Predict market next state:

• 32-dimensional anonymized time series features with low autocorrelation and SNR
• Each time series consists of 1000 timesteps
• Predictions should be provided for steps >= 101
• First 100 steps can be used to "warm up" the model
• Features like EMA and spectral entropy may be useful (according to Granger causality test), but adding them to the model can be tricky

---

Install dependencies:

poetry install --no-root

---