def run_everything():
return run_everything_imported(run_states,
MovingWindowModel,
max_date_str,
load_data,
state_models_filename=state_models_filename,
state_report_filename=state_report_filename,
moving_window_size=moving_window_size,
n_bootstraps=n_bootstraps,
n_likelihood_samples=n_likelihood_samples,
load_data_obj=load_data,
sorted_param_names=sorted_param_names,
sorted_init_condit_names=sorted_init_condit_names,
curve_fit_bounds=curve_fit_bounds,
priors=priors,
test_params=test_params,
static_params=static_params,
opt_calc=opt_calc,
opt_force_plot=opt_force_plot,
logarithmic_params=logarithmic_params,
extra_params=extra_params,
plot_param_names=plot_param_names,
opt_statsmodels=True,
opt_simplified=opt_simplified
)
def _run_everything_sub(region=Region.US_states, override_run_states=None):
if type(load_data.current_cases_ranked_us_counties) == tuple:
load_data.current_cases_ranked_us_counties = load_data.current_cases_ranked_us_counties[0]
if type(load_data.current_cases_ranked_non_us_provinces) == tuple:
load_data.current_cases_ranked_non_us_provinces = load_data.current_cases_ranked_non_us_provinces[0]
load_data.current_cases_ranked_non_us_provinces = [x for x in load_data.current_cases_ranked_non_us_provinces \
if not x.startswith('US:')]
# Remove provinces without enough data
new_provinces = list()
for province in load_data.current_cases_ranked_non_us_provinces:
tmp_dict = load_data.get_state_data(province)
if tmp_dict['series_data'].shape[1] < 3 or tmp_dict['series_data'].shape[0] < 30 or province.startswith('China:'):
print(f'Removing province {province}')
if tmp_dict['series_data'].shape[1] >= 3:
print(f' with tmp_dict["series_data"].shape = {tmp_dict["series_data"].shape}')
else:
print(f'Keeping province {province}')
print(f' with tmp_dict["series_data"].shape = {tmp_dict["series_data"].shape}')
new_provinces.append(province)
load_data.current_cases_ranked_non_us_provinces = new_provinces
print('load_data.current_cases_ranked_non_us_provinces')
[print(x) for x in sorted(load_data.current_cases_ranked_non_us_provinces)]
if override_run_states is None:
if region == Region.US_states:
override_run_states = load_data.current_cases_ranked_us_states
elif region == Region.US_counties:
override_run_states = load_data.current_cases_ranked_us_counties
elif region == Region.countries:
override_run_states = load_data.current_cases_ranked_non_us_states[:100]
elif region == Region.provinces:
override_run_states = load_data.current_cases_ranked_non_us_provinces[:200]
override_run_states = [x for x in override_run_states if not x.startswith(' ')]
print('Gonna run these states:')
[print(x) for x in sorted(override_run_states)]
model_type_name = f'moving_window_{moving_window_size}_days_{region}_region'
if override_max_date_str is None:
hyperparameter_max_date_str = datetime.datetime.today().strftime('%Y-%m-%d')
else:
hyperparameter_max_date_str = override_max_date_str
state_models_filename = f'state_models_smoothed_moving_window_{region}_{n_bootstraps}_bootstraps_{n_likelihood_samples}_likelihood_samples_{hyperparameter_max_date_str.replace("-", "_")}_max_date.joblib'
state_report_filename = f'state_report_smoothed_moving_window_{region}_{n_bootstraps}_bootstraps_{n_likelihood_samples}_likelihood_samples_{hyperparameter_max_date_str.replace("-", "_")}_max_date.joblib'
# fixing parameters I don't want to train for saves a lot of computer power
extra_params = dict()
static_params = {'day0_positive_multiplier': 1,
'day0_deceased_multiplier': 1}
logarithmic_params = ['positive_intercept',
'deceased_intercept',
'sigma_positive',
'sigma_deceased',
# 'day0_positive_multiplier',
'day1_positive_multiplier',
'day2_positive_multiplier',
'day3_positive_multiplier',
'day4_positive_multiplier',
