def __init__(self, service, name='', position=geopy.Point(45.4641, 9.1919),
ageDiffusionIn={}, scaleIn=1, attributesIn={}):
assert isinstance(position, geopy.Point), 'Position must be a geopy Point'
assert isinstance(service, ServiceType), 'Service must belong to the Eum'
assert isinstance(name, str), 'Name must be a string'
assert (np.isscalar(scaleIn)) & (scaleIn>0) , 'Scale must be a positive scalar'
assert isinstance(attributesIn, dict), 'Attributes can be provided in a dict'
self.name = name
self.service = service
# A ServiceType can have many sites, so each unit has its own.
# Moreover, a site is not uniquely assigned to a service
self.site = position
self.scale = scaleIn # store scale info
self.attributes = attributesIn# dictionary
# how the service availablity area varies for different age groups
if ageDiffusionIn==None:
self.ageDiffusion = {g: (
1 + .005*np.round(np.random.normal(),2))*self.scale for g in AgeGroup.all()}
else:
assert set(ageDiffusionIn.keys()) <= set(AgeGroup.all()), 'Diffusion keys should be AgeGroups'
#assert all
self.ageDiffusion = ageDiffusionIn
# define kernel taking scale into account
self.kernel = {g: gaussKern(length_scale=l*self.scale) for g, l in self.ageDiffusion.items()}
def __init__(self, mappedPositions):
assert isinstance(mappedPositions, MappedPositionsFrame), 'Expected MappedPositionsFrame'
self.mappedPositions = mappedPositions
# initialise for all service types
super().__init__({service: pd.DataFrame(
np.zeros([mappedPositions.shape[0], len(AgeGroup.all())]),
index=mappedPositions.index, columns=AgeGroup.all())
for service in ServiceType})
def compute_kpi_for_localized_services(self):
assert self.bEvaluated, 'Have you evaluated service values before making averages for KPIs?'
# get mean service levels by quartiere, weighting according to the number of citizens
for service, data in self.serviceValues.items():
checkRange = {}
for col in self.agesFrame.columns: # iterate over columns as Enums are not orderable...
if col in service.demandAges:
self.weightedValues[service][col] = pd.Series.multiply(
data[col], self.agesFrame[col])
else:
self.weightedValues[service][col] = np.nan*data[col]
checkRange = (data.groupby(common_cfg.IdQuartiereColName).min()-np.finfo(float).eps,
data.groupby(common_cfg.IdQuartiereColName).max()+np.finfo(float).eps)
# sum weighted fractions by neighbourhood
weightedSums = self.weightedValues[service].groupby(common_cfg.IdQuartiereColName).sum()
# set to NaN value the AgeGroups that have no people or there is no demand for the service
weightedSums[self.agesTotals == 0] = np.nan
weightedSums.iloc[:, ~weightedSums.columns.isin(service.demandAges)] = np.nan
self.quartiereKPI[service] = (weightedSums/self.agesTotals).reindex(
columns=AgeGroup.all(), copy=False)
# check that the weighted mean lies between min and max in the neighbourhood
for col in self.quartiereKPI[service].columns:
bGood = (self.quartiereKPI[service][col].between(
checkRange[0][col], checkRange[1][col]) | self.quartiereKPI[service][col].isnull())
assert all(bGood), 'Unexpected error in mean computation'
return self.quartiereKPI
def __init__(self, dfIn, bDuplicatesCheck=True):
assert isinstance(dfIn, pd.DataFrame), 'Input DataFrame expected'
# initialise and assign base DataFrame properties
super().__init__()
self.__dict__.update(dfIn.copy().__dict__)
# prepare the AgeGroups cardinalities
groupsCol = 'ageGroup'
