def deploy_dod_projections(disable_validation, input_state_dir, input_county_dir, output):
"""Generates and runs dod data projections from model outputs.
Used for manual trigger
# triggering persistance to s3
AWS_PROFILE=covidactnow BUCKET_NAME=covidactnow-deleteme python run.py deploy-dod-projections
# deploy to the data bucket
AWS_PROFILE=covidactnow BUCKET_NAME=data.covidactnow.org python run.py deploy-dod-projections
# triggering persistance to local
python run.py deploy-dod-projections
"""
for intervention in list(Intervention):
# TODO(issues/#259): Split up the dod pipeline to process states/counties since states support inference
# TODO(issues/#258): remove check once counties support inferrence
if intervention in Intervention.county_supported_interventions():
logger.info(f"Starting to generate files for {intervention.name}.")
state_result, county_result = dod_pipeline.run_projections(
input_state_dir, input_county_dir, intervention, run_validation=not disable_validation
)
dod_pipeline.deploy_results(state_result, output)
dod_pipeline.deploy_results(county_result, output)
logger.info('finished dod job')
def deploy_counties_api(disable_validation, input_dir, output, summary_output):
"""The entry function for invocation"""
for intervention in list(Intervention):
# TODO(issues/#258): remove check once counties support inferrence
if intervention in Intervention.county_supported_interventions():
county_result = api_pipeline.run_projections(
input_dir,
AggregationLevel.COUNTY,
intervention,
run_validation=not disable_validation,
)
county_summaries, county_timeseries = api_pipeline.generate_api(
county_result, input_dir
)
api_pipeline.deploy_results([*county_summaries, *county_timeseries], output)
counties_summary = api_pipeline.build_counties_summary(county_summaries, intervention)
counties_timeseries = api_pipeline.build_counties_timeseries(county_timeseries, intervention)
summarized_timeseries = api_pipeline.build_prediction_header_timeseries_data(counties_timeseries)
api_pipeline.deploy_prediction_timeseries_csvs(summarized_timeseries, summary_output)
api_pipeline.deploy_results([counties_summary], summary_output, write_csv=True)
api_pipeline.deploy_results([counties_timeseries], summary_output)
logger.info("finished top counties job")
def _get_intervention_type(intervention_type, state, state_interventions_df):
if intervention_type == Intervention.SELECTED_MITIGATION.value:
state_intervention_results = state_interventions_df.loc[
state_interventions_df["state"] == state]["intervention"]
if not state_intervention_results.empty:
intervention_string = state_intervention_results.values[0]
return Intervention.from_str(intervention_string).value
return intervention_type
def __init__(self, data):
data = data.reset_index(drop=True)
self.fips = data[schema.FIPS].iloc[0]
self.data = data
self.intervention = Intervention(data[schema.INTERVENTION].iloc[0])
self.data["short_fall"] = _calculate_shortfall(
self.data[schema.ALL_HOSPITALIZED], self.data[schema.BEDS])
def get_intervention_for_state(state) -> Intervention:
return Intervention.from_str(get_interventions()[state])
def write_region(self, regional_input: RegionalInput) -> None:
"""Generates the CAN UI output format for a given region.
Args:
regional_input: the region and its data
"""
# Get the latest observed values to use in calculating shims
observed_latest_dict = regional_input.latest
state = observed_latest_dict[CommonFields.STATE]
log.info("Mapping output to WebUI.",
state=state,
fips=regional_input.fips)
shim_log = structlog.getLogger(fips=regional_input.fips)
pyseir_outputs = regional_input.ensemble_results()
try:
fit_results = regional_input.inference_result()
t0_simulation = datetime.fromisoformat(fit_results["t0_date"])
except (KeyError, ValueError):
log.error("Fit result not found for fips. Skipping...",
fips=regional_input.fips)
return
population = regional_input.population
# We will shim all suppression policies by the same amount (since historical tracking error
# for all policies is the same).
baseline_policy = "suppression_policy__inferred" # This could be any valid policy
