def plot_error_dist_many_heights(error_matrix, heights_m_agl,
min_error_to_plot, max_error_to_plot,
axes_object):
"""Plots error distribution at each height on the same axes.
E = number of examples
H = number of heights
:param error_matrix: E-by-H numpy array of signed errors.
:param heights_m_agl: length-H numpy array of heights (metres above ground
level).
:param min_error_to_plot: Lower limit for x-axis.
:param max_error_to_plot: Upper limit for x-axis.
:param axes_object: Will plot on these axes (instance of
`matplotlib.axes._subplots.AxesSubplot`).
"""
error_checking.assert_is_numpy_array_without_nan(error_matrix)
error_checking.assert_is_numpy_array(error_matrix, num_dimensions=2)
num_heights = error_matrix.shape[1]
error_checking.assert_is_geq_numpy_array(heights_m_agl, 0.)
error_checking.assert_is_numpy_array(heights_m_agl,
exact_dimensions=numpy.array(
[num_heights], dtype=int))
error_checking.assert_is_greater(max_error_to_plot, min_error_to_plot)
boxplot_style_dict = {'color': 'k', 'linewidth': 2}
y_values = numpy.linspace(0, num_heights - 1, num=num_heights, dtype=float)
for j in range(num_heights):
axes_object.boxplot(error_matrix[:, j],
widths=0.9,
vert=False,
notch=False,
sym='o',
whis=(5, 95),
medianprops=boxplot_style_dict,
boxprops=boxplot_style_dict,
whiskerprops=boxplot_style_dict,
capprops=boxplot_style_dict,
positions=y_values[[j]])
axes_object.set_xlim(min_error_to_plot, max_error_to_plot)
y_tick_strings = profile_plotting.create_height_labels(
tick_values_km_agl=METRES_TO_KM * heights_m_agl, use_log_scale=False)
for j in range(len(y_tick_strings)):
if numpy.mod(j, 5) == 0:
continue
y_tick_strings[j] = ' '
axes_object.set_yticklabels(y_tick_strings)
axes_object.set_ylabel('Height (km AGL)')
def test_create_height_labels_linear(self):
"""Ensures correct output from create_height_labels.
In this case, assuming that height axis is linear.
"""
these_tick_strings = profile_plotting.create_height_labels(
tick_values_km_agl=TICK_VALUES_KM_AGL, use_log_scale=False)
self.assertTrue(these_tick_strings == TICK_STRINGS_LINEAR_SCALE)
def _plot_one_example(
gradcam_dict, example_index, model_metadata_dict, use_log_scale,
title_string, output_dir_name):
"""Plots class-activation map for one example.
:param gradcam_dict: Dictionary read by `gradcam.read_file`.
:param example_index: Will plot class-activation map for example with this
array index.
:param model_metadata_dict: Dictionary read by `neural_net.read_metafile`.
:param use_log_scale: See documentation at top of file.
:param title_string: Figure title.
:param output_dir_name: Name of output directory. Figure will be saved
here.
"""
figure_object, axes_object = pyplot.subplots(
1, 1, figsize=(FIGURE_WIDTH_INCHES, FIGURE_HEIGHT_INCHES)
)
if use_log_scale:
axes_object.set_yscale('log')
class_activations = (
gradcam_dict[gradcam.CLASS_ACTIVATIONS_KEY][example_index, ...]
)
example_id_string = gradcam_dict[gradcam.EXAMPLE_IDS_KEY][example_index]
generator_option_dict = model_metadata_dict[neural_net.TRAINING_OPTIONS_KEY]
heights_km_agl = METRES_TO_KM * (
generator_option_dict[neural_net.HEIGHTS_KEY]
)
axes_object.plot(
class_activations, heights_km_agl, color=LINE_COLOUR,
linewidth=LINE_WIDTH
)
y_tick_strings = profile_plotting.create_height_labels(
tick_values_km_agl=axes_object.get_yticks(), use_log_scale=use_log_scale
)
axes_object.set_yticklabels(y_tick_strings)
axes_object.set_xlabel('Class activation')
axes_object.set_ylabel('Height (km AGL)')
axes_object.set_title(title_string, fontsize=TITLE_FONT_SIZE)
output_file_name = '{0:s}/{1:s}.jpg'.format(
output_dir_name, example_id_string.replace('_', '-')
)
print('Saving figure to: "{0:s}"...'.format(output_file_name))
figure_object.savefig(
output_file_name, dpi=FIGURE_RESOLUTION_DPI, pad_inches=0,
bbox_inches='tight'
)
pyplot.close(figure_object)
def _plot_saliency_one_example(saliency_dict, example_index,
model_metadata_dict, colour_map_object,
max_colour_percentile, output_dir_name):
"""Plots saliency maps for one example.