peopleBySampleAge = common_cfg.fill_sample_ages_in_cpa_columns(self)
dataByGroup = peopleBySampleAge.rename(AgeGroup.find_AgeGroup, axis='columns').T
dataByGroup.index.name = groupsCol # index is now given by AgeGroup items
dataByGroup = dataByGroup.reset_index() # extract to convert to categorical and groupby
dataByGroup[groupsCol] = dataByGroup[groupsCol].astype('category')
agesBySection = dataByGroup.groupby(groupsCol).sum().T
# self['Ages'] = pd.Series(agesBySection.T.to_dict()) # assign dict to each section
self['PeopleTot'] = agesBySection.sum(axis=1)
# report all ages
for col in AgeGroup.all():
self[col] = agesBySection.get(col, np.zeros_like(self.iloc[:, 0]))
# assign centroid as position
geopyValues = self['geometry'].apply(
lambda pos: geopy.Point(pos.centroid.y, pos.centroid.x))
self[common_cfg.positionsCol] = geopyValues
if bDuplicatesCheck:
# check no location is repeated - takes a while
assert not any(self[common_cfg.positionsCol].duplicated()), 'Repeated position found'
def __init__(self, df_input, b_duplicates_check=True):
assert isinstance(df_input, pd.DataFrame), 'Input DataFrame expected'
# initialise and assign base DataFrame properties
# FIXME: this is not nice at all. Refactor to properly inherit from df
super().__init__()
self.__dict__.update(df_input.copy().__dict__)
# prepare the AgeGroups cardinalities
groups_col = 'ageGroup'
people_by_sample_age = common_cfg.fill_sample_ages_in_cpa_columns(self)
data_by_group = people_by_sample_age.rename(AgeGroup.find_age_group,
axis='columns').T
# index is now given by AgeGroup items
data_by_group.index.name = groups_col
# extract to convert to categorical and groupby
data_by_group = data_by_group.reset_index()
data_by_group[groups_col] = \
data_by_group[groups_col].astype('category')
ages_by_section = data_by_group.groupby(groups_col).sum().T
self['PeopleTot'] = ages_by_section.sum(axis=1)
# report all ages
for col in self.OUTPUT_AGES:
self[col] = ages_by_section.get(col, np.zeros_like(self.iloc[:,
0]))
# extract long and lat and build geopy locations
self[common_cfg.coord_col_names[0]] = self['geometry'].apply(
lambda pos: pos.centroid.x)
self[common_cfg.coord_col_names[1]] = self['geometry'].apply(
lambda pos: pos.centroid.y)
self[common_cfg.positions_col] = [
geopy.Point(yx)
for yx in zip(self[common_cfg.coord_col_names[::-1]].as_matrix())
]
if b_duplicates_check:
# check no location is repeated - takes a while
assert not any(self[common_cfg.positions_col].duplicated()),\
'Repeated position found'
# cache ages frame and mapped positions for quicker access
age_multi_index = [
self[common_cfg.id_quartiere_col_name],
self[common_cfg.positions_col].apply(tuple)
]
self.ages_frame = self[AgeGroup.all()].set_index(age_multi_index)
self.mapped_positions = MappedPositionsFrame(
long=self[common_cfg.coord_col_names[0]],
lat=self[common_cfg.coord_col_names[1]],
geopy_pos=self[common_cfg.positions_col].tolist(),
id_quartiere=self[common_cfg.id_quartiere_col_name].tolist())
def compute_kpi_for_istat_values(self):
allQuartiere = self.demand.groupby(common_cfg.IdQuartiereColName).sum()
dropColumns = [c for c in AgeGroup.all()+common_cfg.excludedColumns \
if c in allQuartiere.columns]
quartiereData = allQuartiere.drop(dropColumns , axis=1)
kpiFrame = istat_kpi.wrangle_istat_cpa2011(quartiereData, self.city)
self.istatKPI = kpiFrame
