Quickstart#

Get from zero to a working forecast in under a minute. This page provides a copy-paste-ready minimal example using OpenSTEF’s workflow presets. No explanations—just working code. For the why behind each step, see Your First Forecast.

        graph LR
    A[Configure WorkflowConfig] --> B[create_forecasting_workflow]
    B --> C[workflow.fit]
    C --> D[workflow.predict]
    D --> E[(TimeSeriesDataset)]
    classDef primary fill:#00D9C5,stroke:#1E3A5F,stroke-width:2px,color:#000
    classDef secondary fill:#1E3A5F,stroke:#00D9C5,stroke-width:2px,color:#fff
    classDef accent fill:#e6f7f5,stroke:#00D9C5,stroke-width:2px,color:#000
    class A accent
    class B,C,D primary
    class E secondary

Prerequisites#

Make sure OpenSTEF is installed. If not, see Installation.

pip install openstef-models

Minimal Working Example#

The following script creates synthetic energy load data, configures a forecasting workflow, trains a model, and produces a forecast—all in one go.

import numpy as np
import pandas as pd
from datetime import timedelta

from openstef_models.presets.forecasting_workflow import (
    ForecastingWorkflowConfig,
    create_forecasting_workflow,
)
from openstef_models.data import TimeSeriesDataset

# --- Step 1: Create synthetic data ---
# 30 days of 15-minute resolution load data with a daily pattern
periods = 30 * 24 * 4  # 30 days at 15-min intervals
index = pd.date_range("2024-01-01", periods=periods, freq="15min", tz="UTC")

np.random.seed(42)
hour = index.hour + index.minute / 60.0
load = 50 + 20 * np.sin(2 * np.pi * hour / 24) + np.random.normal(0, 2, periods)

df = pd.DataFrame({"load": load}, index=index)
df.index.name = "datetime"

# --- Step 2: Configure the workflow ---
config = ForecastingWorkflowConfig(
    target_column="load",
    horizons=[timedelta(hours=24)],
    selected_features=["load"],
    model_type="xgb_quantile",
    quantiles=(0.1, 0.5, 0.9),
)

# --- Step 3: Create the workflow ---
workflow = create_forecasting_workflow(config)

# --- Step 4: Split data into train / predict sets ---
train_df = df.iloc[:-96]   # all but last day
predict_df = df.iloc[-96:]  # last day (96 quarter-hours)

train_data = TimeSeriesDataset(train_df)
predict_data = TimeSeriesDataset(predict_df)

# --- Step 5: Fit the model ---
fit_result = workflow.fit(train_data)

# --- Step 6: Generate a forecast ---
forecast = workflow.predict(predict_data)

# --- Step 7: Inspect results ---
print(forecast.to_dataframe().head())

What Just Happened#

In six lines of meaningful code you:

  • Created a ForecastingWorkflowConfig that defines the target column, forecast horizon, features, model type, and output quantiles.

  • Called create_forecasting_workflow which assembled the full pipeline—preprocessing, feature engineering, model, and postprocessing—from that single config object.

  • Trained the model with workflow.fit().

  • Produced probabilistic forecasts with workflow.predict().

The output is a TimeSeriesDataset containing columns for each requested quantile (p10, p50, p90), giving you both a point forecast and confidence intervals.

Key Configuration Options#

The ForecastingWorkflowConfig accepts many parameters. Here are the ones you’ll adjust most often:

config = ForecastingWorkflowConfig(
    target_column="load",                    # column name to forecast
    horizons=[timedelta(hours=24)],          # forecast horizons
    selected_features=["load", "temperature", "wind_speed"],  # input features
    model_type="xgb_quantile",              # model backend
    quantiles=(0.05, 0.25, 0.5, 0.75, 0.95),  # output quantiles
    completeness_threshold=0.7,             # minimum data completeness
    flatliner_threshold=24,                 # hours before flagging flatline
    verbosity=1,                            # 0=silent, 1=warning, 2=info, 3=debug
)

Next Steps#