'day5_positive_multiplier',
'day6_positive_multiplier',
# 'day0_deceased_multiplier',
'day1_deceased_multiplier',
'day2_deceased_multiplier',
'day3_deceased_multiplier',
'day4_deceased_multiplier',
'day5_deceased_multiplier',
'day6_deceased_multiplier',
]
plot_param_names = ['positive_slope',
'positive_intercept',
'deceased_slope',
'deceased_intercept',
'sigma_positive',
'sigma_deceased'
]
if opt_simplified:
plot_param_names = ['positive_slope',
'deceased_slope',
'positive_intercept',
'deceased_intercept', ]
sorted_init_condit_names = list()
sorted_param_names = ['positive_slope',
'positive_intercept',
'deceased_slope',
'deceased_intercept',
'sigma_positive',
'sigma_deceased',
# 'day0_positive_multiplier',
'day1_positive_multiplier',
'day2_positive_multiplier',
'day3_positive_multiplier',
'day4_positive_multiplier',
'day5_positive_multiplier',
'day6_positive_multiplier',
# 'day0_deceased_multiplier',
'day1_deceased_multiplier',
'day2_deceased_multiplier',
'day3_deceased_multiplier',
'day4_deceased_multiplier',
'day5_deceased_multiplier',
'day6_deceased_multiplier']
curve_fit_bounds = {'positive_slope': (-10, 10),
'positive_intercept': (0, 1000000),
'deceased_slope': (-10, 10),
'deceased_intercept': (0, 1000000),
'sigma_positive': (0, 100),
'sigma_deceased': (0, 100),
# 'day0_positive_multiplier': (0, 10),
'day1_positive_multiplier': (0, 10),
'day2_positive_multiplier': (0, 10),
'day3_positive_multiplier': (0, 10),
'day4_positive_multiplier': (0, 10),
'day5_positive_multiplier': (0, 10),
'day6_positive_multiplier': (0, 10),
# 'day0_deceased_multiplier': (0, 10),
'day1_deceased_multiplier': (0, 10),
'day2_deceased_multiplier': (0, 10),
'day3_deceased_multiplier': (0, 10),
'day4_deceased_multiplier': (0, 10),
'day5_deceased_multiplier': (0, 10),
'day6_deceased_multiplier': (0, 10)
}
test_params = {'positive_slope': 0,
'positive_intercept': 2500,
'deceased_slope': 0,
'deceased_intercept': 250,
'sigma_positive': 0.05,
'sigma_deceased': 0.1,
# 'day0_positive_multiplier': 1,
'day1_positive_multiplier': 1,
'day2_positive_multiplier': 1,
'day3_positive_multiplier': 1,
'day4_positive_multiplier': 1,
'day5_positive_multiplier': 1,
'day6_positive_multiplier': 1,
# 'day0_deceased_multiplier': 1,
'day1_deceased_multiplier': 1,
'day2_deceased_multiplier': 1,
'day3_deceased_multiplier': 1,
'day4_deceased_multiplier': 1,
'day5_deceased_multiplier': 1,
'day6_deceased_multiplier': 1
}
# uniform priors with bounds:
priors = curve_fit_bounds
plot_subfolder = run_everything_imported(override_run_states,
MovingWindowModel,
load_data,
model_type_name=model_type_name,
state_models_filename=state_models_filename,
state_report_filename=state_report_filename,
moving_window_size=moving_window_size,
n_bootstraps=n_bootstraps,
n_likelihood_samples=n_likelihood_samples,
load_data_obj=load_data,
sorted_param_names=sorted_param_names,
sorted_init_condit_names=sorted_init_condit_names,
curve_fit_bounds=curve_fit_bounds,
priors=priors,
test_params=test_params,
static_params=static_params,
opt_force_calc=opt_force_calc,
opt_force_plot=opt_force_plot,
logarithmic_params=logarithmic_params,
extra_params=extra_params,
plot_param_names=plot_param_names,
opt_statsmodels=True,
opt_simplified=opt_simplified,
override_max_date_str=override_max_date_str,
)
return plot_subfolder
def run_everything():
if override_max_date_str is None:
hyperparameter_max_date_str = datetime.datetime.today().strftime('%Y-%m-%d')
else:
hyperparameter_max_date_str = override_max_date_str
state_models_filename = f'state_models_smoothed_convolution_{n_bootstraps}_bootstraps_{n_likelihood_samples}_likelihood_samples_{hyperparameter_max_date_str.replace("-", "_")}_max_date.joblib'