:param saliency_dict: Dictionary read by `saliency.read_all_targets_file`.
:param example_index: Will plot saliency maps for example with this array
index.
:param model_metadata_dict: Dictionary read by `neural_net.read_metafile`.
:param colour_map_object: See documentation at top of file.
:param max_colour_percentile: Same.
:param output_dir_name: Same.
"""
generator_option_dict = model_metadata_dict[
neural_net.TRAINING_OPTIONS_KEY]
scalar_predictor_names = (
generator_option_dict[neural_net.SCALAR_PREDICTOR_NAMES_KEY])
scalar_target_names = (
generator_option_dict[neural_net.SCALAR_TARGET_NAMES_KEY])
vector_predictor_names = (
generator_option_dict[neural_net.VECTOR_PREDICTOR_NAMES_KEY])
vector_target_names = (
generator_option_dict[neural_net.VECTOR_TARGET_NAMES_KEY])
heights_km_agl = (METRES_TO_KM *
generator_option_dict[neural_net.HEIGHTS_KEY])
height_labels = profile_plotting.create_height_labels(
tick_values_km_agl=heights_km_agl, use_log_scale=False)
height_labels = [
height_labels[k] if numpy.mod(k, 4) == 0 else ' '
for k in range(len(height_labels))
]
example_id_string = saliency_dict[saliency.EXAMPLE_IDS_KEY][example_index]
this_matrix = (
saliency_dict[saliency.SALIENCY_SCALAR_P_SCALAR_T_KEY][example_index,
...])
is_dense_net = len(this_matrix.shape) == 2
if this_matrix.size > 0:
if is_dense_net:
_plot_saliency_scalar_p_scalar_t(
saliency_matrix=this_matrix,
predictor_names=scalar_predictor_names,
target_names=scalar_target_names,
example_id_string=example_id_string,
colour_map_object=colour_map_object,
max_colour_percentile=max_colour_percentile,
output_dir_name=output_dir_name)
else:
_plot_saliency_vector_p_scalar_t(
saliency_matrix=this_matrix,
predictor_names=scalar_predictor_names,
target_names=scalar_target_names,
height_labels=height_labels,
example_id_string=example_id_string,
colour_map_object=colour_map_object,
max_colour_percentile=max_colour_percentile,
output_dir_name=output_dir_name)
this_matrix = (
saliency_dict[saliency.SALIENCY_VECTOR_P_SCALAR_T_KEY][example_index,
...])
if this_matrix.size > 0:
_plot_saliency_vector_p_scalar_t(
saliency_matrix=this_matrix,
predictor_names=vector_predictor_names,
target_names=scalar_target_names,
height_labels=height_labels,
example_id_string=example_id_string,
colour_map_object=colour_map_object,
max_colour_percentile=max_colour_percentile,
output_dir_name=output_dir_name)
this_matrix = (
saliency_dict[saliency.SALIENCY_SCALAR_P_VECTOR_T_KEY][example_index,
...])
if this_matrix.size > 0:
if is_dense_net:
_plot_saliency_scalar_p_vector_t(
saliency_matrix=this_matrix,
predictor_names=scalar_predictor_names,
target_names=vector_target_names,
height_labels=height_labels,
example_id_string=example_id_string,
colour_map_object=colour_map_object,
max_colour_percentile=max_colour_percentile,
output_dir_name=output_dir_name)
else:
_plot_saliency_vector_p_vector_t(
saliency_matrix=this_matrix,
predictor_names=scalar_predictor_names,
target_names=vector_target_names,
height_labels=height_labels,
example_id_string=example_id_string,
colour_map_object=colour_map_object,
max_colour_percentile=max_colour_percentile,
output_dir_name=output_dir_name)
this_matrix = (
saliency_dict[saliency.SALIENCY_VECTOR_P_VECTOR_T_KEY][example_index,
...])
