def _selector(name, data_in, data_out, setting):
d = dsp.Dispatcher(
name='%s selector' % name,
description='Select the calibrated %s.' % name,
)
errors, setting = [], setting or {}
_sort_models = setting.pop('sort_models', sort_models)
if 'weights' in setting:
_weights = dsp_utl.map_list(setting['targets'], *setting.pop('weights'))
else:
_weights = None
_get_best_model = functools.partial(
setting.pop('get_best_model', get_best_model),
models_wo_err=setting.pop('models_wo_err', None),
selector_id=d.name
)
d.add_data(
data_id='selector_settings',
default_value={})
node_ids = ['error_settings', 'best_model_settings']
d.add_function(
function=functools.partial(define_selector_settings, node_ids=node_ids),
inputs=['selector_settings'],
outputs=node_ids
)
for i in data_in:
e = 'error/%s' % i
errors.append(e)
d.add_function(
function=_errors(name, i, data_out, setting),
inputs=['error_settings', i] + [k for k in data_out if k != i],
outputs=[e]
)
d.add_function(
function_id='sort_models',
function=functools.partial(_sort_models, weights=_weights),
inputs=errors,
outputs=['rank']
)
d.add_function(
function_id='get_best_model',
function=_get_best_model,
inputs=['rank', 'best_model_settings'],
outputs=['model', 'errors']
)
return dsp_utl.SubDispatch(d, outputs=['model', 'errors'],
output_type='list')
def _comparison2df(comparison):
res = {}
it = co2_utl.stack_nested_keys(comparison, depth=3)
keys = ['usage', 'cycle', 'param']
gen = [(dsp_utl.map_list(keys, *k), k, v) for k, v in it]
for s, k, v in _yield_sorted_params(gen, keys=keys):
l = co2_utl.get_nested_dicts(res, *k[:-1], default=list)
l.append(dsp_utl.combine_dicts({'param_id': k[-1]}, v))
if res:
return _dd2df(res, 'param_id', depth=2)
def selector(*data):
"""
Defines the models' selector model.
.. dispatcher:: dsp
>>> dsp = selector()
:return:
The models' selector model.
:rtype: SubDispatchFunction
"""
data = data or ('wltp_h', 'wltp_l')
dsp = Dispatcher(
name='Models selector',
description='Select the calibrated models.',
)
dsp.add_function(function=partial(dsp_utl.map_list, data),
inputs=data,
outputs=['CO2MPAS_results'])
dsp.add_data(data_id='models', function=combine_outputs, wait_inputs=True)
dsp.add_data(data_id='scores', function=combine_scores, wait_inputs=True)
setting = sub_models()
dsp.add_data(data_id='error_settings', default_value={})
m = list(setting)
dsp.add_function(function=partial(split_error_settings, m),
inputs=['error_settings'],
outputs=['error_settings/%s' % k for k in m])
for k, v in setting.items():
v['dsp'] = v.pop('define_sub_model', define_sub_model)(**v)
v['metrics'] = dsp_utl.map_list(v['targets'], *v['metrics'])
dsp.add_function(function=v.pop('model_selector', _selector)(k, data,
data, v),
function_id='%s selector' % k,
inputs=['CO2MPAS_results',
'error_settings/%s' % k],
outputs=['models', 'scores'])
func = dsp_utl.SubDispatchFunction(dsp=dsp,
function_id='models_selector',
inputs=('error_settings', ) + data,
outputs=['models', 'scores'])
return func
def _format_scores(scores):
res = {}
for k, j in co2_utl.stack_nested_keys(scores, depth=3):
if k[-1] in ('limits', 'errors'):
model_id = k[0]
extra_field = ('score',) if k[-1] == 'errors' else ()
for i, v in co2_utl.stack_nested_keys(j):
i = (model_id, i[-1], k[1],) + i[:-1] + extra_field
co2_utl.get_nested_dicts(res, *i, default=co2_utl.ret_v(v))
sco = {}
for k, v in co2_utl.stack_nested_keys(res, depth=4):