state_report_filename = f'state_report_smoothed_convolution_{n_bootstraps}_bootstraps_{n_likelihood_samples}_likelihood_samples_{hyperparameter_max_date_str.replace("-", "_")}_max_date.joblib'
# fixing parameters I don't want to train for saves a lot of computer power
static_params = {'contagious_to_positive_width': 7,
'contagious_to_deceased_width': 7,
'contagious_to_positive_mult': 0.1}
logarithmic_params = ['I_0',
'contagious_to_deceased_mult',
'sigma_positive',
'sigma_deceased']
sorted_init_condit_names = ['I_0']
sorted_param_names = ['alpha_1',
'alpha_2',
'contagious_to_positive_delay',
'contagious_to_deceased_delay',
'contagious_to_deceased_mult',
'sigma_positive',
'sigma_deceased'
]
plot_param_names = ['alpha_1',
'alpha_2',
'contagious_to_positive_delay',
'positive_to_deceased_delay',
'positive_to_deceased_mult']
def get_positive_to_deceased_delay(x, map_name_to_sorted_ind=None):
return x[map_name_to_sorted_ind['contagious_to_deceased_delay']] - x[
map_name_to_sorted_ind['contagious_to_positive_delay']]
def get_positive_to_deceased_mult(x, map_name_to_sorted_ind=None):
return x[map_name_to_sorted_ind['contagious_to_deceased_mult']] / 0.1
extra_params = {
'positive_to_deceased_delay': get_positive_to_deceased_delay,
'positive_to_deceased_mult': get_positive_to_deceased_mult
}
curve_fit_bounds = {'I_0': (1e-12, 100.0), # starting infections
'alpha_1': (-1, 2),
'alpha_2': (-1, 2),
'sigma_positive': (0, 100),
'sigma_deceased': (0, 100),
'contagious_to_positive_delay': (-14, 21),
# 'contagious_to_positive_width': (0, 14),
'contagious_to_deceased_delay': (-14, 42),
# 'contagious_to_deceased_width': (0, 14),
'contagious_to_deceased_mult': (1e-12, 1),
}
test_params = {'I_0': 2e-3, # starting infections
'alpha_1': 0.23,
'alpha_2': 0.01,
'sigma_positive': 0.01,
'sigma_deceased': 0.2,
'contagious_to_positive_delay': 9,
# 'contagious_to_positive_width': 7,
'contagious_to_deceased_delay': 15,
# 'contagious_to_deceased_width': 7,
'contagious_to_deceased_mult': 0.01,
}
# uniform priors with bounds:
priors = curve_fit_bounds
# cycle over most populous states first
population_ranked_state_names = sorted(load_data.map_state_to_current_case_cnt.keys(),
key=lambda x: -load_data.map_state_to_current_case_cnt[x])
run_states = population_ranked_state_names
if override_run_states is not None:
run_states = override_run_states
return run_everything_imported(run_states,
ConvolutionModel,
max_date_str,
load_data,
state_models_filename=state_models_filename,
state_report_filename=state_report_filename,
n_bootstraps=n_bootstraps,
n_likelihood_samples=n_likelihood_samples,
load_data_obj=load_data,
sorted_param_names=sorted_param_names,
sorted_init_condit_names=sorted_init_condit_names,
curve_fit_bounds=curve_fit_bounds,
priors=priors,
test_params=test_params,
static_params=static_params,
opt_force_calc=opt_force_calc,
opt_force_plot=opt_force_plot,
logarithmic_params=logarithmic_params,
extra_params=extra_params,
plot_param_names=plot_param_names,
)
def _run_everything_sub(region=Region.US_states, override_run_states=None):
if type(load_data.current_cases_ranked_us_counties) == tuple:
load_data.current_cases_ranked_us_counties = load_data.current_cases_ranked_us_counties[
0]
if type(load_data.current_cases_ranked_non_us_provinces) == tuple:
load_data.current_cases_ranked_non_us_provinces = load_data.current_cases_ranked_non_us_provinces[
0]
load_data.current_cases_ranked_non_us_provinces = [x for x in load_data.current_cases_ranked_non_us_provinces \
if not x.startswith('US:')]
# Remove provinces without enough data
new_provinces = list()
for province in load_data.current_cases_ranked_non_us_provinces:
tmp_dict = load_data.get_state_data(province)
if tmp_dict['series_data'].shape[1] < 3 or tmp_dict[
'series_data'].shape[0] < 30 or province.startswith('China:'):
print(f'Removing province {province}')
