def calculate_statistics(model_name, ref_name, data, config=None):
"""
Calculates the statistics for the given model and reference variables located in the data file. Calculation will be
performed according to the provided configuration.
@param model_name: the name of the model variable.
@param ref_name: the name of the reference variable.
@param data: the input data object.
@param config: the optional configuration.
@return: a dictionary of statistics.
"""
if config is None:
config = get_default_config()
is_gridded = len(data.get_reference_dimensions(ref_name)) > 1
if is_gridded:
reference_values, model_values = data.get_values(ref_name, model_name)
unit = data.unit(model_name)
return processor.calculate_statistics(model_values, reference_values, model_name, ref_name, unit, config)
me = MatchupEngine(data, config)
matchups = me.find_all_matchups()
if config.remove_empty_matchups:
matchups = me.remove_empty_matchups(matchups)
if len(matchups) == 0:
print("No matchups found; maybe allow higher maximum time delta.")
return
unit = data.unit(model_name)
return calculate_statistics_from_matchups(matchups, model_name, ref_name, data, unit, config=None)
def create_target_diagram(statistics, config=None):
figure = plt.figure()
if config is None:
config = get_default_config()
statistics = utils.ensure_list(statistics)
diagram = TargetDiagram(figure, config.normalise_target_diagram,
config.show_legends,
config.utilise_stddev_difference)
diagram.setup_axes()
for stats in statistics:
diagram.plot_sample(
stats['bias'], stats['unbiased_rmse'], stats['normalised_rmse'],
stats['rmse'], stats['ref_stddev'], stats['stddev'],
create_sample_name(stats['model_name'], stats['unit']))
if config.normalise_target_diagram:
diagram.plot_correcoeff_marker_line()
diagram.update_legend()
if not config.normalise_target_diagram:
diagram.update_ranges(config.target_diagram_bounds)
return diagram
def __init__(self, config=None):
"""Constructs a new instance of Output.
Keyword arguments:
data -- the data object
config -- the configuration the processors and matchup engine have been run with (optional)
"""
self.config = config if config is not None else get_default_config()
self.separator = self.config.separator
def __init__(self, config=None):
"""Constructs a new instance of Output.
Keyword arguments:
data -- the data object
config -- the configuration the processors and matchup engine have been run with (optional)
"""
self.config = config if config is not None else get_default_config()
self.separator = self.config.separator
def create_taylor_diagrams(statistics, config=None):
if config is None:
config = get_default_config()
statistics = utils.ensure_list(statistics)
statistics_by_name_and_unit = sort_statistics_by_name_and_unit(
config, statistics)
diagrams = []
for name in statistics_by_name_and_unit.keys():
for current_unit in statistics_by_name_and_unit[name].keys():
current_statistics = statistics_by_name_and_unit[name][
current_unit]
ref_names = list(
map(lambda x: x.get('ref_name'), current_statistics))
ref_stddevs = list(
map(lambda x: x.get('ref_stddev'), current_statistics)
) if config.use_absolute_standard_deviation else list(
np.ones(len(current_statistics)))
units = list(map(lambda x: x.get('unit'), current_statistics))
ref = tuple(zip(ref_names, ref_stddevs, units))
model_stddevs = list(
map(lambda x: x.get('stddev'), current_statistics)
) if config.use_absolute_standard_deviation else list(
map(lambda x: x.get('normalised_stddev'), current_statistics))
max_stddev = max(ref_stddevs + model_stddevs) * 1.5
for v in ref_stddevs:
if v == 0.0 or np.isnan(v):
logging.warning(
'Unable to create Taylor diagram from statistics.')
logging.debug('Statistics: %s' % current_statistics)
continue
figure = plt.figure()
diagram = TaylorDiagram(figure, ref,
config.show_negative_corrcoeff,
config.use_absolute_standard_deviation,
config.show_legends, max_stddev)
diagram.setup_axes()
for stats in current_statistics:
model_name = stats[
'model_name'] if 'model_name' in stats else None
stddev = stats[
'stddev'] if config.use_absolute_standard_deviation else stats[
'normalised_stddev']
diagram.plot_sample(stats['corrcoeff'], stddev, model_name,
stats['unit'])
diagram.update_legend()
diagrams.append(diagram)
return diagrams
def calculate_statistics(model_values,
reference_values,
model_name=None,
ref_name=None,
unit=None,
config=None):
"""Calculate the statistics for the given reference and model arrays.
