def statistics(outputs, weights, params):
"""
:param outputs:
An instance of `AssetsIndividualOutputs`
See `classical` for a description of `params`
:returns:
an instance of `StatisticalOutputs`
It makes use of `..base.statistics` to compute curves and maps
"""
ret = []
# traverse the curve matrix on the second dimension (the assets)
# accumulating results in `ret`, then return `ret` unzipped
for loss_ratio_curves in outputs.curve_matrix.transpose(1, 0, 2, 3):
# get the loss ratios only from the first curve
loss_ratios, _poes = loss_ratio_curves[0]
curves_poes = [poes for _losses, poes in loss_ratio_curves]
mean_curve, quantile_curves, mean_maps, quantile_maps = (
base.asset_statistics(
loss_ratios, curves_poes,
params.quantiles, weights, params.conditional_loss_poes))
# compute also mean and quantile loss fractions
mean_fractions = [
conditional_loss_ratio(mean_curve[0], mean_curve[1], poe)
for poe in params.poes_disagg]
quantile_fractions = [[
conditional_loss_ratio(quantile_curve[0], quantile_curve[1], poe)
for poe in params.poes_disagg]
for quantile_curve in quantile_curves]
ret.append((mean_curve, mean_maps, mean_fractions,
quantile_curves, quantile_maps, quantile_fractions))
(mean_curve, mean_maps, mean_fractions,
quantile_curves, quantile_maps, quantile_fractions) = zip(*ret)
# now all the lists keep N items
# transpose maps and fractions to have P/F/Q items of N-sized lists
mean_maps = numpy.array(mean_maps).transpose()
mean_fractions = numpy.array(mean_fractions).transpose()
quantile_curves = numpy.array(quantile_curves).transpose(1, 0, 2, 3)
quantile_maps = numpy.array(quantile_maps).transpose(2, 1, 0)
quantile_fractions = numpy.array(quantile_fractions).transpose(2, 1, 0)
return StatisticalOutputs(
outputs.assets, mean_curve, mean_maps,
mean_fractions, quantile_curves, quantile_maps, quantile_fractions)
def statistics(assets, curve_matrix, weights, params):
ret = []
for curves in curve_matrix:
non_trivial_curves = [(losses, poes)
for losses, poes in curves if losses[-1] > 0]
if not non_trivial_curves: # no damage. all trivial curves
loss_ratios, _poes = curves[0]
curves_poes = [poes for _losses, poes in curves]
else: # standard case
max_losses = [losses[-1] # we assume non-decreasing losses
for losses, _poes in non_trivial_curves]
reference_curve = non_trivial_curves[numpy.argmax(max_losses)]
loss_ratios = reference_curve[0]
curves_poes = [interpolate.interp1d(
losses, poes, bounds_error=False, fill_value=0)(loss_ratios)
for losses, poes in curves]
mean_curve, quantile_curves, mean_maps, quantile_maps = (
base.asset_statistics(
loss_ratios, curves_poes,
params.quantiles, weights, params.conditional_loss_poes))
ret.append((mean_curve, mean_maps, quantile_curves, quantile_maps))
(mean_curves, mean_maps, quantile_curves, quantile_maps) = zip(*ret)
# now all the lists keep N items
# transpose maps and fractions to have P/F/Q items of N-sized lists
mean_maps = numpy.array(mean_maps).transpose()
if (len(quantile_curves) and len(quantile_curves[0])):
quantile_curves = numpy.array(quantile_curves).transpose(1, 0, 2, 3)
else:
quantile_curves = None
if (len(quantile_maps) and len(quantile_maps[0])):
quantile_maps = numpy.array(quantile_maps).transpose(2, 1, 0)
else:
quantile_maps = None
return StatisticalOutputs(
assets, mean_curves, mean_maps, quantile_curves, quantile_maps)
