def _make_query(self):
"""
Default query method implemented in the
metaclass Query().
Returns a sorted calldays table.
"""
relevant_columns = ", ".join(
get_columns_for_level(self.level, self.column_name))
sql = f"""
SELECT * FROM (
SELECT
connections.subscriber,
{relevant_columns},
COUNT(*) AS calldays
FROM (
SELECT DISTINCT locations.subscriber, {relevant_columns}, locations.time::date
FROM ({self.ul.get_query()}) AS locations
) AS connections
GROUP BY connections.subscriber, {relevant_columns}
) calldays
ORDER BY calldays.subscriber ASC, calldays.calldays DESC
"""
return sql
def __get_location_buffer(self):
"""
Protected method for generating SQL
for the buffer areas between a location
(i..e an origin) and all its possible
counterparts (i.e. destinations).
"""
cols = get_columns_for_level(self.level)
from_cols = ", ".join("{c}_from".format(c=c) for c in cols)
to_cols = ", ".join("{c}_to".format(c=c) for c in cols)
sql = """
SELECT
{froms},
{tos},
A.distance AS distance,
A.geom_origin AS geom_origin,
A.geom_destination AS geom_destination,
ST_Buffer(A.geom_destination::geography, A.distance * 1000) AS geom_buffer
FROM ({distance_matrix_table}) AS A
""".format(
distance_matrix_table=self.distance_matrix.get_query(),
froms=from_cols,
tos=to_cols,
)
return sql
def _make_query(self):
"""
Protected method that generates SQL
that calculates the population that is
covered by a buffer.
"""
cols = get_columns_for_level(self.level)
from_cols = ", ".join("B.{c}_from".format(c=c) for c in cols)
outer_from_cols = ", ".join("C.{c}_from".format(c=c) for c in cols)
to_cols = ", ".join("B.{c}_to".format(c=c) for c in cols)
outer_to_cols = ", ".join("C.{c}_to".format(c=c) for c in cols)
pop_join = " AND ".join("A.{c} = B.{c}_to".format(c=c) for c in cols)
sql = """
SELECT row_number() OVER(ORDER BY C.geom_buffer) AS id,
{froms_outer},
{tos_outer},
C.distance,
C.geom_buffer,
sum(C.destination_population) AS buffer_population,
count(*) AS n_sites
FROM
(SELECT
{froms},
{tos},
B.distance,
B.geom_buffer,
A.destination_population
FROM (
SELECT DISTINCT ON ({tos})
{tos},
B.geom_destination,
A.total AS destination_population
FROM ({population_table}) AS A
JOIN ({location_buffer_table}) AS B
ON {pop_join}
) AS A
INNER JOIN ({location_buffer_table}) AS B
ON ST_Intersects(B.geom_buffer::geography,
A.geom_destination::geography)) AS C
GROUP BY {froms_outer},
{tos_outer},
C.distance,
C.geom_buffer
""".format(
population_table=self.population_object.get_query(),
location_buffer_table=self.__get_location_buffer(),
pop_join=pop_join,
froms=from_cols,
tos=to_cols,
froms_outer=outer_from_cols,
tos_outer=outer_to_cols,
)
return sql
def index_cols(self):
"""
A list of columns to use as indexes when storing this query.
Returns
-------
ixen : list
By default, returns the location columns if they are present
and self.level is defined, and the subscriber column.
Examples
--------
>>> daily_location("2016-01-01").index_cols
[['name'], '"subscriber"']
"""
from flowmachine.utils.utils import (
get_columns_for_level, ) # Local import to avoid circular import
cols = self.column_names
ixen = []
try:
# Not all objects define the attribute column_name so we'll fall
# back to the default if it is not defined
try:
loc_cols = get_columns_for_level(self.level, self.column_name)
except AttributeError:
loc_cols = get_columns_for_level(self.level)
if set(loc_cols).issubset(cols):
ixen.append(loc_cols)
except AttributeError:
pass
try:
if self.subscriber_identifier in cols:
ixen.append(self.subscriber_identifier)
else:
ixen.append('"subscriber"')
except AttributeError:
pass
return ixen
def _make_query(self):
cols = get_columns_for_level(self.level)
sql_location_table = "SELECT * FROM infrastructure." + (
"sites" if self.level == "versioned-site" else "cells")
try:
cols.remove("lat")
cols.remove("lon")
except ValueError:
pass # Nothing to remove
from_cols = ", ".join("A.{c_id_safe} AS {c}_from".format(
c_id_safe="id" if c.endswith("id") else c, c=c) for c in cols)
to_cols = ", ".join("B.{c_id_safe} AS {c}_to".format(
c_id_safe="id" if c.endswith("id") else c, c=c) for c in cols)
return_geometry_statement = ""
if self.return_geometry:
return_geometry_statement = """
,
A.geom_point AS geom_origin,
B.geom_point AS geom_destination
"""
sql = """
SELECT
{froms},
{tos},
ST_X(A.geom_point::geometry) AS lon_from,
ST_Y(A.geom_point::geometry) AS lat_from,
ST_X(B.geom_point::geometry) AS lon_to,
ST_Y(B.geom_point::geometry) AS lat_to,
ST_Distance(
A.geom_point::geography,
B.geom_point::geography
) / 1000 AS distance
{return_geometry_statement}
FROM ({location_table_statement}) AS A
CROSS JOIN ({location_table_statement}) AS B
ORDER BY distance DESC
""".format(
location_table_statement=sql_location_table,
froms=from_cols,
tos=to_cols,
return_geometry_statement=return_geometry_statement,
)
return sql
def test_column_list():
"""Test that supplying the column name as a list returns it as a new list."""