self.istatVitality = istat_kpi.compute_vitality_cpa2011(quartiereData)
return self.istatKPI, self.istatVitality
def load(self, meanRadius):
assert meanRadius, 'Please provide a reference radius for the mean library size'
(propertData, locations) = super().extract_locations()
nameCol = 'denominazioni.ufficiale'
typeCol = 'tipologia-funzionale'
# Modifica e specifica che per le fasce d'età
typeAgeDict = {'Specializzata': {group:1 for group in AgeGroup.all()},
'Importante non specializzata': {group:1 for group in AgeGroup.all()},
'Pubblica': {group:1 for group in AgeGroup.all()},
'NON SPECIFICATA': {AgeGroup.ChildPrimary:1},
'Scolastica': {AgeGroup.ChildPrimary:1},
'Istituto di insegnamento superiore': {AgeGroup.ChildPrimary:1},
'Nazionale': {AgeGroup.ChildPrimary:1},}
libraryTypes = propertData[typeCol].unique()
assert set(libraryTypes) <= set(typeAgeDict.keys()), 'Unrecognized types in input'
unitList = []
for libType in libraryTypes:
bThisGroup = propertData[typeCol]==libType
typeData = propertData[bThisGroup]
typeLocations = [l for i,l in enumerate(locations) if bThisGroup[i]]
for iUnit in range(typeData.shape[0]):
rowData = typeData.iloc[iUnit,:]
attrDict = {'level':libType}
thisUnit = ServiceUnit(self.servicetype,
name=rowData[nameCol],
position=typeLocations[iUnit],
ageDiffusionIn=typeAgeDict[libType],
attributesIn=attrDict)
unitList.append(thisUnit)
return unitList
def evaluate_services_at(self, mappedPositions):
assert isinstance(mappedPositions, MappedPositionsFrame), 'Expected MappedPositionsFrame'
# set all age groups as output default
outputAgeGroups = AgeGroup.all()
# initialise output with dedicated class
valuesStore = ServiceValues(mappedPositions)
# loop over different services
for thisServType in self.outputServices:
serviceUnits = [u for u in self.units if u.service == thisServType]
if not serviceUnits:
continue
else:
for thisAgeGroup in outputAgeGroups:
unitValues = np.stack(list(map(
lambda x: x.evaluate(
valuesStore.positions, thisAgeGroup), serviceUnits)), axis=-1)
# aggregate unit contributions according to the service type norm
valuesStore[thisServType][thisAgeGroup] = thisServType.aggregate_units(unitValues)
return valuesStore
def load(self, meanRadius):
assert meanRadius, 'Please provide a reference radius for urban green'
(propertData, locations) = super().extract_locations()
nameCol = 'CODICEIDENTIFICATIVOFARMACIA'
colAttributes = {'Descrizione': 'DESCRIZIONEFARMACIA', 'PartitaIva': 'PARTITAIVA'}
unitList = []
for iUnit in range(propertData.shape[0]):
rowData = propertData.iloc[iUnit, :]
attrDict = {name: rowData[col] for name, col in colAttributes.items()}
thisUnit = ServiceUnit(self.servicetype,
name=rowData[nameCol].astype(str),
position=locations[iUnit],
ageDiffusionIn={g: 1 for g in AgeGroup.all()},
scaleIn=meanRadius,
attributesIn=attrDict)
unitList.append(thisUnit)
return unitList
def make_serviceareas_output(self, precision=4):
out = dict()
# tool to format frame data that does not depend on age
def prepare_frame_data(frame_in):
frame_in = frame_in.round(precision)
orig_type = frame_in.index.dtype.type
data_dict = frame_in.reset_index().to_dict(orient='records')
# restore type as pandas has a bug and casts to float if int
for quartiere_data in data_dict:
old_value = quartiere_data[common_cfg.id_quartiere_col_name]
if orig_type in (np.int32, np.int64, int):
quartiere_data[common_cfg.id_quartiere_col_name] = int(
old_value)
