def get_predictions(start_date_str, end_date_str, pres_df, countries=None):
# Concatenate prescriptions with historical data
raw_df = get_raw_data(HIST_DATA_FILE_PATH)
hist_df = generate_scenario(start_date_str,
end_date_str,
raw_df,
countries=countries,
scenario='Historical')
start_date = pd.to_datetime(start_date_str, format='%Y-%m-%d')
hist_df = hist_df[hist_df.Date < start_date]
ips_df = pd.concat([hist_df, pres_df])
with tempfile.NamedTemporaryFile() as tmp_ips_file:
# Write ips_df to file
ips_df.to_csv(tmp_ips_file.name)
with tempfile.NamedTemporaryFile() as tmp_pred_file:
# Run script to generate predictions
output_str = subprocess.check_output([
'python3', PREDICT_MODULE, '--start_date', start_date_str,
'--end_date', end_date_str, '--interventions_plan',
tmp_ips_file.name, '--output_file', tmp_pred_file.name
],
stderr=subprocess.STDOUT)
# Print output from running script
print(output_str.decode("utf-8"))
# Load predictions to return
# print(f"Generating predictions from {start_date_str} to {start_date_str} from {tmp_pred_file.name} . {tmp_ips_file.name}..")
# predict(start_date_str, start_date_str, tmp_ips_file.name, tmp_pred_file.name)
df = pd.read_csv(tmp_pred_file)
return df
def test_generate_scenario_mind_the_gap_freeze_dates_mismatch(self):
# Check scenario contains all days, for 2 countries, where 1 country has 1 more day of data than the other
# Last known date:
# - Belgium: 20201103
# - Brazil: 20201104
# Make sure we don't skip a day
start_date_str = "2021-01-01"
end_date_str = "2021-01-31"
countries = ["Belgium", "Brazil"]
dates_mismatch_df = get_raw_data(DATES_MISMATCH_DATA_FILE,
latest=False)
scenario_df = generate_scenario(start_date_str,
end_date_str,
dates_mismatch_df,
countries,
scenario="Freeze")
self.assertIsNotNone(scenario_df)
# Misleading name but checks the elements, regardless of order
self.assertCountEqual(countries, scenario_df.CountryName.unique(),
"Not the requested countries")
# Inception is 2020-01-01. 366 days for 2020 + 31 for Jan 2021
nb_days_since_inception = 397
# For each country, assert the scenario contains the expected number of days
for country in countries:
all_regions = dates_mismatch_df[dates_mismatch_df.CountryName ==
country].RegionName.unique()
for region in all_regions:
ips_gdf = scenario_df[(scenario_df.CountryName == country)
& (scenario_df.RegionName == region)]
self.assertEqual(
nb_days_since_inception, len(ips_gdf),
f"Not the expected number of days"
f" for {country} / {region}")
def get_predictions(start_date_str, end_date_str, pres_df, countries=None):
# Concatenate prescriptions with historical data
raw_df = get_raw_data(HIST_DATA_FILE_PATH)
hist_df = generate_scenario(start_date_str,
end_date_str,
raw_df,
countries=countries,
scenario='Historical')
start_date = pd.to_datetime(start_date_str, format='%Y-%m-%d')
hist_df = hist_df[hist_df.Date < start_date]
ips_df = pd.concat([hist_df, pres_df])
with tempfile.NamedTemporaryFile() as tmp_ips_file:
xp = XPrizePredictor()
ips_df.to_csv(tmp_ips_file.name)
x = xp.predict(start_date_str, end_date_str, tmp_ips_file.name)
return x
def get_predictions(start_date_str, end_date_str, pres_df, countries=None):
# Concatenate prescriptions with historical data
raw_df = get_raw_data(HIST_DATA_FILE_PATH)
hist_df = generate_scenario(start_date_str, end_date_str, raw_df,
countries=countries, scenario='Historical')
start_date = pd.to_datetime(start_date_str, format='%Y-%m-%d')
hist_df = hist_df[hist_df.Date < start_date]
ips_df = pd.concat([hist_df, pres_df])
# Write ips_df to file
ips_df.to_csv(TMP_PRESCRIPTION_FILE)
# Use full path of the local file passed as ip_file
ip_file_full_path = os.path.abspath(TMP_PRESCRIPTION_FILE)
# Go to covid-xprize root dir to access predict script
wd = os.getcwd()
os.chdir("../../../..")