v.update(dsp_utl.map_list(['model_id', 'param_id'], *k[:2]))
co2_utl.get_nested_dicts(sco, *k[2:], default=list).append(v)
return sco
def _summary2df(data):
res = []
summary = data.get('summary', {})
if 'results' in summary:
r = {}
index = ['cycle', 'stage', 'usage']
for k, v in dsp_utl.stack_nested_keys(summary['results'], depth=4):
l = dsp_utl.get_nested_dicts(r, k[0], default=list)
l.append(dsp_utl.combine_dicts(dsp_utl.map_list(index, *k[1:]), v))
if r:
df = _dd2df(r,
index=index,
depth=2,
col_key=functools.partial(_sort_key,
p_keys=('param', ) * 2),
row_key=functools.partial(_sort_key, p_keys=index))
df.columns = pd.MultiIndex.from_tuples(_add_units(df.columns))
setattr(df, 'name', 'results')
res.append(df)
if 'selection' in summary:
df = _dd2df(summary['selection'], ['model_id'],
depth=2,
col_key=functools.partial(_sort_key,
p_keys=('stage', 'cycle')),
row_key=functools.partial(_sort_key, p_keys=()))
setattr(df, 'name', 'selection')
res.append(df)
if 'comparison' in summary:
r = {}
for k, v in dsp_utl.stack_nested_keys(summary['comparison'], depth=3):
v = dsp_utl.combine_dicts(v, base={'param_id': k[-1]})
dsp_utl.get_nested_dicts(r, *k[:-1], default=list).append(v)
if r:
df = _dd2df(r, ['param_id'],
depth=2,
col_key=functools.partial(_sort_key,
p_keys=('stage', 'cycle')),
row_key=functools.partial(_sort_key, p_keys=()))
setattr(df, 'name', 'comparison')
res.append(df)
if res:
return {'summary': res}
return {}
def _error(name, data_id, data_out, setting):
d = dsp.Dispatcher(
name='%s-%s error vs %s' % (name, data_id, data_out),
description='Calculates the error of calibrated model of a reference.',
)
default_settings = {
'inputs_map': {},
'targets': [],
'metrics_inputs': {},
'up_limit': None,
'dn_limit': None
}
default_settings.update(setting)
it = dsp_utl.selector(['up_limit', 'dn_limit'], default_settings).items()
for k, v in it:
if v is not None:
default_settings[k] = dsp_utl.map_list(setting['targets'], *v)
d.add_function(
function_id='select_inputs',
function=dsp_utl.map_dict,
inputs=['inputs_map', 'data'],
outputs=['inputs<0>']
)
d.add_function(
function_id='select_inputs',
function=functools.partial(dsp_utl.selector, allow_miss=True),
inputs=['inputs', 'inputs<0>'],
outputs=['inputs<1>']
)
d.add_function(
function=dsp_utl.combine_dicts,
inputs=['calibrated_models', 'inputs<1>'],
outputs=['prediction_inputs']
)
d.add_function(
function_id='select_targets',
function=functools.partial(dsp_utl.selector, allow_miss=True),
inputs=['targets', 'data'],
outputs=['references']
)
d.add_function(
function=functools.partial(
default_settings.pop('dsp', lambda x: x), {}
),
inputs=['prediction_inputs', 'calibrated_models'],
outputs=['results']
)
d.add_function(
function_id='select_outputs',
function=select_outputs,
inputs=['outputs', 'targets', 'results'],
outputs=['predictions']
)
d.add_function(
function_id='select_metrics_inputs',
function=functools.partial(dsp_utl.selector, allow_miss=True),
inputs=['metrics_inputs', 'data'],
outputs=['metrics_args']
)
d.add_function(
function=make_metrics,
inputs=['metrics', 'references', 'predictions', 'metrics_args'],
outputs=['errors']
)
d.add_function(
function=check_limits,
inputs=['errors', 'up_limit', 'dn_limit'],
outputs=['status']
)
for k, v in default_settings.items():
d.add_data(k, v)
func = dsp_utl.SubDispatch(
dsp=d,
outputs=['errors', 'status'],
output_type='list'
)
return func