return data_dict
# make istat layer
out[common_cfg.istat_layer_name] = prepare_frame_data(self.istat_data)
# make vitality layer
out[common_cfg.vitality_layer_name] = prepare_frame_data(
self.vitality_data)
# make layers
for area, layers in self.areas_tree.items():
layer_list = []
for service in layers:
data = self.layers_data[service].round(precision)
layer_list.append(
pd.Series(data[AgeGroup.all()].as_matrix().tolist(),
index=data.index,
name=service.name))
area_data = pd.concat(layer_list, axis=1).reset_index()
out[area.value] = area_data.to_dict(orient='records')
return out
def __init__(self, demandFrame, serviceUnits, cityName):
assert cityName in city_settings.cityNamesList, 'Unrecognized city name %s' % cityName
assert isinstance(demandFrame, DemandFrame),'Demand frame expected'
assert all([isinstance(su, ServiceUnit) for su in serviceUnits]),'Service units list expected'
self.city = cityName
self.demand = demandFrame
self.sources = serviceUnits
# initialise the service evaluator
self.evaluator = ServiceEvaluator(serviceUnits)
self.servicePositions = self.evaluator.servicePositions
# initialise output values
self.serviceValues = ServiceValues(self.demand.mappedPositions)
self.bEvaluated = False
self.weightedValues = ServiceValues(self.demand.mappedPositions)
self.quartiereKPI = {}
self.istatKPI = pd.DataFrame()
# derive Ages frame
ageMIndex = [demandFrame[common_cfg.IdQuartiereColName],
demandFrame[common_cfg.positionsCol].apply(tuple)]
self.agesFrame = demandFrame[AgeGroup.all()].set_index(ageMIndex)
self.agesTotals = self.agesFrame.groupby(level=0).sum()
def load(self, mean_radius=None):
assert mean_radius, 'Please provide a reference radius for pharmacies'
(propert_data, locations) = super().extract_locations()
col_attributes = {
'Descrizione': 'DESCRIZIONEFARMACIA',
'PartitaIva': 'PARTITAIVA'
}
unit_list = []
# We assume all pharmacies share the same scale, so only one
# threshold is necessary
cached_thresholds = None
for i_unit in range(propert_data.shape[0]):
row_data = propert_data.iloc[i_unit, :]
attr_dict = {
name: row_data[col]
for name, col in col_attributes.items()
}
this_unit = ServiceUnit(
self.servicetype,
name=row_data[self.name_col].astype(str),
unit_id=row_data[self.id_col],
position=locations[i_unit],
scale=mean_radius,
age_diffusion={g: 1
for g in AgeGroup.all()},
kernel_thresholds=cached_thresholds,
attributes=attr_dict)
unit_list.append(this_unit)
# if there were no thresholds, cache the computed ones
if not cached_thresholds:
cached_thresholds = this_unit.ker_thresholds
return unit_list
def __init__(self,
service,
name,
unit_id,
position,
scale,
age_diffusion=None,
kernel_thresholds=None,
attributes=None):
assert isinstance(position,
geopy.Point), 'Position must be a geopy Point'
assert isinstance(service,
ServiceType), 'Service must belong to the Eum'
assert isinstance(name, str), 'Name must be a string'
assert (np.isscalar(scale)) & \
(scale > 0), 'Scale must be a positive scalar'
assert set(age_diffusion.keys()) <= set(
AgeGroup.all()), 'Diffusion keys should be AgeGroups'
if not attributes:
attributes = {}
assert isinstance(attributes,
dict), 'Attributes have to be provided in a dict'
if kernel_thresholds:
assert set(kernel_thresholds.keys()) >= set(age_diffusion.keys()),\
'Kernel thresholds if provided must' \
' be defined for every age diffusion key'
b_thresholds_input = True
else:
b_thresholds_input = False
self.name = name
self.id = unit_id
self.service = service
# A ServiceType can have many sites, so each unit has its own.
# Moreover, a site is not uniquely assigned to a service
self.site = position
self.coord_tuple = (position.latitude, position.longitude)
self.scale = scale # store scale info
self.attributes = attributes # dictionary
# how the service availability area varies for different age groups
self.age_diffusion = age_diffusion
# define kernel taking scale into account
self.kernel = {
g: gaussKern(length_scale=l * self.scale)
for g, l in self.age_diffusion.items()
}
# precompute kernel threshold per AgeGroup
# initialise to Inf
self.ker_thresholds = {g: np.Inf for g in AgeGroup.all()}
if b_thresholds_input:
assert all([isinstance(kern, gaussKern)
for kern in self.kernel.values()]),\
'Unexpected kernel type in ServiceUnit'
assert all([val > 0 for val in kernel_thresholds.values()]), \
'Thresholds must be positive'
self.ker_thresholds.update(kernel_thresholds)
else:
self._compute_kernel_thresholds()
# initialise attendance
self.attendance = np.nan
class DemandFrame(pd.DataFrame):
"""A class to store demand units in row and
make them available for aggregation"""
OUTPUT_AGES = AgeGroup.all()
_metadata = ['ages_frame', 'mapped_positions']
def __init__(self, df_input, b_duplicates_check=True):
assert isinstance(df_input, pd.DataFrame), 'Input DataFrame expected'
# initialise and assign base DataFrame properties
# FIXME: this is not nice at all. Refactor to properly inherit from df
super().__init__()
self.__dict__.update(df_input.copy().__dict__)
# prepare the AgeGroups cardinalities
groups_col = 'ageGroup'
people_by_sample_age = common_cfg.fill_sample_ages_in_cpa_columns(self)
data_by_group = people_by_sample_age.rename(AgeGroup.find_age_group,
axis='columns').T
# index is now given by AgeGroup items
data_by_group.index.name = groups_col
# extract to convert to categorical and groupby
data_by_group = data_by_group.reset_index()
data_by_group[groups_col] = \
data_by_group[groups_col].astype('category')
ages_by_section = data_by_group.groupby(groups_col).sum().T
self['PeopleTot'] = ages_by_section.sum(axis=1)
# report all ages
for col in self.OUTPUT_AGES:
self[col] = ages_by_section.get(col, np.zeros_like(self.iloc[:,
0]))
# extract long and lat and build geopy locations
self[common_cfg.coord_col_names[0]] = self['geometry'].apply(
lambda pos: pos.centroid.x)
self[common_cfg.coord_col_names[1]] = self['geometry'].apply(
lambda pos: pos.centroid.y)
self[common_cfg.positions_col] = [
geopy.Point(yx)
for yx in zip(self[common_cfg.coord_col_names[::-1]].as_matrix())
]
if b_duplicates_check:
# check no location is repeated - takes a while
assert not any(self[common_cfg.positions_col].duplicated()),\
'Repeated position found'
# cache ages frame and mapped positions for quicker access
age_multi_index = [
self[common_cfg.id_quartiere_col_name],
self[common_cfg.positions_col].apply(tuple)
]
self.ages_frame = self[AgeGroup.all()].set_index(age_multi_index)
self.mapped_positions = MappedPositionsFrame(
long=self[common_cfg.coord_col_names[0]],
lat=self[common_cfg.coord_col_names[1]],
geopy_pos=self[common_cfg.positions_col].tolist(),
id_quartiere=self[common_cfg.id_quartiere_col_name].tolist())
def get_age_sample(self, age_group=None, n_sample=1000):
if age_group is not None:
coord, n_repeat = self.mapped_positions.align(
self.ages_frame[age_group], axis=0)
else:
coord, n_repeat = self.mapped_positions.align(
self.ages_frame.sum(axis=1), axis=0)
idx = np.repeat(range(coord.shape[0]), n_repeat)
coord = coord[common_cfg.coord_col_names].iloc[idx]
sample = coord.sample(int(n_sample)).as_matrix()
return sample[:, 0], sample[:, 1]
@classmethod
def create_from_istat_cpa(cls, city_name):
"""Constructor caller for DemandFrame"""
city_config = city_settings.get_city_config(city_name)
return cls(city_config.istat_cpa_data, b_duplicates_check=False)
def agesFrame(self):
ageMIndex = [self[common_cfg.IdQuartiereColName],
self[common_cfg.positionsCol].apply(tuple)]
return self[AgeGroup.all()].set_index(ageMIndex)