if this_matrix.size > 0:
_plot_saliency_vector_p_vector_t(
saliency_matrix=this_matrix,
predictor_names=vector_predictor_names,
target_names=vector_target_names,
height_labels=height_labels,
example_id_string=example_id_string,
colour_map_object=colour_map_object,
max_colour_percentile=max_colour_percentile,
output_dir_name=output_dir_name)
def plot_taylor_diagram_many_heights(
target_stdevs,
prediction_stdevs,
correlations,
heights_m_agl,
figure_object,
colour_map_object=DEFAULT_HEIGHT_CMAP_OBJECT):
"""Plots Taylor diagram for many heights on the same axes.
Each point should be for the same variable, just at different
heights.
H = number of heights
:param target_stdevs: length-H numpy array with standard deviations of
target (actual) values.
:param prediction_stdevs: length-H numpy array with standard deviations of
predicted values.
:param correlations: length-H numpy array of correlations.
:param heights_m_agl: length-H numpy array of heights (metres above ground
level).
:param figure_object: Will plot on this figure (instance of
`matplotlib.figure.Figure`).
:param colour_map_object: Colour map (instance of `matplotlib.pyplot.cm` or
similar). Will be used to colour points in Taylor diagram by height.
:return: taylor_diagram_object: Handle for Taylor diagram (instance of
`taylor_diagram.TaylorDiagram`).
"""
error_checking.assert_is_geq_numpy_array(target_stdevs, 0.)
error_checking.assert_is_numpy_array(target_stdevs, num_dimensions=1)
num_heights = len(target_stdevs)
expected_dim = numpy.array([num_heights], dtype=int)
error_checking.assert_is_geq_numpy_array(prediction_stdevs, 0.)
error_checking.assert_is_numpy_array(prediction_stdevs,
exact_dimensions=expected_dim)
error_checking.assert_is_geq_numpy_array(correlations, -1., allow_nan=True)
error_checking.assert_is_leq_numpy_array(correlations, 1., allow_nan=True)
error_checking.assert_is_numpy_array(correlations,
exact_dimensions=expected_dim)
error_checking.assert_is_geq_numpy_array(heights_m_agl, 0.)
error_checking.assert_is_numpy_array(heights_m_agl,
exact_dimensions=expected_dim)
heights_km_agl = heights_m_agl * METRES_TO_KM
colour_norm_object = matplotlib.colors.LogNorm(
vmin=numpy.min(heights_km_agl), vmax=numpy.max(heights_km_agl))
mean_target_stdev = numpy.mean(target_stdevs)
this_ratio = numpy.maximum(
numpy.max(target_stdevs),
numpy.max(prediction_stdevs)) / mean_target_stdev
taylor_diagram_object = taylor_diagram.TaylorDiagram(
refstd=mean_target_stdev,
fig=figure_object,
srange=(0, this_ratio),
extend=False,
plot_reference_line=False)
this_marker_object = taylor_diagram_object.samplePoints[0]
this_marker_object.set_visible(False)
for j in range(num_heights):
if numpy.isnan(correlations[j]):
continue
this_colour = colour_map_object(colour_norm_object(heights_km_agl[j]))
taylor_diagram_object.add_sample(stddev=target_stdevs[j], corrcoef=1.)
