simpliml 1.5.0

SimpliML is a Python machine-learning library designed to simplify the data lifecycle. The project brings data preparation, exploratory analysis, forecasting, model evaluation, and result interpretation together in one consistent workflow. It is intended for data scientists, analysts, students, and developers who want practical forecasting tools without building the entire modeling pipeline from scratch.

SimpliML currently focuses on time-series forecasting. It accepts historical observations, prepares regular date-based datasets, evaluates multiple model families, and returns comparable results for selecting a forecast. The package is designed to work with familiar Python data tools, especially pandas DataFrame objects.

Key Features

• Time-Series Data Preparation
  Convert date columns, generate regular date ranges, preserve historical observations, handle missing dates, add calendar features, and create future periods for prediction.

• Data Analysis
  Explore trends, seasonality, distributions, autocorrelation, and partial autocorrelation with interactive and statistical visualizations.

• Multiple Model Families
  Train statistical models through statsmodels, ensemble and regression models through scikit-learn, and neural-network forecasting models through PyTorch.

• Flexible Execution
  Run model evaluation sequentially, with multiple threads, or with multiple processes depending on the workload and environment.

• Model Comparison
  Compare successful models using MAE, MSE, RMSE, RMSLE, MAPE, R2, SMAPE, MFE, NMSE, and Theil's U statistic.

• Forecast Result Review
  Select the best available model by MAPE or inspect a specific model, then combine historical, test, and forecast values for analysis and visualization.

Forecasting Workflow

The package follows a simple end-to-end workflow:

1. Prepare historical date and target-value data.
2. Generate a regular time series and future dates.
3. Analyze trends, seasonality, and statistical behavior.
4. Train supported forecasting models with a selected seasonal period.
5. Evaluate model performance on held-out observations.
6. Select the best successful model and review its forecast.

The public time-series API includes generateTSData, analysisData, runModel, runThreadModel, runProcessModel, buildModel, and modelResult.

Supported Data and Models

Input data is represented with pandas dataframes. The forecasting workflow uses a date column and a numeric target column named Data after preparation. Regular data frequencies such as daily, monthly, and quarterly observations are supported.

The model pipeline can include:

• Statistical forecasting models from statsmodels, including ARIMA, AutoReg, exponential smoothing, Holt, SARIMA/SARIMAX, VAR, and VARMAX.
• Scikit-learn regression and ensemble estimators for feature-based forecasting.
• PyTorch LSTM-based forecasting models for neural-network workflows.

Model execution can be configured with FAST for a reduced search or BEST for a broader search. The selected seasonal period should match the data frequency, such as 7 for weekly seasonality in daily data or 12 for yearly seasonality in monthly data.

Intended Uses

SimpliML can support demand and sales forecasting, operational planning, monthly or daily KPI prediction, time-series experimentation, classroom projects, and rapid comparisons of statistical and machine-learning models. It is a forecasting toolkit rather than a replacement for domain validation, data quality checks, or production monitoring.

Documentation

The official documentation is available on Read the Docs.

Project Links

• PyPI
• GitHub Repository
• Issue Tracker