# We need the index in the model's temporal frame.
idx_offset = int(fit_results["t_today"] - fit_results["t0"])
observed_death_latest = observed_latest_dict[CommonFields.DEATHS]
observed_total_hosps_latest = observed_latest_dict[
CommonFields.CURRENT_HOSPITALIZED]
observed_icu_latest = observed_latest_dict[CommonFields.CURRENT_ICU]
# For Deaths
model_death_latest = pyseir_outputs[baseline_policy]["total_deaths"][
"ci_50"][idx_offset]
model_acute_latest = pyseir_outputs[baseline_policy]["HGen"]["ci_50"][
idx_offset]
model_icu_latest = pyseir_outputs[baseline_policy]["HICU"]["ci_50"][
idx_offset]
model_total_hosps_latest = model_acute_latest + model_icu_latest
death_shim = shim.calculate_strict_shim(
model=model_death_latest,
observed=observed_death_latest,
log=shim_log.bind(type=CommonFields.DEATHS),
)
total_hosp_shim = shim.calculate_strict_shim(
model=model_total_hosps_latest,
observed=observed_total_hosps_latest,
log=shim_log.bind(type=CommonFields.CURRENT_HOSPITALIZED),
)
# For ICU This one is a little more interesting since we often don't have ICU. In this case
# we use information from the same aggregation level (intralevel) to keep the ratios
# between general hospitalization and icu hospitalization
icu_shim = shim.calculate_intralevel_icu_shim(
model_acute=model_acute_latest,
model_icu=model_icu_latest,
observed_icu=observed_icu_latest,
observed_total_hosps=observed_total_hosps_latest,
log=shim_log.bind(type=CommonFields.CURRENT_ICU),
)
# Iterate through each suppression policy.
# Model output is interpolated to the dates desired for the API.
suppression_policies = [
key for key in pyseir_outputs.keys()
if key.startswith("suppression_policy")
]
for suppression_policy in suppression_policies:
output_for_policy = pyseir_outputs[suppression_policy]
output_model = pd.DataFrame()
t_list = output_for_policy["t_list"]
t_list_downsampled = range(0, int(max(t_list)),
self.output_interval_days)
output_model[schema.DAY_NUM] = t_list_downsampled
output_model[schema.DATE] = [
(t0_simulation + timedelta(days=t)).date().strftime("%Y-%m-%d")
for t in t_list_downsampled
]
output_model[schema.TOTAL] = population
output_model[schema.TOTAL_SUSCEPTIBLE] = np.interp(
t_list_downsampled, t_list, output_for_policy["S"]["ci_50"])
output_model[schema.EXPOSED] = np.interp(
t_list_downsampled, t_list, output_for_policy["E"]["ci_50"])
output_model[schema.INFECTED] = np.interp(
t_list_downsampled,
t_list,
np.add(output_for_policy["I"]["ci_50"],
output_for_policy["A"]["ci_50"]),
) # Infected + Asympt.
output_model[schema.INFECTED_A] = output_model[schema.INFECTED]
interpolated_model_acute_values = np.interp(
t_list_downsampled, t_list, output_for_policy["HGen"]["ci_50"])
output_model[schema.INFECTED_B] = interpolated_model_acute_values
raw_model_icu_values = output_for_policy["HICU"]["ci_50"]
interpolated_model_icu_values = np.interp(t_list_downsampled,
t_list,
raw_model_icu_values)
# 21 August 2020: The line assigning schema.INFECTED_C in the output_model is
# commented out while the Linear Regression estimator is patched through this pipeline
# to be consumed downstream by the ICU utilization calculations. It is left here as a
# marker for the future if the ICU utilization calculations is dis-entangled from the
# PySEIR model outputs.
output_model[schema.INFECTED_C] = (
icu_shim + interpolated_model_icu_values).clip(min=0)
# General + ICU beds. don't include vent here because they are also counted in ICU
output_model[schema.ALL_HOSPITALIZED] = (
interpolated_model_acute_values +
interpolated_model_icu_values + total_hosp_shim).clip(min=0)
output_model[schema.ALL_INFECTED] = output_model[schema.INFECTED]
# Shim Deaths to Match Observed
raw_model_deaths_values = output_for_policy["total_deaths"][
"ci_50"]
interp_model_deaths_values = np.interp(t_list_downsampled, t_list,
raw_model_deaths_values)
output_model[schema.DEAD] = (interp_model_deaths_values +
death_shim).clip(min=0)
# Continue mapping
final_beds = np.mean(output_for_policy["HGen"]["capacity"])
output_model[schema.BEDS] = final_beds
output_model[schema.CUMULATIVE_INFECTED] = np.interp(
t_list_downsampled,
t_list,
np.cumsum(output_for_policy["total_new_infections"]["ci_50"]),
)
if fit_results:
output_model[schema.Rt] = np.interp(
t_list_downsampled,
t_list,
fit_results["eps2"] * fit_results["R0"] *
np.ones(len(t_list)),
)
output_model[schema.Rt_ci90] = np.interp(
t_list_downsampled,
t_list,
2 * fit_results["eps2_error"] * fit_results["R0"] *
np.ones(len(t_list)),
)
else:
output_model[schema.Rt] = 0
output_model[schema.Rt_ci90] = 0
output_model[schema.CURRENT_VENTILATED] = (
icu_shim +
np.interp(t_list_downsampled, t_list,
output_for_policy["HVent"]["ci_50"])).clip(min=0)