It is recommended to provide model_name and ref_name in order to allow for meaningful output.
"""
if config is None:
config = get_default_config()
model_percentiles = percentiles(model_values.ravel(), config.alpha,
config.beta)
ref_percentiles = percentiles(reference_values.ravel(), config.alpha,
config.beta)
model_minmax = minmax(model_values)
ref_minmax = minmax(reference_values)
stats = dict()
stats['model_name'] = model_name
stats['ref_name'] = ref_name
stats['unit'] = unit
stats['rmse'] = rmse(reference_values, model_values)
stats['corrcoeff'] = correlation(reference_values, model_values)
stats['unbiased_rmse'] = unbiased_rmse(reference_values, model_values)
stats['normalised_rmse'] = normalised_rmse(reference_values, model_values,
config.ddof, stats['corrcoeff'])
stats['pbias'] = percentage_model_bias(reference_values, model_values)
stats['bias'] = bias(reference_values, model_values)
stats['reliability_index'] = reliability_index(reference_values,
model_values)
stats['model_efficiency'] = model_efficiency(reference_values,
model_values)
stats['mean'] = mean(model_values)
stats['ref_mean'] = mean(reference_values)
stats['stddev'] = stddev(model_values, config.ddof)
stats['ref_stddev'] = stddev(reference_values, config.ddof)
stats['normalised_stddev'] = stats['stddev'] / stats['ref_stddev']
stats['median'] = model_percentiles[0]
stats['ref_median'] = ref_percentiles[0]
stats['p90'] = model_percentiles[1]
stats['ref_p90'] = ref_percentiles[1]
stats['p95'] = model_percentiles[2]
stats['ref_p95'] = ref_percentiles[2]
stats['min'] = model_minmax[0]
stats['ref_min'] = ref_minmax[0]
stats['max'] = model_minmax[1]
stats['ref_max'] = ref_minmax[1]
return stats
def calculate_statistics(model_values, reference_values, model_name=None, ref_name=None, unit=None, config=None):
"""Calculate the statistics for the given reference and model arrays.
It is recommended to provide model_name and ref_name in order to allow for meaningful output.
"""
if config is None:
config = get_default_config()
model_percentiles = percentiles(model_values.ravel(), config.alpha, config.beta)
ref_percentiles = percentiles(reference_values.ravel(), config.alpha, config.beta)
model_minmax = minmax(model_values)
ref_minmax = minmax(reference_values)
stats = dict()
stats['model_name'] = model_name
stats['ref_name'] = ref_name
stats['unit'] = unit
stats['rmse'] = rmse(reference_values, model_values)
stats['corrcoeff'] = correlation(reference_values, model_values)
stats['unbiased_rmse'] = unbiased_rmse(reference_values, model_values)
stats['normalised_rmse'] = normalised_rmse(reference_values, model_values, config.ddof, stats['corrcoeff'])
stats['pbias'] = percentage_model_bias(reference_values, model_values)
stats['bias'] = bias(reference_values, model_values)
stats['reliability_index'] = reliability_index(reference_values, model_values)
stats['model_efficiency'] = model_efficiency(reference_values, model_values)
stats['mean'] = mean(model_values)
stats['ref_mean'] = mean(reference_values)
stats['stddev'] = stddev(model_values, config.ddof)
stats['ref_stddev'] = stddev(reference_values, config.ddof)
stats['normalised_stddev'] = stats['stddev'] / stats['ref_stddev']
stats['median'] = model_percentiles[0]
stats['ref_median'] = ref_percentiles[0]
stats['p90'] = model_percentiles[1]
stats['ref_p90'] = ref_percentiles[1]
stats['p95'] = model_percentiles[2]
stats['ref_p95'] = ref_percentiles[2]
stats['min'] = model_minmax[0]
stats['ref_min'] = ref_minmax[0]
stats['max'] = model_minmax[1]
stats['ref_max'] = ref_minmax[1]
return stats
def create_target_diagram(statistics, config=None):
figure = plt.figure()
if config is None:
config = get_default_config()
statistics = utils.ensure_list(statistics)
