Financial models (fynance.models)¶
Several financial models — deep learning, econometric and statistical architectures.
Scaled dot-product and multi-head attention modules for Transformer-based architectures.
Time-series models: MA, ARMA, ARMA-GARCH, ARMAX-GARCH.
Multi-layer perceptron and base class for PyTorch neural network models.
RNN, GRU and LSTM models with walk-forward training support.
Causal dilated convolutional network for sequences.
Causal Transformer encoder with positional encoding.
Direction + magnitude stacking with an out-of-fold meta-model.
Deep ensembles and MC Dropout predictive-uncertainty wrappers.
Differentiable Sharpe/Sortino/Calmar/Omega/directional losses.
Sample weighting and early stopping.
Walk-forward evaluation wrappers for time-series models.
Grid/random hyperparameter search scored on purged walk-forward folds.