output_model[schema.POPULATION] = population
# Average capacity.
output_model[schema.ICU_BED_CAPACITY] = np.mean(
output_for_policy["HICU"]["capacity"])
output_model[schema.VENTILATOR_CAPACITY] = np.mean(
output_for_policy["HVent"]["capacity"])
# Truncate date range of output.
output_dates = pd.to_datetime(output_model["date"])
output_model = output_model[
(output_dates >= datetime(month=3, day=3, year=2020))
& (output_dates < datetime.today() + timedelta(days=90))]
# Fill in results for the Rt indicator.
rt_results = regional_input.inferred_infection_rate()
if rt_results is None or rt_results.empty:
log.warning(
"No Rt Results found, clearing Rt in output.",
fips=regional_input.fips,
suppression_policy=suppression_policy,
)
output_model[schema.RT_INDICATOR] = "NaN"
output_model[schema.RT_INDICATOR_CI90] = "NaN"
else:
rt_results.index = rt_results["date"].dt.strftime("%Y-%m-%d")
merged = output_model.merge(
rt_results[["Rt_MAP_composite", "Rt_ci95_composite"]],
right_index=True,
left_on="date",
how="left",
)
output_model[schema.RT_INDICATOR] = merged["Rt_MAP_composite"]
# With 90% probability the value is between rt_indicator - ci90
# to rt_indicator + ci90
output_model[schema.RT_INDICATOR_CI90] = (
merged["Rt_ci95_composite"] - merged["Rt_MAP_composite"])
output_model[[schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90]] = output_model[[
schema.RT_INDICATOR, schema.RT_INDICATOR_CI90
]].fillna("NaN")
int_columns = [
col for col in output_model.columns if col not in (
schema.DATE,
schema.Rt,
schema.Rt_ci90,
schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90,
schema.FIPS,
)
]
# Casing floats to ints and then replacing filled in zeros with NaN values instead of
# propagating zeros.
na_int_columns = output_model.loc[:, int_columns].isna()
output_model.loc[:,
int_columns] = output_model[int_columns].fillna(
0).astype(int)
output_model[na_int_columns] = np.nan
output_model.loc[:, [
schema.Rt, schema.Rt_ci90, schema.RT_INDICATOR, schema.
RT_INDICATOR_CI90
]] = output_model[[
schema.Rt, schema.Rt_ci90, schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90
]]
output_model[schema.FIPS] = regional_input.fips
intervention = Intervention.from_webui_data_adaptor(
suppression_policy)
output_model[schema.INTERVENTION] = intervention.value
output_path = get_run_artifact_path(regional_input.region,
RunArtifact.WEB_UI_RESULT,
output_dir=self.output_dir)
output_path = output_path.replace("__INTERVENTION_IDX__",
str(intervention.value))
output_model.to_json(output_path, orient=OUTPUT_JSON_ORIENT)
def map_fips(self, fips):
"""
For a given county fips code, generate the CAN UI output format.
Parameters
----------
fips: str
County FIPS code to map.