if tmp_dict['series_data'].shape[1] >= 3:
print(
f' with tmp_dict["series_data"].shape = {tmp_dict["series_data"].shape}'
)
else:
print(f'Keeping province {province}')
print(
f' with tmp_dict["series_data"].shape = {tmp_dict["series_data"].shape}'
)
new_provinces.append(province)
load_data.current_cases_ranked_non_us_provinces = new_provinces
print('load_data.current_cases_ranked_non_us_provinces')
[print(x) for x in sorted(load_data.current_cases_ranked_non_us_provinces)]
if override_run_states is None:
if region == Region.US_states:
override_run_states = load_data.current_cases_ranked_us_states
elif region == Region.US_counties:
override_run_states = load_data.current_cases_ranked_us_counties[:
70]
elif region == Region.countries:
override_run_states = load_data.current_cases_ranked_non_us_states[:
70]
elif region == Region.provinces:
override_run_states = load_data.current_cases_ranked_non_us_provinces[:
70]
override_run_states = [
x for x in override_run_states if not x.startswith(' ')
]
print('Gonna run these states:')
[print(x) for x in sorted(override_run_states)]
model_type_name = f'frechet_{region}_region'
if override_max_date_str is None:
hyperparameter_max_date_str = datetime.datetime.today().strftime(
'%Y-%m-%d')
else:
hyperparameter_max_date_str = override_max_date_str
state_models_filename = f'state_models_smoothed_frechet_{n_bootstraps}_bootstraps_{n_likelihood_samples}_likelihood_samples_{hyperparameter_max_date_str.replace("-", "_")}_max_date.joblib'
state_report_filename = f'state_report_smoothed_frechet_{n_bootstraps}_bootstraps_{n_likelihood_samples}_likelihood_samples_{hyperparameter_max_date_str.replace("-", "_")}_max_date.joblib'
# fixing parameters I don't want to train for saves a lot of computer power
static_params = dict() #{'mult_positive': 3400,
#'mult_deceased': 3400})
logarithmic_params = [
'alpha_positive',
's_positive',
'alpha_deceased',
's_deceased',
'sigma_positive',
'sigma_deceased',
]
sorted_init_condit_names = list()
sorted_param_names = [
'alpha_positive',
's_positive',
'm_positive',
'mult_positive',
'alpha_deceased',
's_deceased',
'm_deceased',
'mult_deceased',
'sigma_positive',
'sigma_deceased',
]
plot_param_names = [
'alpha_positive',
's_positive',
'm_positive',
'mult_positive',
'alpha_deceased',
's_deceased',
'm_deceased',
'mult_deceased',
'sigma_positive',
'sigma_deceased',
]
extra_params = dict()
curve_fit_bounds = {
'alpha_positive': (1e-8, 1000),
's_positive': (1e-3, 1000),
'm_positive': (-1000, 1000),
'mult_positive': (1e-8, 1e12),
'alpha_deceased': (1e-8, 1000),
's_deceased': (1e-3, 1000),
'm_deceased': (-1000, 1000),
'mult_deceased': (1e-8, 1e12),
'sigma_positive': (0, 100),
'sigma_deceased': (0, 100),
}
test_params = {
'alpha_positive': 8,
's_positive': 200,
'm_positive': -50,
'mult_positive': 2e6,
'alpha_deceased': 8,
's_deceased': 140,
'm_deceased': -50,
'mult_deceased': 1e5,
'sigma_positive': 0.5,
'sigma_deceased': 0.3,
}
# uniform priors with bounds:
priors = curve_fit_bounds
plot_subfolder = run_everything_imported(
override_run_states,
FrechetModel,
load_data,
model_type_name=model_type_name,
state_models_filename=state_models_filename,
state_report_filename=state_report_filename,
n_bootstraps=n_bootstraps,
n_likelihood_samples=n_likelihood_samples,
load_data_obj=load_data,
sorted_param_names=sorted_param_names,
sorted_init_condit_names=sorted_init_condit_names,
curve_fit_bounds=curve_fit_bounds,
priors=priors,
test_params=test_params,
static_params=static_params,
opt_force_calc=opt_force_calc,
opt_force_plot=opt_force_plot,
logarithmic_params=logarithmic_params,
extra_params=extra_params,
plot_param_names=plot_param_names,
opt_statsmodels=True,
override_max_date_str=override_max_date_str,
)
return plot_subfolder