orient='records')
def dataframe_add_key(data: pandas.DataFrame):
''' Adds the Key column to all rows of the dataframe '''
data['Key'] = data.apply(compute_record_key, axis=1)
return data.set_index('Key')
# Root path of the project
ROOT = Path(os.path.dirname(__file__)) / '..'
# Read the full data file
data = read_csv(ROOT / 'output' / 'data.csv')
for col in data.columns:
data[col] = series_converter(data[col])
# Backwards compatibility: China dataset and Region -> RegionName
china = data[data['CountryCode'] == 'CN']
china = china[~china['RegionCode'].isna()]
dataframe_split(china, ('RegionCode', ), ROOT, 'cn')
china['Region'] = china['RegionName']
dataframe_split(china, ('RegionCode', ), ROOT, 'china')
# Backwards compatibility: Usa dataset and RegionName -> RegionCode
usa = data[data['CountryCode'] == 'US']
usa = usa[~usa['RegionCode'].isna()]
dataframe_split(usa, ('RegionCode', ), ROOT, 'us')
usa['Region'] = usa['RegionCode']
dataframe_split(usa, ('RegionCode', ), ROOT, 'usa')
# Estimate daily counts per category assuming ratio is constant
df['NewCases'] = df['Confirmed'].diff().astype('Int64')
df['NewDeaths'] = df['Deaths'].diff().astype('Int64')
df['NewMild'] = df['NewCases'] * mild_ratio
df['NewSevere'] = df['NewCases'] * severe_ratio
df['NewCritical'] = df['NewCases'] * critical_ratio
df = df[['NewCases', 'NewDeaths', 'NewMild', 'NewSevere', 'NewCritical']]
# Compute the rolling windows for count of active (current) categories
df['CurrentlyMild'] = df['NewMild'].rolling(
round(mild_recovery_days)).sum()
df['CurrentlySevere'] = df['NewSevere'].rolling(
round(severe_recovery_days)).sum()
df['CurrentlyCritical'] = df['NewCritical'].rolling(
round(critical_recovery_days)).sum()
# Get rid of the first columns which are useless because of the windowing function
df = df.iloc[
max(mild_recovery_days, severe_recovery_days, critical_recovery_days):]
# Make sure all columns have the appropriate type
for col in df.columns:
df[col] = series_converter(df[col])
# Output resulting dataframe
if print_header_flag:
df.to_csv(sys.stdout)
print_header_flag = False
else:
df.to_csv(sys.stdout, header=None)
# Early exit: If there are less than DATAPOINT_COUNT output datapoints
if len(subset) < DATAPOINT_COUNT - PREDICT_WINDOW: continue
# Perform forecast
forecast_data = compute_forecast(subset['Confirmed'], PREDICT_WINDOW)
# Capture only the last DATAPOINT_COUNT days
forecast_data = forecast_data.sort_index().iloc[-DATAPOINT_COUNT:]
# Fill out the corresponding index in the output forecast
for idx in forecast_data.index:
df_forecast.loc[(idx, key), 'ForecastDate'] = forecast_date
df_forecast.loc[(idx, key),
'Estimated'] = '%.03f' % forecast_data.loc[idx]
if idx in subset.index:
df_forecast.loc[(idx, key),
'Confirmed'] = int(subset.loc[idx, 'Confirmed'])
# Do data cleanup here
data = df_forecast.reset_index()
forecast_columns = ['ForecastDate', 'Date', 'Key', 'Estimated', 'Confirmed']
data = data.sort_values(['Key', 'Date'])[forecast_columns]
# Make sure the core columns have the right data type
for col in data.columns:
data[col] = series_converter(data[col])
# Output resulting dataframe
data.to_csv(sys.stdout, index=False)