this_marker_object = taylor_diagram_object.samplePoints[-1]
this_marker_object.set_marker(TAYLOR_TARGET_MARKER_TYPE)
this_marker_object.set_markersize(TAYLOR_TARGET_MARKER_SIZE)
this_marker_object.set_markerfacecolor(this_colour)
this_marker_object.set_markeredgewidth(0)
taylor_diagram_object.add_sample(stddev=prediction_stdevs[j],
corrcoef=correlations[j])
this_marker_object = taylor_diagram_object.samplePoints[-1]
this_marker_object.set_marker(TAYLOR_PREDICTION_MARKER_TYPE)
this_marker_object.set_markersize(TAYLOR_PREDICTION_MARKER_SIZE)
this_marker_object.set_markerfacecolor(this_colour)
this_marker_object.set_markeredgewidth(0)
crmse_contour_object = taylor_diagram_object.add_contours(levels=5,
colors='0.5')
pyplot.clabel(crmse_contour_object, inline=1, fmt='%.0f')
taylor_diagram_object.add_grid()
taylor_diagram_object._ax.axis[:].major_ticks.set_tick_out(True)
colour_bar_object = plotting_utils.plot_colour_bar(
axes_object_or_matrix=figure_object.axes[0],
data_matrix=heights_km_agl,
colour_map_object=colour_map_object,
colour_norm_object=colour_norm_object,
orientation_string='vertical',
extend_min=False,
extend_max=False,
fraction_of_axis_length=0.85,
font_size=FONT_SIZE)
tick_values = colour_bar_object.get_ticks()
tick_strings = profile_plotting.create_height_labels(
tick_values_km_agl=tick_values, use_log_scale=True)
colour_bar_object.set_ticks(tick_values)
colour_bar_object.set_ticklabels(tick_strings)
colour_bar_object.set_label('Height (km AGL)', fontsize=FONT_SIZE)
return taylor_diagram_object
def plot_rel_curve_many_heights(mean_target_matrix,
mean_prediction_matrix,
heights_m_agl,
min_value_to_plot,
max_value_to_plot,
axes_object,
colour_map_object=DEFAULT_HEIGHT_CMAP_OBJECT):
"""Plots reliability curves for many heights on the same axes.
Reliability curves should be for the same variable, just at different
heights.
B = number of forecast bins
H = number of heights
:param mean_target_matrix: H-by-B numpy array of mean target (actual)
values.
:param mean_prediction_matrix: H-by-B numpy array of mean predicted values.
:param heights_m_agl: length-H numpy array of heights (metres above ground
level).
:param min_value_to_plot: Minimum value to plot (for both x- and y-axes).
:param max_value_to_plot: Max value to plot (for both x- and y-axes).
:param axes_object: Will plot on these axes (instance of
`matplotlib.axes._subplots.AxesSubplot`).
:param colour_map_object: Colour map (instance of `matplotlib.pyplot.cm` or
similar). Will be used to colour reliability curves by height.
"""
error_checking.assert_is_numpy_array(mean_target_matrix, num_dimensions=2)
error_checking.assert_is_numpy_array(mean_prediction_matrix,
exact_dimensions=numpy.array(
mean_target_matrix.shape,
dtype=int))
num_heights = mean_target_matrix.shape[0]
error_checking.assert_is_geq_numpy_array(heights_m_agl, 0.)
error_checking.assert_is_numpy_array(heights_m_agl,
exact_dimensions=numpy.array(
[num_heights], dtype=int))
error_checking.assert_is_greater(max_value_to_plot, min_value_to_plot)
heights_km_agl = heights_m_agl * METRES_TO_KM
colour_norm_object = matplotlib.colors.LogNorm(
vmin=numpy.min(heights_km_agl), vmax=numpy.max(heights_km_agl))
for j in range(num_heights):
this_colour = colour_map_object(colour_norm_object(heights_km_agl[j]))
_plot_reliability_curve(axes_object=axes_object,
mean_predictions=mean_prediction_matrix[j, :],
mean_observations=mean_target_matrix[j, :],
line_colour=this_colour,
min_value_to_plot=min_value_to_plot,
max_value_to_plot=max_value_to_plot)
axes_object.set_xlim(min_value_to_plot, max_value_to_plot)
axes_object.set_ylim(min_value_to_plot, max_value_to_plot)
colour_bar_object = plotting_utils.plot_colour_bar(
axes_object_or_matrix=axes_object,
data_matrix=heights_km_agl,
colour_map_object=colour_map_object,
colour_norm_object=colour_norm_object,
orientation_string='vertical',
extend_min=False,
extend_max=False,
font_size=FONT_SIZE)
tick_values = colour_bar_object.get_ticks()
tick_strings = profile_plotting.create_height_labels(
tick_values_km_agl=tick_values, use_log_scale=True)
colour_bar_object.set_ticks(tick_values)
colour_bar_object.set_ticklabels(tick_strings)
colour_bar_object.set_label('Height (km AGL)', fontsize=FONT_SIZE)
def plot_score_profile(heights_m_agl,
score_values,
score_name,
line_colour,
line_width,
use_log_scale,
axes_object,
are_axes_new,
line_style='solid'):
"""Plots vertical profile of one score.