"""
logging.info(f"Mapping output to WebUI for {self.state}, {fips}")
pyseir_outputs = load_data.load_ensemble_results(fips)
if len(fips) == 5 and fips not in self.df_whitelist.fips.values:
logging.info(f"Excluding {fips} due to white list...")
return
try:
fit_results = load_inference_result(fips)
t0_simulation = datetime.fromisoformat(fit_results["t0_date"])
except (KeyError, ValueError):
logging.error(f"Fit result not found for {fips}. Skipping...")
return
# ---------------------------------------------------------------------
# Rescale hosps based on the population ratio... Could swap this to
# infection ratio later?
# ---------------------------------------------------------------------
hosp_times, current_hosp, _ = load_data.load_hospitalization_data_by_state(
state=self.state_abbreviation,
t0=t0_simulation,
convert_cumulative_to_current=True,
category="hospitalized",
)
_, current_icu, _ = load_data.load_hospitalization_data_by_state(
state=self.state_abbreviation,
t0=t0_simulation,
convert_cumulative_to_current=True,
category="icu",
)
if len(fips) == 5:
population = self.population_data.get_record_for_fips(fips)[CommonFields.POPULATION]
else:
population = self.population_data.get_record_for_state(self.state_abbreviation)[
CommonFields.POPULATION
]
# logging.info(f'Mapping output to WebUI for {self.state}, {fips}')
# pyseir_outputs = load_data.load_ensemble_results(fips)
# if pyseir_outputs is None:
# logging.warning(f'Fit result not found for {fips}: Skipping county')
# return None
policies = [key for key in pyseir_outputs.keys() if key.startswith("suppression_policy")]
if current_hosp is not None:
t_latest_hosp_data, current_hosp = hosp_times[-1], current_hosp[-1]
t_latest_hosp_data_date = t0_simulation + timedelta(days=int(t_latest_hosp_data))
state_hosp_gen = load_data.get_compartment_value_on_date(
fips=fips[:2], compartment="HGen", date=t_latest_hosp_data_date
)
state_hosp_icu = load_data.get_compartment_value_on_date(
fips=fips[:2], compartment="HICU", date=t_latest_hosp_data_date
)
if len(fips) == 5:
# Rescale the county level hospitalizations by the expected
# ratio of county / state hospitalizations from simulations.
# We use ICU data if available too.
county_hosp = load_data.get_compartment_value_on_date(
fips=fips,
compartment="HGen",
date=t_latest_hosp_data_date,
ensemble_results=pyseir_outputs,
)
county_icu = load_data.get_compartment_value_on_date(
fips=fips,
compartment="HICU",
date=t_latest_hosp_data_date,
ensemble_results=pyseir_outputs,
)
current_hosp *= (county_hosp + county_icu) / (state_hosp_gen + state_hosp_icu)
hosp_rescaling_factor = current_hosp / (state_hosp_gen + state_hosp_icu)
# Some states have covidtracking issues. We shouldn't ground ICU cases
# to zero since so far these have all been bad reporting.
if current_icu is not None and current_icu[-1] > 0:
icu_rescaling_factor = current_icu[-1] / state_hosp_icu
else:
icu_rescaling_factor = current_hosp / (state_hosp_gen + state_hosp_icu)
else:
hosp_rescaling_factor = 1.0
icu_rescaling_factor = 1.0
# Iterate through each suppression policy.
# Model output is interpolated to the dates desired for the API.
for i_policy, suppression_policy in enumerate(
[key for key in pyseir_outputs.keys() if key.startswith("suppression_policy")]
):
output_for_policy = pyseir_outputs[suppression_policy]
output_model = pd.DataFrame()
t_list = output_for_policy["t_list"]
t_list_downsampled = range(0, int(max(t_list)), self.output_interval_days)
output_model[schema.DAY_NUM] = t_list_downsampled
output_model[schema.DATE] = [
(t0_simulation + timedelta(days=t)).date().strftime("%m/%d/%y")
for t in t_list_downsampled
]
output_model[schema.TOTAL] = population
output_model[schema.TOTAL_SUSCEPTIBLE] = np.interp(
t_list_downsampled, t_list, output_for_policy["S"]["ci_50"]
)
output_model[schema.EXPOSED] = np.interp(
t_list_downsampled, t_list, output_for_policy["E"]["ci_50"]
)
output_model[schema.INFECTED] = np.interp(
t_list_downsampled,
t_list,
np.add(output_for_policy["I"]["ci_50"], output_for_policy["A"]["ci_50"]),
) # Infected + Asympt.
