast count
],
dtype=np.float32,
)
result = Plugin()._enforce_min_count_and_montonicity(
reliability_table_slice.copy())
assert_array_equal(result.data, expected_data)
assert_allclose(
result.coord("probability_bin").points, expected_bin_coord_points)
assert_allclose(
result.coord("probability_bin").bounds, expected_bin_coord_bounds)
def setUp(self):
"""Set up monotonic bins as default and plugin for testing."""
super().setUp()
self.obs_count = np.array([0, 250, 500, 750, 1000], dtype=np.float32)
self.forecast_probability_sum = np.array([0, 250, 500, 750, 1000],
dtype=np.float32)
self.plugin = Plugin()
def process(
reliability_table: cli.inputcube,
*,
minimum_forecast_count: int = 200,
):
"""
Manipulate a reliability table to ensure sufficient sample counts in
as many bins as possible by combining bins with low sample counts.
Also enforces a monotonic observation frequency.
Args:
reliability_table (iris.cube.Cube):
The reliability table that needs to be manipulated after the
spatial dimensions have been aggregated.
minimum_forecast_count (int):
The minimum number of forecast counts in a forecast probability
bin for it to be used in calibration.
The default value of 200 is that used in Flowerdew 2014.
Returns:
iris.cube.CubeList:
The reliability table that has been manipulated to ensure
sufficient sample counts in each probability bin and a monotonic
observation frequency.
The cubelist contains a separate cube for each threshold in
the original reliability table.
"""
from improver.calibration.reliability_calibration import ManipulateReliabilityTable
plugin = ManipulateReliabilityTable(
minimum_forecast_count=minimum_forecast_count)
return plugin(reliability_table)
class Test__combine_undersampled_bins(Test_setup):
"""Test the _combine_undersampled_bins method."""
def setUp(self):
"""Set up monotonic bins as default and plugin for testing."""
super().setUp()
self.obs_count = np.array([0, 250, 500, 750, 1000], dtype=np.float32)
self.forecast_probability_sum = np.array([0, 250, 500, 750, 1000],
dtype=np.float32)
self.plugin = Plugin()
def test_no_undersampled_bins(self):
"""Test no bins are combined when no bins are under-sampled."""
forecast_count = np.array([1000, 1000, 1000, 1000, 1000],
dtype=np.float32)
result = self.plugin._combine_undersampled_bins(
self.obs_count,
self.forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(
result[:3],
[self.obs_count, self.forecast_probability_sum, forecast_count])
self.assertEqual(result[3], self.probability_bin_coord)
def test_poorly_sampled_bins(self):
"""Test when all bins are poorly sampled and the minimum forecast count
cannot be reached."""
obs_count = np.array([0, 2, 5, 8, 10], dtype=np.float32)
forecast_probability_sum = np.array([0, 2, 5, 8, 10], dtype=np.float32)
forecast_count = np.array([10, 10, 10, 10, 10], dtype=np.float32)
expected = np.array([
[25], # Observation count
[25], # Sum of forecast probability
[50], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.5], dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_one_undersampled_bin_at_top(self):
"""Test when the highest probability bin is under-sampled."""
obs_count = np.array([0, 250, 500, 750, 100], dtype=np.float32)
forecast_probability_sum = np.array([0, 250, 500, 750, 100],
dtype=np.float32)
forecast_count = np.array([1000, 1000, 1000, 1000, 100],
dtype=np.float32)
expected = np.array([
[0, 250, 500, 850], # Observation count
[0, 250, 500, 850], # Sum of forecast probability
[1000, 1000, 1000, 1100], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.1, 0.3, 0.5, 0.8],
dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.2], [0.2, 0.4], [0.4, 0.6], [0.6, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_one_undersampled_bin_at_bottom(self):
"""Test when the lowest probability bin is under-sampled."""
