def test_normal_ppf_compare_with_scipy(self):
"""Compare the percentiles function to the scipy ppf function."""
from helpers import metrics
cdf_fun = norm.cdf
# PDF parameters
mean = 0
standard_deviation = 1
# Percentiles to estimate
percentiles = np.arange(1, 99, 1) / 100
# Percentile search parameters
search_start_value = -50
search_increment_value = 0.01
search_probability_threshold = 0.001
# Do comparison
target_percentiles = \
norm.ppf(percentiles, loc=mean, scale=standard_deviation)
comparison_percentiles = \
metrics.percentiles(
cdf_fun, percentiles,
search_start_value, search_increment_value,
search_probability_threshold)
assert_almost_equal(target_percentiles, comparison_percentiles, 2)
def add_model_predictions(dataframe, model, test_observation,
persistence_forecast, row_index,
element_name, percentiles):
"""Add model predictions to the dataframe."""
# Local constants
ZERO_DEGREES_KELVIN = 273.15
THRESHOLDS = np.arange(-30, 30, 0.5) + 273.15
# Ensemble PDF
dataframe.loc[row_index, element_name + '_ENSEMBLE_PDF'] = \
model.pdf(test_observation)
# Ensemble CDF
dataframe.loc[row_index, element_name + '_ENSEMBLE_CDF'] = \
model.cdf(test_observation)
# Ensemble CDF freezing
dataframe.loc[row_index, element_name + '_ENSEMBLE_CDF_FREEZE'] = \
model.cdf(ZERO_DEGREES_KELVIN)
# CRPS
threshold_cdfs = model.cdf(THRESHOLDS)
dataframe.loc[row_index, element_name + '_CRPS'] = \
metrics.crps(THRESHOLDS, threshold_cdfs, test_observation)
# Ensemble mean
dataframe.loc[row_index, element_name + '_ENSEMBLE_MEAN'] = \
model.mean()
# Persistence forecast
dataframe.loc[row_index, element_name + '_PERSISTENCE_FORECAST'] = \
persistence_forecast
# Observation rank
dataframe.loc[row_index, element_name + '_OBS_RANK'] = \
metrics.rank(test_observation, model.get_member_means())
# Percentiles
search_initialization = test_observation - 15
perc_values = metrics.percentiles(
model.cdf, percentiles / 100, search_initialization)
for percentile, value in zip(percentiles, perc_values):
name = element_name + '_ENSEMBLE_PERC' + str(percentile)
dataframe.loc[row_index, name] = value