output_model[schema.INFECTED_A] = output_model[schema.INFECTED]
output_model[schema.INFECTED_B] = hosp_rescaling_factor * np.interp(
t_list_downsampled, t_list, output_for_policy["HGen"]["ci_50"]
) # Hosp General
output_model[schema.INFECTED_C] = icu_rescaling_factor * np.interp(
t_list_downsampled, t_list, output_for_policy["HICU"]["ci_50"]
) # Hosp ICU
# General + ICU beds. don't include vent here because they are also counted in ICU
output_model[schema.ALL_HOSPITALIZED] = np.add(
output_model[schema.INFECTED_B], output_model[schema.INFECTED_C]
)
output_model[schema.ALL_INFECTED] = output_model[schema.INFECTED]
output_model[schema.DEAD] = np.interp(
t_list_downsampled, t_list, output_for_policy["total_deaths"]["ci_50"]
)
final_beds = np.mean(output_for_policy["HGen"]["capacity"])
output_model[schema.BEDS] = final_beds
output_model[schema.CUMULATIVE_INFECTED] = np.interp(
t_list_downsampled,
t_list,
np.cumsum(output_for_policy["total_new_infections"]["ci_50"]),
)
if fit_results:
output_model[schema.Rt] = np.interp(
t_list_downsampled,
t_list,
fit_results["eps"] * fit_results["R0"] * np.ones(len(t_list)),
)
output_model[schema.Rt_ci90] = np.interp(
t_list_downsampled,
t_list,
2 * fit_results["eps_error"] * fit_results["R0"] * np.ones(len(t_list)),
)
else:
output_model[schema.Rt] = 0
output_model[schema.Rt_ci90] = 0
output_model[schema.CURRENT_VENTILATED] = icu_rescaling_factor * np.interp(
t_list_downsampled, t_list, output_for_policy["HVent"]["ci_50"]
)
output_model[schema.POPULATION] = population
# Average capacity.
output_model[schema.ICU_BED_CAPACITY] = np.mean(output_for_policy["HICU"]["capacity"])
output_model[schema.VENTILATOR_CAPACITY] = np.mean(
output_for_policy["HVent"]["capacity"]
)
# Truncate date range of output.
output_dates = pd.to_datetime(output_model["date"])
output_model = output_model[
(output_dates >= datetime(month=3, day=3, year=2020))
& (output_dates < datetime.today() + timedelta(days=90))
]
output_model = output_model.fillna(0)
# Fill in results for the Rt indicator.
try:
rt_results = load_Rt_result(fips)
rt_results.index = rt_results["Rt_MAP_composite"].index.strftime("%m/%d/%y")
merged = output_model.merge(
rt_results[["Rt_MAP_composite", "Rt_ci95_composite"]],
right_index=True,
left_on="date",
how="left",
)
output_model[schema.RT_INDICATOR] = merged["Rt_MAP_composite"]
# With 90% probability the value is between rt_indicator - ci90 to rt_indicator + ci90
output_model[schema.RT_INDICATOR_CI90] = (
merged["Rt_ci95_composite"] - merged["Rt_MAP_composite"]
)
except (ValueError, KeyError) as e:
output_model[schema.RT_INDICATOR] = "NaN"
output_model[schema.RT_INDICATOR_CI90] = "NaN"
output_model[[schema.RT_INDICATOR, schema.RT_INDICATOR_CI90]] = output_model[
[schema.RT_INDICATOR, schema.RT_INDICATOR_CI90]
].fillna("NaN")
# Truncate floats and cast as strings to match data model.
int_columns = [
col
for col in output_model.columns
if col
not in (
schema.DATE,
schema.Rt,
schema.Rt_ci90,
schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90,
)
]
output_model.loc[:, int_columns] = (
output_model[int_columns].fillna(0).astype(int).astype(str)
)
output_model.loc[
:, [schema.Rt, schema.Rt_ci90, schema.RT_INDICATOR, schema.RT_INDICATOR_CI90]
] = (
output_model[
[schema.Rt, schema.Rt_ci90, schema.RT_INDICATOR, schema.RT_INDICATOR_CI90]
]
.fillna(0)
.round(decimals=4)
.astype(str)
)
# Convert the records format to just list(list(values))
output_model = [
[val for val in timestep.values()]
for timestep in output_model.to_dict(orient="records")
]
output_path = get_run_artifact_path(
fips, RunArtifact.WEB_UI_RESULT, output_dir=self.output_dir
)
policy_enum = Intervention.from_webui_data_adaptor(suppression_policy)
output_path = output_path.replace("__INTERVENTION_IDX__", str(policy_enum.value))
with open(output_path, "w") as f:
json.dump(output_model, f)
def map_fips(self, fips):
"""
For a given county fips code, generate the CAN UI output format.
