tools/forecaster.py

import pandas as pd
import plotly.graph_objects as go
from statsmodels.tsa.arima.model import ARIMA

def forecast_tool(file_path: str, date_col: str) -> str:
    """
    Forecast next 3 periods of 'Sales'. Returns text summary and saves forecast_plot.png.
    """
    df = pd.read_csv(file_path)

    try:
        df[date_col] = pd.to_datetime(df[date_col])
    except Exception:
        return f"❌ Column '{date_col}' cannot be parsed as dates."

    if "Sales" not in df.columns:
        return "❌ CSV must contain a 'Sales' column."

    df.set_index(date_col, inplace=True)
    model = ARIMA(df["Sales"], order=(1, 1, 1))
    model_fit = model.fit()
    forecast = model_fit.forecast(steps=3)

    # Interactive Plotly forecast with confidence interval
    conf_int = model_fit.get_forecast(steps=3).conf_int()
    future_index = forecast.index

    fig = go.Figure()
    fig.add_scatter(x=df.index, y=df["Sales"], mode="lines", name="Sales")
    fig.add_scatter(x=future_index, y=forecast, mode="lines", name="Forecast")
    fig.add_scatter(
        x=future_index,
        y=conf_int.iloc[:, 0],
        mode="lines",
        fill=None,
        line=dict(width=0),
        showlegend=False,
    )
    fig.add_scatter(
        x=future_index,
        y=conf_int.iloc[:, 1],
        mode="lines",
        fill="tonexty",
        name="95% CI",
        line=dict(width=0),
    )
    fig.update_layout(title="Sales Forecast", template="plotly_dark")
    fig.write_image("forecast_plot.png")

    return forecast.to_frame(name="Forecast").to_string()