TSPLab (Time Series Prediction Lab) is a command center for forecasting experiment management. Audit, monitor, tune, and compare time series forecasting experiments — with built-in support for NSE optimization and power market use cases.

Forecasting experiments are hard to track: which hyperparameters were used? How does this model compare to last week's? Is the experiment reproducible? TSPLab gives you a structured way to audit, monitor, tune, and compare forecasting experiments.

• Experiment Audit — structured logging of all experiment parameters and results
• Monitor — real-time tracking of experiment progress and metric convergence
• Tuner — hyperparameter optimization with experiment comparison
• Comparator — side-by-side model comparison with NSE/MAE/R2 metrics
• Reproducibility — every experiment is fully parameterized and replayable
• Power Market Ready — built-in support for electricity price forecasting benchmarks

# Clone and install
git clone https://github.com/disdorqin/tsplab.git
cd tsplab
pip install -e ".[dev]"

# Run an experiment
python -m tsplab experiment run --config examples/lstm_electricity.yaml

# Compare experiments
python -m tsplab compare --exp1 runs/001 --exp2 runs/002

# Launch monitor dashboard
python -m tsplab monitor --port 8080

# examples/lstm_electricity.yaml
model: LSTM
data: electricity_price.csv
features:
  - lag_24
  - lag_168
  - day_of_week
  - temperature
hyperparams:
  hidden_size: 64
  num_layers: 2
  learning_rate: 0.001
  epochs: 100
metrics:
  - NSE
  - MAE
  - R2
output: runs/lstm_v1

• Experiment audit logging
• Basic monitor
• Hyperparameter tuner
• Experiment comparator dashboard
• Integration with DARIS for automated experiment pipelines
• Cloud experiment tracking

Python · pandas · numpy · PyTorch · scikit-learn · YAML

See CONTRIBUTING.md.

MIT — see LICENSE.