Parameters
----------
fips: str
County FIPS code to map.
"""
if len(fips) == 5:
population = self.population_data.get_county_level(
'USA', state=self.state_abbreviation, fips=fips)
else:
population = self.population_data.get_state_level(
'USA', state=self.state_abbreviation)
logging.info(f'Mapping output to WebUI for {self.state}, {fips}')
pyseir_outputs = load_data.load_ensemble_results(fips)
policies = [
key for key in pyseir_outputs.keys()
if key.startswith('suppression_policy')
]
all_hospitalized_today = None
try:
fit_results = load_inference_result(fips)
except ValueError:
fit_results = None
logging.error(
f'Fit result not found for {fips}: Skipping inference elements'
)
for i_policy, suppression_policy in enumerate(policies):
if suppression_policy == 'suppression_policy__full_containment': # No longer shipping this.
continue
output_for_policy = pyseir_outputs[suppression_policy]
output_model = pd.DataFrame()
if suppression_policy == 'suppression_policy__inferred' and fit_results:
if len(fips) == 5 and fips not in self.df_whitelist.fips:
continue
t0 = datetime.fromisoformat(fit_results['t0_date'])
# Hospitalizations need to be rescaled by the inferred factor to match observations for display.
now_idx = int(
(datetime.today() -
datetime.fromisoformat(fit_results['t0_date'])).days)
total_hosps = output_for_policy['HGen']['ci_50'][
now_idx] + output_for_policy['HICU']['ci_50'][now_idx]
hosp_fraction = all_hospitalized_today / total_hosps
else:
t0 = datetime.today()
hosp_fraction = 1.
t_list = output_for_policy['t_list']
t_list_downsampled = range(0, int(max(t_list)),
self.output_interval_days)
# Col 0
output_model['days'] = t_list_downsampled
# Col 1
output_model['date'] = [
(t0 + timedelta(days=t)).date().strftime('%m/%d/%y')
for t in t_list_downsampled
]
# Col 2
output_model['t'] = population
# Col 3
output_model['b'] = np.interp(t_list_downsampled, t_list,
output_for_policy['S']['ci_50'])
# Col 4
output_model['c'] = np.interp(t_list_downsampled, t_list,
output_for_policy['E']['ci_50'])
# Col 5
output_model['d'] = np.interp(
t_list_downsampled, t_list,
np.add(output_for_policy['I']['ci_50'],
output_for_policy['A']['ci_50'])) # Infected + Asympt.
# Col 6
output_model['e'] = output_model['d']
# Col 7
output_model['f'] = np.interp(
t_list_downsampled, t_list,
output_for_policy['HGen']['ci_50']) # Hosp General
# Col 8
output_model['g'] = np.interp(
t_list_downsampled, t_list,
output_for_policy['HICU']['ci_50']) # Hosp ICU
# Col 9
output_model['all_hospitalized'] = hosp_fraction * np.add(
output_model['f'], output_model['g'])
# Col 10
output_model['all_infected'] = output_model['d']
# Col 11
output_model['dead'] = np.interp(
t_list_downsampled, t_list,
output_for_policy['total_deaths']['ci_50'])
# Col 12
final_beds = np.mean(
output_for_policy['HGen']['capacity']) + np.mean(
output_for_policy['HICU']['capacity'])
output_model['beds'] = final_beds
output_model['cumulative_infected'] = np.interp(
t_list_downsampled, t_list,
np.cumsum(output_for_policy['total_new_infections']['ci_50']))
if fit_results:
output_model['R_t'] = np.interp(
t_list_downsampled, t_list, fit_results['eps'] *
fit_results['R0'] * np.ones(len(t_list)))
output_model['R_t_stdev'] = np.interp(
t_list_downsampled, t_list, fit_results['eps_error'] *
fit_results['R0'] * np.ones(len(t_list)))
else:
output_model['R_t'] = 0
output_model['R_t_stdev'] = 0
# Record the current number of hospitalizations in order to rescale the inference results.
all_hospitalized_today = output_model['all_hospitalized'][0]
# Don't backfill inferences
if suppression_policy != 'suppression_policy__inferred':
backfill = self.backfill_output_model_fips(
fips, t0, final_beds, output_model)
output_model = pd.concat([
backfill, output_model
])[output_model.columns].reset_index(drop=True)
