def one_run(t):
params = prepare_params(t)
# t days **before** lockdown (Jan 23)
# simulation finishes at Jan 27 (after lockdown for 5 days)
# t=1, 2, 3, ... means patient zero appeared in Jan 22, 21, 20
p0_time = lockdown_date - timedelta(days=t)
total, _, _, _, stats = do_simulation(t + 5,
bed_info,
params,
p0_time=p0_time,
verbose=0)
pred_I = np.array([total[-1, STATE.I]])
pred_E = np.array([total[-1, STATE.E]])
pred_M = np.array([total[-1, STATE.M]])
mse_I = mean_absolute_error(actual_I, pred_I)
mse_E = mean_absolute_error(actual_E, pred_E)
mse_M = mean_absolute_error(actual_M, pred_M)
mse_IM = mean_absolute_error(actual_M + actual_I, pred_M + pred_I)
mse_IEM = mean_absolute_error(actual_I + actual_E + actual_M,
pred_I + pred_E + pred_M)
return (t, actual_I[0], pred_I[0], mse_I, mse_E, mse_M, mse_IM, mse_IEM,
stats)
def one_run(initial_num_I, I2E_factor, I2M_factor, alpha, beta, mu_ei, mu_mo,
k_days, x0_pt):
initial_num_E = initial_num_I * I2E_factor
initial_num_M = initial_num_I * I2M_factor
initial_num_M = min(bed_info[0][1], initial_num_M)
params = Params(initial_num_I=initial_num_I,
initial_num_E=initial_num_E,
initial_num_M=initial_num_M,
alpha=alpha,
beta=beta,
mu_ei=mu_ei,
mu_mo=mu_mo,
k_days=k_days,
x0_pt=x0_pt)
total, delta, increase, trans_data, ax = do_simulation(total_days + 3,
bed_info,
params,
p0_time=start_date)
I_mae = mean_absolute_error(I_true, increase[1:(total_days + 1), STATE.I])
O_mae = mean_absolute_error(O_true, increase[1:(total_days + 1), STATE.O])
is_decreasing_after_Feb09 = True
for i in range((T('09/02/2020') - start_date).days, total.shape[0] - 1):
if total[i, STATE.I] < total[i + 1, STATE.I]:
is_decreasing_after_Feb09 = False
row = (initial_num_I, initial_num_E, initial_num_M, alpha, beta, k_days,
mu_ei, mu_mo, x0_pt, I_mae, O_mae, is_decreasing_after_Feb09)
return row
def one_run(params_after, bed_info, delta_t, total_days=360):
delta = timedelta(days=delta_t)
assumed_ld_date = lockdown_time + delta
days_to_p0 = (assumed_ld_date - p0_time).days
print('lockdown date', assumed_ld_date)
params = Params(total_population=params_before.total_population,
initial_num_E=params_before.initial_num_E,
initial_num_I=params_before.initial_num_I,
initial_num_M=params_before.initial_num_M,
alpha=[(0, params_before.alpha),
(days_to_p0, params_after.alpha)],
beta=[(0, params_before.beta),
(days_to_p0, params_after.beta)],
mu_ei=params_after.mu_ei,
mu_mo=params_after.mu_mo,
k_days=params_after.k_days,
x0_pt=params_after.x0_pt,
stages=[days_to_p0])
total, delta, increase, trans, stats = do_simulation(total_days,
bed_info,
params,
p0_time=p0_time)
bundle = [total, delta, increase, trans]
return assumed_ld_date, delta_t, bundle, stats
def test_equivalence():
p0_time, total_days, bed_info, params = get_input()
total_actual, delta_actual, increase_actual, trans_data_actual, stats_actual = do_simulation(
total_days,
bed_info,
params,
p0_time=p0_time,
verbose=0,
show_bar=True)
total_expected, delta_expected, increase_expected, trans_data_expected, stats_expected = do_simulation_old(
total_days,
bed_info,
params,
p0_time=p0_time,
verbose=0,
show_bar=True)
assert_areq(total_expected, total_actual)
assert_areq(delta_expected, delta_actual)
assert_areq(increase_expected, increase_actual)
assert_areq(trans_data_expected, trans_data_actual)
assert stats_actual == stats_expected
if d <= end_date]
params = Params(initial_num_I=br.initial_num_I,
initial_num_E=br.initial_num_E,
initial_num_M=br.initial_num_M,
alpha=br.alpha,
beta=br.beta,
mu_ei=br.mu_ei,
mu_mo=br.mu_mo,
k_days=int(br.k_days),
x0_pt=br.x0_pt)
pkl.dump(params, open('output/params_after_lockdown.pkl', 'wb'))
total, delta, increase, trans_data, aux = do_simulation(total_days + 60,
bed_info,
params,
p0_time=start_date)
I_true_all = df[(df['date'] > start_date)]['infected'].values
I_pred_all = increase[1:len(I_true_all) + 1, STATE.I]
I_pred = increase[1:(len(I_true) + 1), STATE.I]
dates = pd.date_range(start_date + timedelta(days=1),
end_date + timedelta(days=2))
data = {
'date': dates,
'true_I': I_true_all,
'pred_I': I_pred_all,
'abs_error': np.abs(I_true_all - I_pred_all),
'squared_error': np.power(I_true_all - I_pred_all, 2),
'used_in_fitting': [(d >= start_date) & (d <= end_date) for d in dates]
initial_num_E=params_before.initial_num_E,
initial_num_I=params_before.initial_num_I,
initial_num_M=params_before.initial_num_M,
mu_ei=params_after.mu_ei,
mu_mo=params_after.mu_mo,
k_days=params_after.k_days,
x0_pt=params_after.x0_pt,
alpha=fine_grained_alpha,
beta=fine_grained_beta,
stages=[days_before_ld] +
[(days_before_ld + i)
for i in range(offset, offset * n_offsets + 1, offset)])
total, delta, increase, trans_data, stats = do_simulation(total_days,
bed_info,
params,
p0_time=p0_time,
verbose=0)
# In[12]:
stats
# In[13]:
params
# In[14]:
stats
makedir_if_not_there('output/tbl/p0-time/')
df.to_csv('output/tbl/p0-time/error.csv', index=False)
pkl.dump(p0_time, open('output/p0_time.pkl', 'wb'))
params = copy(params_jan27)
params.alpha = infection_factor * params.alpha
params.beta = infection_factor * params.beta
params.initial_num_E = 1
params.initial_num_I = 0
params.initial_num_M = 0
total, delta, increase, trans, stats = do_simulation(best_t,
bed_info,
params,
verbose=0,
p0_time=p0_time)
makedir_if_not_there('output/tbl/before-lockdown/')
save_bundle([total, delta, increase, trans], p0_time, best_t,
'output/tbl/before-lockdown')
save_to_json(stats, 'output/tbl/before-lockdown/stat.txt')
pkl.dump(params, open('output/params_before_lockdown.pkl', 'wb'))
fig, ax = plot_total(total, p0_time, best_t)
makedir_if_not_there('figs/')
fig.savefig('figs/before-lockdown.pdf')
