def main(load_inter_flag=False):
if not load_inter_flag:
real_data = list(pd.read_csv(get_data_path(True))['adcode'].unique())
start_date = date(2020, 1, 24)
real_data1 = pd.read_csv(get_data_path(True))
history_real = prepareData(real_data1)
rerun_cities = None
for i in range(get_seed_num()):
total_params = pd.read_csv(
os.path.join(os.path.dirname(__file__),
'./params_foreign{}.csv'.format(int(i))))
for ind, city in enumerate(real_data):
print(i, ind, city, str(get_important_date(city)))
x = get_params(total_params, city)
if rerun_cities is None or city in rerun_cities:
run_simulation2(x,
city,
60,
60,
start_date,
get_important_date(city),
history_real,
unob_flow_num=None,
json_name='data_run_foreign{}.json'.format(
int(i)))
# choose the final parameters
max_data, min_data, mean_data, best_data, \
std_data, middle_data, xmin_data, xmax_data, \
xmean_data, xstd_data = load_and_save('./data_run_foreign{}.json', './data_run_foreign_{}.json')
"""
def run_simulation(x,
city,
simulate_day1,
simulate_day2,
start_date,
simulate_date,
incubation=3,
unob_flow_num=None,
json_name='data_run.json'):
real_data = pd.read_csv(get_data_path(True))
history_real = prepareData(real_data)
run_simulation2(x, city, simulate_day1, simulate_day2, start_date,
simulate_date, history_real, incubation, unob_flow_num,
json_name)
def prepare_data_china(load_inter_flag=False):
real_data = pd.read_csv(get_data_path())['adcode'].unique()
rerun_cities = None
if rerun_cities is not None:
real_data = rerun_cities.copy()
start_date = date(2020, 1, 11)
real_data1 = pd.read_csv(get_data_path())
history_real = prepareData(real_data1)
if not load_inter_flag:
print('re-simulate the progress')
province_travel_dict = flowHubei()
for i in range(get_seed_num()):
total_params = pd.read_csv(
os.path.join(os.path.dirname(__file__),
'./params{}.csv'.format(int(i))))
unob_flow_num = initHubei(
get_params(total_params, 420000),
start_date,
important_date=[get_important_date(420000)],
travel_from_hubei=province_travel_dict)
for ind, city in enumerate(real_data):
print(i, ind, city)
x = get_params(total_params, city)
run_simulation2(x,
city,
90,
90,
start_date,
get_important_date(city),
history_real,
unob_flow_num=unob_flow_num,
json_name='data_run{}.json'.format(int(i)))
else:
print('directly load the pre-simulated data')
for i in range(get_seed_num()):
get_total_province('data_run{}.json'.format(int(i)))
# choose the final parameters
max_data, min_data, mean_data, best_data, std_data, middle_data, \
xmin_data, xmax_data, xmean_data, xstd_data = load_and_save(
'./data_run{}.json', './data_run_{}.json')
"""
def run_opt(city, budget, start_date, important_dates, infectratio_range=None,
dummy_range=None, unob_flow_num=None, repeat_time=1, init_samples=None,
training_end_date=None, json_name='data_run.json', seed=3, loss_ord=0.0,
touch_range=None,iso_range=None):
assert infectratio_range is not None and dummy_range is not None
days_predict = 0
# load data
real_data = pd.read_csv(get_data_path(True))
history_real = prepareData(real_data)
flow_out_data = None
# initialize models
infectratio = InfectRatio(1, [infectratio_range], [True])
if touch_range is None:
touchratio = TouchRatio(1, [[0.999, 1.0000]], [True])
else:
touchratio = TouchRatio(1, [touch_range], [True])
touchratiointra = TouchRatio(1, [[0, 1]], [True])
obs = ObservationRatio(1, [[0.0, 0.3]], [True])
dead = DeadRatio(1, [[0., 0.01]], [True])
if iso_range is None:
isoratio = IsolationRatio(1, [[0.03, 0.12]], [True])
else:
isoratio = IsolationRatio(1, [iso_range], [True])
dummy = DummyModel(1, [dummy_range], [True, True])
cure_ratio = InfectRatio(1, [[0., 0.15]], [True])
simulator = Simulator(city, infectratio, touchratio, obs, dead, dummy, isoratio, touchratiointra, cure_ratio, important_dates,
unob_flow_num=unob_flow_num, flow_out_data=flow_out_data, training_date_end=training_end_date)
test_date = datetime.strptime(history_real['date'].max(), '%Y-%m-%d').date() - timedelta(days_predict)
history_real = history_real[history_real['adcode'] == city]
history_real = history_real[history_real['date'] >= str(start_date)]
history_train = history_real[history_real['date'] <= str(test_date)]
x, y = simulator.fit(history_train, budget=budget, server_num=get_core_num(),
repeat=repeat_time, seed=seed, intermediate_freq=10000, init_samples=init_samples,
loss_ord=loss_ord)
print('best_solution: x = ', x, 'y = ', y)
simulator.set_param(x)
run_simulation(x, city, 60, 60, start_date, get_important_date(city), unob_flow_num=unob_flow_num, json_name=json_name)
return x
def run_opt(city,
budget,
start_date,
important_dates,
