fnHoltWinter.pq

(
    // table to pass in
    tbl as table, 
    // name of the date column
    date_column as text, 
    // name of the column you are projecting on
    projection_column as text, 
    // "add", "mul", "None"
    trend as text, 
    // "add", "mul", "None"
    seasonal as text, 
    // how many forecast periods you want to project
    forecast_periods as number, 
    // how many periods constitute a season (e.g. if the data is monthly, then 12 periods could be a year)
    seasonal_periods as number
)=>

let 
    python = 
"
import pandas as pd
from statsmodels.tsa.holtwinters import ExponentialSmoothing

# Ensure dataset is sorted
dataset[""" & date_column & """] = pd.to_datetime(dataset[""" & date_column & """])
dataset = dataset.sort_values(""" & date_column & """)

# Define the forecast period (e.g., next 6 months)
forecast_periods = " & Text.From(forecast_periods) & "

# Fit the Holt-Winters model (Triple Exponential Smoothing)
model = ExponentialSmoothing(dataset[""" & projection_column& """], trend=""" & trend & """, seasonal=""" & seasonal & """, seasonal_periods=" & Text.From(seasonal_periods) & ")
fitted_model = model.fit()

# Create forecast
future_dates = pd.date_range(start=dataset[""" & date_column & """].max(), periods=forecast_periods + 1, freq=""M"")[1:]
forecast_values = fitted_model.forecast(forecast_periods)

# Create DataFrame with forecast
forecast_df = pd.DataFrame({""" & date_column & """: future_dates, "" ForecastedPayroll "": forecast_values})

# Append forecast to original dataset
output_df = pd.concat([dataset[[""" & date_column & """, """ & projection_column & """]], forecast_df], ignore_index=True)

# Return the result to Power BI
output_df",
    run = Python.Execute(python,[dataset=tbl])
in 
    run