# Truncate date range of output.
output_dates = pd.to_datetime(output_model['date'])
output_model = output_model[
(output_dates > datetime(month=3, day=3, year=2020))
& (output_dates < datetime.today() + timedelta(days=90))]
output_model = output_model.fillna(0)
for col in ['l']:
output_model[col] = 0
output_model['population'] = population
for col in ['m', 'n']:
output_model[col] = 0
# Truncate floats and cast as strings to match data model.
int_columns = [
col for col in output_model.columns
if col not in ('date', 'R_t', 'R_t_stdev')
]
output_model.loc[:,
int_columns] = output_model[int_columns].fillna(
0).astype(int).astype(str)
output_model.loc[:, ['R_t', 'R_t_stdev']] = output_model[[
'R_t', 'R_t_stdev'
]].fillna(0).round(decimals=4).astype(str)
# Convert the records format to just list(list(values))
output_model = [[
val for val in timestep.values()
] for timestep in output_model.to_dict(orient='records')]
output_path = get_run_artifact_path(fips,
RunArtifact.WEB_UI_RESULT,
output_dir=self.output_dir)
policy_enum = Intervention.from_webui_data_adaptor(
suppression_policy)
output_path = output_path.replace('__INTERVENTION_IDX__',
str(policy_enum.value))
with open(output_path, 'w') as f:
json.dump(output_model, f)
def map_fips(self, fips: str) -> None:
"""
For a given fips code, for either a county or state, generate the CAN UI output format.
Parameters
----------
fips: str
FIPS code to map.
"""
log.info("Mapping output to WebUI.", state=self.state, fips=fips)
pyseir_outputs = load_data.load_ensemble_results(fips)
try:
fit_results = load_inference_result(fips)
t0_simulation = datetime.fromisoformat(fit_results["t0_date"])
except (KeyError, ValueError):
log.error("Fit result not found for fips. Skipping...", fips=fips)
return
population = self._get_population(fips)
# Get multiplicative conversion factors to scale model output to fit dataset current values
hosp_rescaling_factor, icu_rescaling_factor = self._get_model_to_dataset_conversion_factors(
t0_simulation=t0_simulation,
fips=fips,
pyseir_outputs=pyseir_outputs,
)
# Iterate through each suppression policy.
# Model output is interpolated to the dates desired for the API.
suppression_policies = [
key for key in pyseir_outputs.keys()
if key.startswith("suppression_policy")
]
for suppression_policy in suppression_policies:
output_for_policy = pyseir_outputs[suppression_policy]
output_model = pd.DataFrame()
t_list = output_for_policy["t_list"]
t_list_downsampled = range(0, int(max(t_list)),
self.output_interval_days)
output_model[schema.DAY_NUM] = t_list_downsampled
output_model[schema.DATE] = [
(t0_simulation + timedelta(days=t)).date().strftime("%m/%d/%y")
for t in t_list_downsampled
]
output_model[schema.TOTAL] = population
output_model[schema.TOTAL_SUSCEPTIBLE] = np.interp(
t_list_downsampled, t_list, output_for_policy["S"]["ci_50"])
output_model[schema.EXPOSED] = np.interp(
t_list_downsampled, t_list, output_for_policy["E"]["ci_50"])
output_model[schema.INFECTED] = np.interp(
t_list_downsampled,
t_list,
np.add(output_for_policy["I"]["ci_50"],
output_for_policy["A"]["ci_50"]),
) # Infected + Asympt.