infectratio_range=None,
dummy_range=None,
unob_flow_num=None,
repeat_time=1,
init_samples=None,
training_date_end=None,
json_name='data_run.json',
seed=1,
loss_ord=0.0,
unob_period=None,
obs_period=None,
iso_period=None,
cure_period=None,
isoratio_it=None):
if city == 420000:
infectratio_range = [0., 0.05]
dummy_range = [0.0000, 400.00001]
else:
assert infectratio_range is not None and dummy_range is not None
days_predict = 0
# load data
real_data = pd.read_csv(get_data_path())
history_real = prepareData(real_data)
flow_out_data = flowOutData()
# initialize models
infectratio = InfectRatio(1, [infectratio_range], [True])
touchratio = TouchRatio(1, [[0.0, 0.6]], [True])
touchratiointra = TouchRatio(1, [[0, 1]], [True])
obs = ObservationRatio(1, [[0.0, 0.3]], [True])
dead = DeadRatio(1, [[0., 0.01]], [True])
if isoratio_it is None:
isoratio = IsolationRatio(1, [[0.2, 0.5]], [True])
else:
isoratio = IsolationRatio(1, [isoratio_it], [True])
dummy = DummyModel(1, [dummy_range], [True, True])
cure_ratio = InfectRatio(1, [[0., 0.1]], [True])
# set the time of applying touchratio
simulator = Simulator(city,
infectratio,
touchratio,
obs,
dead,
dummy,
isoratio,
touchratiointra,
cure_ratio,
important_dates,
unob_flow_num=unob_flow_num,
flow_out_data=flow_out_data,
training_date_end=training_date_end)
# set period here
simulator.set_period()
test_date = datetime.strptime(history_real['date'].max(),
'%Y-%m-%d').date() - timedelta(days_predict)
history_real = history_real[history_real['adcode'] == city]
history_real = history_real[history_real['date'] >= str(start_date)]
history_train = history_real[history_real['date'] <= str(test_date)]
x, y = simulator.fit(history_train,
budget=budget,
server_num=get_core_num(),
repeat=repeat_time,
seed=seed,
intermediate_freq=10000,
init_samples=init_samples,
loss_ord=loss_ord)
print('best_solution: x = ', x, 'y = ', y)
simulator.set_param(x)
run_simulation(x,
city,
60,
60,
start_date,
get_important_date(city),
unob_flow_num=unob_flow_num,
json_name=json_name)
duration = len(real_data["date"].unique()) - 1
sim_res, _ = simulator.simulate(str(start_date), duration)
print('RMSE: ', get_newly_loss(sim_res, history_real))
return x, sim_res
def initHubei(x, start_date, important_date, travel_from_hubei):
if travel_from_hubei is None:
return None
days_predict = 0
# load data
real_data = pd.read_csv(get_data_path())
history_real = prepareData(real_data)
infectratio = InfectRatio(1, [[0, 1]], [True])
touchratio = TouchRatio(1, [[0., 0.3]], [True])
touchratiointra = TouchRatio(1, [[0, 10]], [True])
obs = ObservationRatio(1, [[0.0, 1.]], [True])
dead = DeadRatio(1, [[0., 0.1]], [True])
isoratio = IsolationRatio(1, [[0., 1]], [True])
cure_ratio = InfectRatio(1, [[0., 100]], [True])
dummy = DummyModel(1, [[0, 200000]], [True, True])
flow_out_data = flowOutData()
test_date = datetime.strptime(history_real['date'].max(),
'%Y-%m-%d').date() - timedelta(days_predict)
history_real = history_real[history_real['adcode'] == 420000]
history_real = history_real[history_real['date'] >= str(start_date)]
history_train = history_real[history_real['date'] <= str(test_date)]
duration = len(history_train["date"].unique())
city = 420000
simulator = Simulator(city,
infectratio,
touchratio,
obs,
dead,
dummy,
isoratio,
touchratiointra,
cure_ratio,
important_date,
flow_out_data=flow_out_data)
simulator.set_param(x)
total_population = get_populations()[420000]
simulated_result, detailed_result = simulator.simulate(
str(start_date), duration + 60)
init_unob = [item for item in reversed(detailed_result.unob_total_init)]
unob_ratio = {}
for i, item in enumerate(init_unob):
date_now = start_date - timedelta(i + 1)
unob_ratio[date_now] = item / total_population
for it_date, unob_num in zip(simulated_result['date'],
simulated_result['total_unobserved']):
unob_ratio[datetime.strptime(
it_date, '%Y-%m-%d').date()] = unob_num / total_population
unob_flow_num = {}
lst_travel = {}
lst_date = None
for it_date in travel_from_hubei:
if it_date not in unob_ratio:
continue
if it_date not in unob_flow_num:
unob_flow_num[it_date] = {}
for it_code in travel_from_hubei[it_date]:
unob_flow_num[it_date][it_code] = travel_from_hubei[it_date][
it_code] * unob_ratio[it_date]
lst_travel = travel_from_hubei[it_date]
lst_date = it_date
lst_date = lst_date + timedelta(1)
while True:
if lst_date in unob_ratio:
if lst_date not in unob_flow_num:
unob_flow_num[lst_date] = {}
for it_code in lst_travel:
unob_flow_num[lst_date][
it_code] = lst_travel[it_code] * unob_ratio[lst_date]
else:
break
lst_date = lst_date + timedelta(1)
return unob_flow_num