H = number of heights
:param heights_m_agl: length-H numpy array of heights (metres above ground
level).
:param score_values: length-H numpy array with values of score.
:param score_name: Name of score (must be accepted by `_check_score_name`).
:param line_colour: Line colour (in any format accepted by matplotlib).
:param line_width: Line width (in any format accepted by matplotlib).
:param use_log_scale: Boolean flag. If True, will plot height (y-axis) in
logarithmic scale. If False, will plot height in linear scale.
:param axes_object: Will plot on these axes (instance of
`matplotlib.axes._subplots.AxesSubplot`).
:param are_axes_new: Boolean flag. If True, axes are new (do not already
contain plot).
:param line_style: Line style (in any format accepted by matplotlib).
:return: main_line_handle: Handle for main line (score profile).
"""
error_checking.assert_is_numpy_array(heights_m_agl, num_dimensions=1)
error_checking.assert_is_geq_numpy_array(heights_m_agl, 0.)
num_heights = len(heights_m_agl)
error_checking.assert_is_numpy_array(score_values,
exact_dimensions=numpy.array(
[num_heights], dtype=int))
_check_score_name(score_name)
error_checking.assert_is_boolean(use_log_scale)
error_checking.assert_is_boolean(are_axes_new)
if are_axes_new:
orig_x_limits = []
orig_y_limits = []
else:
orig_x_limits = axes_object.get_xlim()
orig_y_limits = axes_object.get_ylim()
if use_log_scale:
axes_object.set_yscale('log')
heights_km_agl = heights_m_agl * METRES_TO_KM
min_height_km_agl = numpy.min(heights_km_agl)
max_height_km_agl = numpy.max(heights_km_agl)
if not are_axes_new:
min_height_km_agl = numpy.minimum(min_height_km_agl, orig_y_limits[0])
max_height_km_agl = numpy.maximum(max_height_km_agl, orig_y_limits[1])
skill_score_names = [MAE_SKILL_SCORE_NAME, MSE_SKILL_SCORE_NAME]
possibly_negative_score_names = (skill_score_names +
[BIAS_NAME, CORRELATION_NAME, KGE_NAME])
if score_name in possibly_negative_score_names:
reference_x_coords = numpy.full(2, 0.)
reference_y_coords = numpy.array([0, max_height_km_agl], dtype=float)
axes_object.plot(reference_x_coords,
reference_y_coords,
color=REFERENCE_LINE_COLOUR,
linestyle='dashed',
linewidth=REFERENCE_LINE_WIDTH)
finite_indices = numpy.where(numpy.isfinite(score_values))[0]
main_line_handle = axes_object.plot(score_values[finite_indices],
heights_km_agl[finite_indices],
color=line_colour,
linestyle=line_style,
linewidth=line_width)[0]
x_max = numpy.maximum(numpy.max(score_values[finite_indices]), 0.)
if not are_axes_new:
x_max = numpy.maximum(x_max, orig_x_limits[1])
if score_name in possibly_negative_score_names:
x_min = numpy.minimum(numpy.min(score_values[finite_indices]), 0.)
if not are_axes_new:
x_min = numpy.minimum(x_min, orig_x_limits[0])
if score_name in skill_score_names:
x_min = numpy.maximum(x_min, -1.)
else:
x_min = 0.
axes_object.set_xlim(x_min, x_max)
if use_log_scale:
axes_object.set_ylim(min_height_km_agl, max_height_km_agl)
else:
axes_object.set_ylim(0, max_height_km_agl)
y_tick_strings = profile_plotting.create_height_labels(
tick_values_km_agl=axes_object.get_yticks(),
use_log_scale=use_log_scale)
axes_object.set_yticklabels(y_tick_strings)
axes_object.set_ylabel('Height (km AGL)')
return main_line_handle
def _plot_gradcam_one_example(
gradcam_dict, example_index, model_metadata_dict, colour_map_object,
max_colour_percentile, output_dir_name):
"""Plots class-activation map for one example, all target variables.
