def test_create_nice_predictor_names_ninth(self):
"""Ensures correct output from create_nice_predictor_names.
In this case, using ninth set of inputs.
"""
these_names_by_matrix = permutation.create_nice_predictor_names(
predictor_matrices=NINTH_PREDICTOR_MATRICES,
cnn_metadata_dict=NINTH_METADATA_DICT)
these_predictor_names = sum(these_names_by_matrix, [])
self.assertTrue(these_predictor_names == NINTH_NICE_PREDICTOR_NAMES)
def test_create_nice_predictor_names_eighth(self):
"""Ensures correct output from create_nice_predictor_names.
In this case, using eighth set of inputs.
"""
these_names_by_matrix = permutation.create_nice_predictor_names(
predictor_matrices=EIGHTH_PREDICTOR_MATRICES,
cnn_metadata_dict=EIGHTH_METADATA_DICT,
separate_radar_heights=EIGHTH_SEPARATE_HEIGHTS_FLAG)
these_predictor_names = sum(these_names_by_matrix, [])
self.assertTrue(these_predictor_names == EIGHTH_NICE_PREDICTOR_NAMES)
def get_pearson_correlations(predictor_matrices,
cnn_metadata_dict,
separate_radar_heights=False):
"""Computes Pearson correlation between each pair of predictors.
P = total number of predictors (over all matrices)
:param predictor_matrices: See doc for
`permutation.create_nice_predictor_names`.
:param cnn_metadata_dict: Same.
:param separate_radar_heights: Same.
:return: correlation_matrix: P-by-P numpy array of Pearson correlations.
:return: predictor_names: length-P list of predictor names.
"""
error_checking.assert_is_boolean(separate_radar_heights)
first_num_dimensions = len(predictor_matrices[0].shape)
separate_radar_heights = (separate_radar_heights
and first_num_dimensions == 5)
predictor_names_by_matrix = permutation.create_nice_predictor_names(
predictor_matrices=predictor_matrices,
cnn_metadata_dict=cnn_metadata_dict,
separate_radar_heights=separate_radar_heights)
num_matrices = len(predictor_names_by_matrix)
for i in range(num_matrices):
print('Predictors in {0:d}th matrix:\n{1:s}\n'.format(
i + 1, str(predictor_names_by_matrix[i])))
print(SEPARATOR_STRING)
predictor_matrices_to_use = copy.deepcopy(predictor_matrices)
if separate_radar_heights:
predictor_matrices_to_use[0] = permutation_utils.flatten_last_two_dim(
predictor_matrices_to_use[0])[0]
num_predictors_by_matrix = numpy.array(
[len(n) for n in predictor_names_by_matrix], dtype=int)
predictor_names = sum(predictor_names_by_matrix, [])
num_predictors = len(predictor_names)
correlation_matrix = numpy.full((num_predictors, num_predictors),
numpy.nan)
for i in range(num_predictors):
for j in range(i, num_predictors):
if i == j:
correlation_matrix[i, j] = 1.
continue
i_matrix, i_channel = _linear_idx_to_matrix_channel_idxs(
linear_index=i,
num_predictors_by_matrix=num_predictors_by_matrix)
j_matrix, j_channel = _linear_idx_to_matrix_channel_idxs(
linear_index=j,
num_predictors_by_matrix=num_predictors_by_matrix)
these_first_values = _take_spatial_mean(
predictor_matrices_to_use[i_matrix][..., i_channel])
these_second_values = _take_spatial_mean(
predictor_matrices_to_use[j_matrix][..., j_channel])
correlation_matrix[i, j] = pearsonr(these_first_values,
these_second_values)[0]
correlation_matrix[j, i] = correlation_matrix[i, j]
return correlation_matrix, predictor_names