passed_cols = ["frogs", "dogs"]
returned_cols = get_columns_for_level("admin0", passed_cols)
assert passed_cols == returned_cols
assert id(passed_cols) != id(returned_cols)
def test_columns_for_level_errors(level, column_name, error):
"""Test that get_columns_for_level raises correct errors"""
with pytest.raises(error):
get_columns_for_level(level, column_name)
def column_names(self) -> List[str]:
return (["subscriber"] +
get_columns_for_level(self.level, self.column_name) +
["calldays"])
def run(
self,
uniform_departure_rate=0.1,
departure_rate_vector=None,
ignore_missing=False,
):
"""
Runs model.
Parameters
----------
uniform_departure_rate : float
Proportion of population from location i
that will be departing in observed time period.
This proportion applies to all locations
uniformly.
departure_rate_vector : dict
A dictionary that contains the proportion
of the population from locations i that have
departed those locations. The keys of the
dictionaries must be the location identifier
and the values the departure rate.
If passed, this will be used over the
`uniform_departure_rate` parameter.
ignore_missing : bool
If True, existing locations that are not
found in the departure_rate_vector dictionary
will be computed using zero departures.
Returns
-------
A pandas dataframe with a mobility matrix.
"""
if "population_buffer" not in self.__dict__.keys():
logger.warn(" Computing Population() and DistanceMatrix() " +
"objects. This can take a few minutes.")
population_df = self.population_object.get_dataframe()
population_buffer = self.population_buffer_object.get_dataframe()
ix = get_columns_for_level(self.level)
ix = ["{}_{}".format(c, d) for d in ("from", "to") for c in ix]
population_buffer.set_index(ix, inplace=True)
M = population_df["total"].sum()
N = len(population_df[get_columns_for_level(
self.level)].drop_duplicates())
beta = 1 / M
locations = population_df[get_columns_for_level(
self.level)].values.tolist()
population_df.set_index(get_columns_for_level(self.level),
inplace=True)
if not departure_rate_vector:
logger.warn(" Using an uniform departure " +
"rate of {} for ".format(uniform_departure_rate) +
"all locations.")
elif not ignore_missing and len(departure_rate_vector) != len(
locations):
raise ValueError("Locations missing from " +
"`departure_rate_vector`. Use " +
"ignore_missing=True if locations " +
"without rates should be ignored.")
results = []
for i in locations:
sigma = 0
m_i = self.__get_population(population_df, i)
if departure_rate_vector:
try:
T_i = m_i * departure_rate_vector[i[0]]
except KeyError:
try:
T_i = m_i * departure_rate_vector[tuple(i[:1])]
except KeyError:
T_i = 0
else:
T_i = m_i * uniform_departure_rate
for k in [l for l in locations if l != i]:
m_k = self.__get_population(population_df, k)
S_ik = self.__get_buffer_population(population_buffer, i, k)
sigma += m_k * ((1 / S_ik) - beta)
for j in [l for l in locations if l != i]:
m_j = self.__get_population(population_df, j)
S_ij = self.__get_buffer_population(population_buffer, i, j)
T_ij = (T_i * m_j * ((1 / S_ij) - beta)) / sigma
if T_i != 0:
probability = T_ij / T_i
else:
probability = 0
results.append(i + j + [T_ij, probability])
ix = get_columns_for_level(self.level)
ix = ["{}_{}".format(c, d) for d in ("from", "to") for c in ix]
ix += ["prediction", "probability"]
res = pd.DataFrame(results, columns=ix)
return res