for tb, sb in itertools.product([30, 50, 70, 90], [3, 4, 5, 6, 7]):
weight = open_cp.prohotspot.ClassicWeight(time_bandwidth=tb,
space_bandwidth=sb)
distance = open_cp.prohotspot.DistanceCircle()
state.add_prediction(scripted.ProHotspotProvider(weight, distance),
time_range)
for tb, sb in itertools.product([30, 50, 70, 90], [0.5, 1, 1.5]):
weight = open_cp.prohotspot.ClassicWeight(time_bandwidth=tb,
space_bandwidth=sb)
state.add_prediction(scripted.ProHotspotCtsProvider(weight, 150),
time_range)
for max_s in [100, 200, 500, 1000]:
for max_t in [7, 14, 21, 8 * 7]:
provider = scripted.STScanProvider(max_s,
datetime.timedelta(days=max_t),
False)
state.add_prediction(provider, time_range)
state.add_prediction(provider.with_new_max_cluster(True),
time_range)
for tw in [12, 16, 20]:
for bw in [50, 75, 100]:
tk = open_cp.kde.ExponentialTimeKernel(tw * 7)
sk = open_cp.kde.GaussianFixedBandwidthProvider(bw)
state.add_prediction(scripted.KDEProvider(tk, sk), time_range)
state.score(scripted.HitCountEvaluator)
state.process(scripted.HitCountSave("hotspots2.csv"))
state.timed_points,
tk,
sk,
bk,
cutoff=time_range.first)
state.add_prediction_range(provider, time_range)
for omega in [0.1, 0.5, 1]:
for sigma in [20, 50, 100]:
tk = sepp.sepp_fixed.ExpTimeKernel(omega)
sk = sepp.sepp_fixed.GaussianSpaceKernel(sigma)
provider = predictors.GridFixedTriggerProvider(
state.grid,
state.timed_points,
tk,
sk,
cutoff=time_range.first)
state.add_prediction_range(provider, time_range)
for omega in [0.1, 0.5, 1]:
for sigma in [20, 50, 100]:
tk = sepp.sepp_fixed.ExpTimeKernel(omega)
sk = sepp.sepp_fixed.GaussianSpaceKernel(sigma)
bk = sepp.kernels.FixedBandwidthKernelProvider(50)
provider = predictors.KDEFixedTriggerProvider(
state.timed_points, tk, sk, bk, cutoff=time_range.first)
state.add_prediction_range(provider, time_range)
state.score(scripted.HitCountEvaluator)
state.process(scripted.HitCountSave("{}_all.csv".format(side)))
return open_cp.sources.chicago.load(f, "BURGLARY", type="all")
def load_geometry():
"""Load the geometry for Chicago; we'll use Southside, as ever..."""
open_cp.sources.chicago.set_data_directory(datadir)
return open_cp.sources.chicago.get_side("South")
## Run
import datetime
with scripted.Data(load_points,
load_geometry,
start=datetime.datetime(2016, 1, 1)) as state:
time_range = scripted.TimeRange(datetime.datetime(2016, 10, 1),
datetime.datetime(2017, 1, 1),
datetime.timedelta(days=1))
provider = scripted.STScanProvider(500, datetime.timedelta(days=21), False)
state.add_prediction(provider, time_range)
state.add_prediction(provider.with_new_max_cluster(True), time_range)
state.score(scripted.HitCountEvaluator)
state.save_predictions("stscan_preds.pic.xz")
state.process(scripted.HitCountSave("stscan_counts.csv"))
return open_cp.sources.chicago.load(f, "BURGLARY", type="all")
def load_geometry():
"""Load the geometry for Chicago; we'll use Southside, as ever..."""
open_cp.sources.chicago.set_data_directory(datadir)
return open_cp.sources.chicago.get_side("South")
## Run
import datetime
with scripted.Data(load_points,
load_geometry,
start=datetime.datetime(2016, 1, 1)) as state:
time_range = scripted.TimeRange(datetime.datetime(2016, 10, 1),
datetime.datetime(2017, 1, 1),
datetime.timedelta(days=1))
state.add_prediction(scripted.NaiveProvider, time_range)
state.add_prediction(scripted.ScipyKDEProvider, time_range)
state.score(scripted.HitRateEvaluator)
state.score(scripted.HitCountEvaluator)
state.save_predictions("naive_preds.pic.xz")
state.process(scripted.HitRateSave("naive.csv"))
state.process(scripted.HitCountSave("naive_counts.csv"))
time_range = scripted.TimeRange(datetime.datetime(2016, 10, 1),
datetime.datetime(2017, 1, 1),
datetime.timedelta(days=1))
state.add_prediction(scripted.NaiveProvider, time_range)
for omegai in [1, 2, 5, 10, 20, 100]:
for sigma in [10, 20, 50, 100]:
tk = sepp.sepp_fixed.ExpTimeKernel(1 / omegai)
sk = sepp.sepp_fixed.GaussianSpaceKernel(sigma)
provider = predictors.GridFixedTriggerProvider(
state.grid,
state.timed_points,
tk,
sk,
cutoff=time_range.first)
state.add_prediction_range(provider, time_range)
for omegai in [1, 2, 5, 10, 20, 100]:
for sigma in [10, 20, 50, 100]:
for bkb in [20, 50, 100, 200]:
tk = sepp.sepp_fixed.ExpTimeKernel(1 / omegai)
sk = sepp.sepp_fixed.GaussianSpaceKernel(sigma)
bk = sepp.kernels.FixedBandwidthKernelProvider(bkb)
provider = predictors.KDEFixedTriggerProvider(
state.timed_points, tk, sk, bk, cutoff=time_range.first)
state.add_prediction_range(provider, time_range)
state.score(scripted.HitCountEvaluator)
state.process(scripted.HitCountSave("{}_mod_fixed.csv".format(side)))
def load_geometry():
frame = gpd.read_file("SouthSide")
return frame.geometry[0]
def stringToDatetime(s):
return datetime.datetime(*(s.split("-")))
earliest_time = "2016-01-01"
start_time = "2016-10-01"
end_time = "2017-01-01"
# Perform the predictions; see `scripted_intro.md`
with scripted.Data(load_points,
load_geometry,
start=stringToDatetime(earliest_time)) as state:
time_range = scripted.TimeRange(stringToDatetime(start_time),
stringToDatetime(end_time),
datetime.timedelta(days=1))
state.add_prediction(scripted.NaiveProvider, time_range)
state.score(scripted.HitRateEvaluator)
state.score(scripted.HitCountEvaluator)
state.process(scripted.HitRateSave("ratesTest.csv"))
state.process(scripted.HitCountSave("countsTest.csv"))