def rolling_mean_by_date_by_group(data: dd = None,
groupby_columns: List[str] = None,
metric_columns: List[str] = None,
date_column: str = None,
window: int = None) -> dd:
"""
Split input dateframe into groups and preform a rolling average on the metric columns for each group
:param data: input dataframe
:param groupby_columns: list of columns to group by
:param metric_columns: columns to calculate rolling average on
:param date_column: name of date column
:param window: window size to be used on rolling average
:return: modified dask dataframe
"""
data = data.set_index(date_column, sorted=True)
output_schema = dict(data.dtypes)
for metric_column in metric_columns:
output_schema[f'{metric_column}_rolling_mean'] = 'float32'
output_schema = list(output_schema.items())
data = data.groupby(by=groupby_columns).apply(
lambda df_g: rolling_mean_by_date(
data=df_g, metric_columns=metric_columns, window=window),
meta=output_schema)
data = data.reset_index().rename(columns={'index': date_column})
return data
def fill_missing_dates(data: dd = None,
date_column: str = None,
fill_method: str = None,
columns=None,
date_range: Tuple[str] = None,
fill_value=None,
groupby_columns=None) -> dd:
"""
Preform date fill on single group
"""
all_dates = pd.date_range(date_range[0], date_range[1])
metric_data = data[[
col for col in data.columns if col not in groupby_columns
]]
data = data[groupby_columns].reindex(all_dates, method='nearest')
metric_data = metric_data.reindex(all_dates,
method=fill_method,
fill_value=fill_value)
data = dd.merge(data, metric_data, left_index=True, right_index=True)
data = data.reset_index().rename(columns={'index': date_column})[columns]
return data
def yoy_percent_change_by_group(data: dd = None,
groupby_columns: List[str] = None,
metric_columns: List[str] = None,
date_column: str = None) -> dd:
"""
Split dataframe into groups and calculate year over year percent change for the etric columns in each group
:param data: input dataframe
:param groupby_columns: list of columns to group by
:param metric_columns: columns to calculate rolling average on
:param date_column: name of date column
:return: modified dataframe
"""
data = data.set_index(date_column, sorted=True)
output_schema = dict(data.dtypes)
for metric_column in metric_columns:
output_schema[f'{metric_column}_yoy_pct_change'] = 'float32'
output_schema = list(output_schema.items())
data = data.groupby(by=groupby_columns).apply(
lambda df_g: yoy_percent_change(data=df_g,
metric_columns=metric_columns),
meta=output_schema)
data = data.reset_index().rename(columns={'index': date_column})
return data