Skip to main content

Basic Time Series helpers

These are the helpers which are going to be useful regardless of the model you want to use

Data Reshaping​

For simple Time Series forecasting, you generally want a dataframe or series with just the metric and the date in the index, evenly spaced apart

For example, The count of records (Metric) in each month (Date)

We have a helper function which can help you transform your data into this format

from scripts.helpers.time_series_helpers import reshape_time_series_data
def reshape_time_series_data(
  pdf: ps.DataFrame,
  date_col: str,
  var_cols: list,
  dateformat: str
)

returns ps.DataFrame
  • pdf: Your Dataframe
  • date_col: The name of the date column, this will be converted into the index
  • var_cols: The name of the metrics you want to use, in a list. A lot of the time you only want a single metric
  • dateformat: Format of your date. For example YYYY/MM/DD would be %Y/%m%d

Usage Example​

Lets say you have a dataframe, with various metric columns but you wanted to just keep one of them, in particular the one called total_call_time

reshaped_df = reshape_time_series_data(pdf,"date",["total_call_time"], "%Y/%M/%d")

Finding Start and End Date​

Some forecast functions will need a start and end date. If you need to quickly find these out from your data and the amount of prediction points, we have a function for that

from scripts.helpers.time_series_helpers import reshape_time_series_data
def get_start_end_date(
  dataframe: pd.DataFrame,
  period: str,
  forecast_count: int
)

returns start_date date,end_date date

So an example of calling this would be

start_date, end_date = get_start_end_date(df,"M",12)

If the final date in the index of DF is "01/12/2022"

start_date = "01/01/2023"

end_date = "01/12/2023"

Making Train Test subsets by count​

When evaluating models, we use something called Train Test as a testing methodology

Why would we want to do this?​

This function will split your data into 2 parts for this purpose

from scripts.helpers.time_series_helpers import get_train_test_subsets
def get_train_test_subsets(
  time_series: ps.DataFrame,
  periods: int):

Returns:
  Train(DataFrame),Test(DataFrame)

Usage Example​

train, test = get_train_test_subsets(pdf,6)

If pdf is of length 10, train will be of length 4 and test will be of length 6 If pdf is of length 100, train will be of length 94 and test will be of length 6