:param gradcam_dict: Dictionary read by `gradcam.read_all_targets_file`.
:param example_index: Will plot class-activation maps for example with this
array index.
:param model_metadata_dict: Dictionary read by `neural_net.read_metafile`.
:param colour_map_object: See documentation at top of file.
:param max_colour_percentile: Same.
:param output_dir_name: Same.
"""
generator_option_dict = model_metadata_dict[neural_net.TRAINING_OPTIONS_KEY]
target_names = generator_option_dict[neural_net.VECTOR_TARGET_NAMES_KEY]
target_names_verbose = [
TARGET_NAME_TO_VERBOSE[n] for n in target_names
]
heights_km_agl = (
METRES_TO_KM * generator_option_dict[neural_net.HEIGHTS_KEY]
)
height_labels = profile_plotting.create_height_labels(
tick_values_km_agl=heights_km_agl, use_log_scale=False
)
height_labels = [
height_labels[k] if numpy.mod(k, 4) == 0 else ' '
for k in range(len(height_labels))
]
example_id_string = gradcam_dict[gradcam.EXAMPLE_IDS_KEY][example_index]
class_activation_matrix_3d = (
gradcam_dict[gradcam.CLASS_ACTIVATIONS_KEY][example_index, ...]
)
num_targets = len(target_names)
num_heights = len(height_labels)
for k in range(num_targets):
class_activation_matrix_2d = class_activation_matrix_3d[..., k]
max_colour_value = numpy.percentile(
class_activation_matrix_2d, max_colour_percentile
)
max_colour_value = numpy.maximum(max_colour_value, 0.001)
figure_object, axes_object = pyplot.subplots(
1, 1, figsize=(FIGURE_WIDTH_INCHES, FIGURE_HEIGHT_INCHES)
)
axes_object.imshow(
numpy.transpose(class_activation_matrix_2d),
cmap=colour_map_object, vmin=0., vmax=max_colour_value,
origin='lower'
)
tick_values = numpy.linspace(
0, num_heights - 1, num=num_heights, dtype=float
)
axes_object.set_xticks(tick_values)
axes_object.set_yticks(tick_values)
axes_object.set_xticklabels(
height_labels, fontsize=TICK_LABEL_FONT_SIZE, rotation=90.
)
axes_object.set_yticklabels(
height_labels, fontsize=TICK_LABEL_FONT_SIZE
)
axes_object.set_xlabel('Predictor height (km AGL)')
axes_object.set_ylabel('Target height (km AGL)')
axes_object.plot(
axes_object.get_xlim(), axes_object.get_ylim(),
color=REFERENCE_LINE_COLOUR, linestyle='dashed',
linewidth=REFERENCE_LINE_WIDTH
)
colour_bar_object = plotting_utils.plot_linear_colour_bar(
axes_object_or_matrix=axes_object,
data_matrix=class_activation_matrix_2d,
colour_map_object=colour_map_object,
min_value=0., max_value=max_colour_value,
orientation_string='horizontal', padding=0.1,
extend_min=False, extend_max=True,
fraction_of_axis_length=0.8, font_size=DEFAULT_FONT_SIZE
)
tick_values = colour_bar_object.get_ticks()
tick_strings = ['{0:.1f}'.format(v) for v in tick_values]
colour_bar_object.set_ticks(tick_values)
colour_bar_object.set_ticklabels(tick_strings)
title_string = 'Class-activation map for {0:s}'.format(
target_names_verbose[k]
)
axes_object.set_title(title_string, fontsize=DEFAULT_FONT_SIZE)
output_file_name = '{0:s}/{1:s}_{2:s}.jpg'.format(
output_dir_name, example_id_string.replace('_', '-'),
target_names[k].replace('_', '-')
)
print('Saving figure to: "{0:s}"...'.format(output_file_name))
figure_object.savefig(
output_file_name, dpi=FIGURE_RESOLUTION_DPI, pad_inches=0,
bbox_inches='tight'
)
pyplot.close(figure_object)