forecast_count = np.array([100, 1000, 1000, 1000, 1000],
dtype=np.float32)
expected = np.array([
[250, 500, 750, 1000], # Observation count
[250, 500, 750, 1000], # Sum of forecast probability
[1100, 1000, 1000, 1000], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
self.obs_count,
self.forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.2, 0.5, 0.7, 0.9],
dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.4], [0.4, 0.6], [0.6, 0.8], [0.8, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_one_undersampled_bin_lower_neighbour(self):
"""Test for one under-sampled bin that is combined with its lower
neighbour."""
obs_count = np.array([0, 250, 50, 1500, 1000], dtype=np.float32)
forecast_probability_sum = np.array([0, 250, 50, 1500, 1000],
dtype=np.float32)
forecast_count = np.array([1000, 1000, 100, 2000, 1000],
dtype=np.float32)
expected = np.array([
[0, 300, 1500, 1000], # Observation count
[0, 300, 1500, 1000], # Sum of forecast probability
[1000, 1100, 2000, 1000], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.1, 0.4, 0.7, 0.9],
dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.2], [0.2, 0.6], [0.6, 0.8], [0.8, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_one_undersampled_bin_upper_neighbour(self):
"""Test for one under-sampled bin that is combined with its upper
neighbour."""
obs_count = np.array([0, 500, 50, 750, 1000], dtype=np.float32)
forecast_probability_sum = np.array([0, 500, 50, 750, 1000],
dtype=np.float32)
forecast_count = np.array([1000, 2000, 100, 1000, 1000],
dtype=np.float32)
expected = np.array([
[0, 500, 800, 1000], # Observation count
[0, 500, 800, 1000], # Sum of forecast probability
[1000, 2000, 1100, 1000], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.1, 0.3, 0.6, 0.9],
dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.2], [0.2, 0.4], [0.4, 0.8], [0.8, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_two_undersampled_bins(self):
"""Test when two bins are under-sampled."""
obs_count = np.array([0, 12, 250, 75, 250], dtype=np.float32)
forecast_probability_sum = np.array([0, 12, 250, 75, 250],
dtype=np.float32)
forecast_count = np.array([1000, 50, 500, 100, 250], dtype=np.float32)
expected = np.array([
[0, 262, 325], # Observation count
[0, 262, 325], # Sum of forecast probability
[1000, 550, 350], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.1, 0.4, 0.8], dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.2], [0.2, 0.6], [0.6, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_two_equal_undersampled_bins(self):
"""Test when two bins are under-sampled and the under-sampled bins have
an equal forecast count."""
obs_count = np.array([0, 25, 250, 75, 250], dtype=np.float32)
forecast_probability_sum = np.array([0, 25, 250, 75, 250],
dtype=np.float32)
forecast_count = np.array([1000, 100, 500, 100, 250], dtype=np.float32)
expected = np.array([
[0, 275, 325], # Observation count
[0, 275, 325], # Sum of forecast probability
[1000, 600, 350], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.1, 0.4, 0.8], dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.2], [0.2, 0.6], [0.6, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)
def test_three_equal_undersampled_bin_neighbours(self):
"""Test when three neighbouring bins are under-sampled."""
obs_count = np.array([0, 25, 50, 75, 250], dtype=np.float32)
forecast_probability_sum = np.array([0, 25, 50, 75, 250],
dtype=np.float32)
forecast_count = np.array([1000, 100, 100, 100, 250], dtype=np.float32)
expected = np.array([
[0, 150, 250], # Observation count
[0, 150, 250], # Sum of forecast probability
[1000, 300, 250], # Forecast count
])
result = self.plugin._combine_undersampled_bins(
obs_count,
forecast_probability_sum,
forecast_count,
self.probability_bin_coord,
)
assert_array_equal(result[:3], expected)
expected_bin_coord_points = np.array([0.1, 0.5, 0.9], dtype=np.float32)
expected_bin_coord_bounds = np.array(
[[0.0, 0.2], [0.2, 0.8], [0.8, 1.0]],
dtype=np.float32,
)
assert_allclose(expected_bin_coord_points, result[3].points)
assert_allclose(expected_bin_coord_bounds, result[3].bounds)