output_model[schema.INFECTED_A] = output_model[schema.INFECTED]
output_model[
schema.INFECTED_B] = hosp_rescaling_factor * np.interp(
t_list_downsampled, t_list,
output_for_policy["HGen"]["ci_50"]) # Hosp General
raw_model_icu_values = output_for_policy["HICU"]["ci_50"]
interpolated_model_icu_values = np.interp(t_list_downsampled,
t_list,
raw_model_icu_values)
final_derived_model_value = icu_rescaling_factor * interpolated_model_icu_values
output_model[schema.INFECTED_C] = final_derived_model_value
# General + ICU beds. don't include vent here because they are also counted in ICU
output_model[schema.ALL_HOSPITALIZED] = np.add(
output_model[schema.INFECTED_B],
output_model[schema.INFECTED_C])
output_model[schema.ALL_INFECTED] = output_model[schema.INFECTED]
output_model[schema.DEAD] = np.interp(
t_list_downsampled, t_list,
output_for_policy["total_deaths"]["ci_50"])
final_beds = np.mean(output_for_policy["HGen"]["capacity"])
output_model[schema.BEDS] = final_beds
output_model[schema.CUMULATIVE_INFECTED] = np.interp(
t_list_downsampled,
t_list,
np.cumsum(output_for_policy["total_new_infections"]["ci_50"]),
)
if fit_results:
output_model[schema.Rt] = np.interp(
t_list_downsampled,
t_list,
fit_results["eps"] * fit_results["R0"] *
np.ones(len(t_list)),
)
output_model[schema.Rt_ci90] = np.interp(
t_list_downsampled,
t_list,
2 * fit_results["eps_error"] * fit_results["R0"] *
np.ones(len(t_list)),
)
else:
output_model[schema.Rt] = 0
output_model[schema.Rt_ci90] = 0
output_model[
schema.CURRENT_VENTILATED] = icu_rescaling_factor * np.interp(
t_list_downsampled, t_list,
output_for_policy["HVent"]["ci_50"])
output_model[schema.POPULATION] = population
# Average capacity.
output_model[schema.ICU_BED_CAPACITY] = np.mean(
output_for_policy["HICU"]["capacity"])
output_model[schema.VENTILATOR_CAPACITY] = np.mean(
output_for_policy["HVent"]["capacity"])
# Truncate date range of output.
output_dates = pd.to_datetime(output_model["date"])
output_model = output_model[
(output_dates >= datetime(month=3, day=3, year=2020))
& (output_dates < datetime.today() + timedelta(days=90))]
output_model = output_model.fillna(0)
# Fill in results for the Rt indicator.
try:
rt_results = load_Rt_result(fips)
rt_results.index = rt_results[
"Rt_MAP_composite"].index.strftime("%m/%d/%y")
merged = output_model.merge(
rt_results[["Rt_MAP_composite", "Rt_ci95_composite"]],
right_index=True,
left_on="date",
how="left",
)
output_model[schema.RT_INDICATOR] = merged["Rt_MAP_composite"]
# With 90% probability the value is between rt_indicator - ci90
# to rt_indicator + ci90
output_model[schema.RT_INDICATOR_CI90] = (
merged["Rt_ci95_composite"] - merged["Rt_MAP_composite"])
except (ValueError, KeyError) as e:
log.warning("Clearing Rt in output for fips.",
fips=fips,
exc_info=e)
output_model[schema.RT_INDICATOR] = "NaN"
output_model[schema.RT_INDICATOR_CI90] = "NaN"
output_model[[schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90]] = output_model[[
schema.RT_INDICATOR, schema.RT_INDICATOR_CI90
]].fillna("NaN")
# Truncate floats and cast as strings to match data model.
int_columns = [
col for col in output_model.columns if col not in (
schema.DATE,
schema.Rt,
schema.Rt_ci90,
schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90,
)
]
output_model.loc[:, int_columns] = (
output_model[int_columns].fillna(0).astype(int).astype(str))
output_model.loc[:, [
schema.Rt, schema.Rt_ci90, schema.RT_INDICATOR, schema.
RT_INDICATOR_CI90
]] = (output_model[[
schema.Rt, schema.Rt_ci90, schema.RT_INDICATOR,
schema.RT_INDICATOR_CI90
]].fillna(0).round(decimals=4).astype(str))
# Convert the records format to just list(list(values))
output_model = [[
val for val in timestep.values()
] for timestep in output_model.to_dict(orient="records")]
output_path = get_run_artifact_path(fips,
RunArtifact.WEB_UI_RESULT,
output_dir=self.output_dir)
policy_enum = Intervention.from_webui_data_adaptor(
suppression_policy)
output_path = output_path.replace("__INTERVENTION_IDX__",
str(policy_enum.value))
with open(output_path, "w") as f:
json.dump(output_model, f)
def get_intervention_for_state(state):
# TODO: read this from a dataset class
interventions_url = "https://raw.githubusercontent.com/covid-projections/covid-projections/master/src/assets/data/interventions.json"
interventions = requests.get(interventions_url).json()
return Intervention.from_str(interventions[state])