diagram = TargetDiagram(figure, config.normalise_target_diagram, config.show_legends, config.utilise_stddev_difference)
diagram.setup_axes()
for stats in statistics:
diagram.plot_sample(stats['bias'], stats['unbiased_rmse'], stats['normalised_rmse'], stats['rmse'],
stats['ref_stddev'], stats['stddev'], create_sample_name(stats['model_name'], stats['unit']))
if config.normalise_target_diagram:
diagram.plot_correcoeff_marker_line()
diagram.update_legend()
if not config.normalise_target_diagram:
diagram.update_ranges(config.target_diagram_bounds)
return diagram
def create_taylor_diagrams(statistics, config=None):
if config is None:
config = get_default_config()
statistics = utils.ensure_list(statistics)
statistics_by_name_and_unit = sort_statistics_by_name_and_unit(config, statistics)
diagrams = []
for name in statistics_by_name_and_unit.keys():
for current_unit in statistics_by_name_and_unit[name].keys():
current_statistics = statistics_by_name_and_unit[name][current_unit]
ref_names = list(map(lambda x: x.get('ref_name'), current_statistics))
ref_stddevs = list(map(lambda x: x.get('ref_stddev'), current_statistics)) if config.use_absolute_standard_deviation else list(np.ones(len(current_statistics)))
units = list(map(lambda x: x.get('unit'), current_statistics))
ref = tuple(zip(ref_names, ref_stddevs, units))
model_stddevs = list(map(lambda x: x.get('stddev'), current_statistics)) if config.use_absolute_standard_deviation else list(map(lambda x: x.get('normalised_stddev'), current_statistics))
max_stddev = max(ref_stddevs + model_stddevs) * 1.5
for v in ref_stddevs:
if v == 0.0 or np.isnan(v):
logging.warning('Unable to create Taylor diagram from statistics.')
logging.debug('Statistics: %s' % current_statistics)
continue
figure = plt.figure()
diagram = TaylorDiagram(figure, ref, config.show_negative_corrcoeff, config.use_absolute_standard_deviation, config.show_legends, max_stddev)
diagram.setup_axes()
for stats in current_statistics:
model_name = stats['model_name'] if 'model_name' in stats else None
stddev = stats['stddev'] if config.use_absolute_standard_deviation else stats['normalised_stddev']
diagram.plot_sample(stats['corrcoeff'], stddev, model_name, stats['unit'])
diagram.update_legend()
diagrams.append(diagram)
return diagrams
def calculate_statistics(model_name, ref_name, data, config=None):
"""
Calculates the statistics for the given model and reference variables located in the data file. Calculation will be
performed according to the provided configuration.
@param model_name: the name of the model variable.
@param ref_name: the name of the reference variable.
@param data: the input data object.
@param config: the optional configuration.
@return: a dictionary of statistics.
"""
if config is None:
config = get_default_config()
is_gridded = len(data.get_reference_dimensions(ref_name)) > 1
if is_gridded:
reference_values, model_values = data.get_values(ref_name, model_name)
unit = data.unit(model_name)
return processor.calculate_statistics(model_values, reference_values,
model_name, ref_name, unit,
config)
me = MatchupEngine(data, config)
matchups = me.find_all_matchups()
if config.remove_empty_matchups:
matchups = me.remove_empty_matchups(matchups)
if len(matchups) == 0:
print("No matchups found; maybe allow higher maximum time delta.")
return
unit = data.unit(model_name)
return calculate_statistics_from_matchups(matchups,
model_name,
ref_name,
data,
unit,
config=None)
def __init__(self, data, configuration=None):
self.data = data
self.config = configuration if configuration is not None else get_default_config(
)
self.pixel_sizes = {}
def __init__(self, data, configuration=None):
self.data = data
self.config = configuration if configuration is not None else get_default_config()
self.pixel_sizes = {}