# Run script to generate predictions
output_str = subprocess.check_output(
[
'python', PREDICT_MODULE,
'--start_date', start_date_str,
'--end_date', end_date_str,
'--interventions_plan', ip_file_full_path,
'--output_file', TMP_PRED_FILE_NAME
],
stderr=subprocess.STDOUT
)
# Print output from running script
print(output_str.decode("utf-8"))
# Load predictions to return
df = pd.read_csv(TMP_PRED_FILE_NAME)
# Return to prescriptor dir
os.chdir(wd)
return df
def generate_costs(distribution='ones'):
"""
Returns df of costs for each IP for each geo according to distribution.
Costs always sum to #IPS (i.e., len(IP_COLUMNS)).
Available distributions:
- 'ones': cost is 1 for each IP.
- 'uniform': costs are sampled uniformly across IPs independently
for each geo.
"""
assert distribution in ['ones', 'uniform'], \
f'Unsupported distribution {distribution}'
df = get_raw_data(DATA_FILE, latest=False)
# Reduce df to one row per geo
df = df.groupby(['CountryName', 'RegionName']).mean().reset_index()
# Reduce to geo id info
df = df[['CountryName', 'RegionName']]
if distribution == 'ones':
df[IP_COLUMNS] = 1
elif distribution == 'uniform':
# Generate weights uniformly for each geo independently.
nb_geos = len(df)
nb_ips = len(IP_COLUMNS)
samples = np.random.uniform(size=(nb_ips, nb_geos))
weights = nb_ips * samples / samples.sum(axis=0)
df[IP_COLUMNS] = weights.T
# Round weights for better readability with neglible loss of generality.
df = df.round(2)
return df
def setUpClass(cls):
# Load the csv data only once
cls.latest_df = get_raw_data(DATA_FILE, latest=True)
# In[29]:
from datetime import datetime, timedelta
start_date = datetime.now() + timedelta(days=7)
start_date_str = start_date.strftime('%Y-%m-%d')
end_date = start_date + timedelta(days=180)
end_date_str = end_date.strftime('%Y-%m-%d')
print(f"Start date: {start_date_str}")
print(f"End date: {end_date_str}")
# In[30]:
from covid_xprize.validation.scenario_generator import get_raw_data, generate_scenario, NPI_COLUMNS
DATA_FILE = 'data/OxCGRT_latest.csv'
latest_df = get_raw_data(DATA_FILE, latest=True)
scenario_df = generate_scenario(start_date_str,
end_date_str,
latest_df,
countries=None,
scenario="Freeze")
scenario_file = "predictions/180_days_future_scenario.csv"
scenario_df.to_csv(scenario_file, index=False)
print(f"Saved scenario to {scenario_file}")
# ### Check it
# In[ ]:
get_ipython().run_cell_magic(
'time', '',
import pandas as pd
from datetime import datetime, timedelta
from covid_xprize.validation.scenario_generator import get_raw_data, generate_scenario, NPI_COLUMNS
#start_date = datetime.now() + timedelta(days=7)
#start_date_str = start_date.strftime('%Y-%m-%d')
start_date_str = '2020-10-01'
start_date = pd.to_datetime(start_date_str, format='%Y-%m-%d')
end_date = start_date + timedelta(days=180)
end_date_str = end_date.strftime('%Y-%m-%d')
print(f"Start date: {start_date_str}")
print(f"End date: {end_date_str}")
DATA_FILE = '../data/OxCGRT_latest.csv'
latest_df = get_raw_data(DATA_FILE, latest=False)
scenario_df = generate_scenario(start_date_str,
end_date_str,
latest_df,
countries=None,
scenario="Freeze")
scenario_file = "180_days_future_scenario.csv"
scenario_df.to_csv(scenario_file, index=False)
print(f"Saved scenario to {scenario_file}")