This package proposes an easy application of the paper Deep Regressions Ensemble.
PAPER: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4049493
pip install DeepRegressionEnsembles
- Antoine Didisheim (Swiss Finance Institute, antoine.didisheim@unil.ch)
- Bryan T. Kelly (Yale SOM; AQR Capital Management, LLC; National Bureau of Economic Research (NBER), bryan.kelly@yale.edu)
- Semyon Malamud (Ecole Polytechnique Federale de Lausanne; Centre for Economic Policy Research (CEPR); Swiss Finance Institute, semyon.malamud@epfl.ch)
See demo.py for a simple example.
The main class DeepRegressionEnsemble() can be defined with the following parameters:
- depth: maximum depth
- k: number of ensembles
- p: number of features per ensemble
- lbd: a grid of lambda
- seed: starting TensorFlow seed
- save_dir: specify one to save the model in training (or as a default loading directory)
- perf_measure: 'R2', 'MSE', or 'MAE' --> only define how we print performance
- gamma_min: minimum value of the scaling parameters
- gamma_max: maximum value of the scaling parameter
- gamma_type: 'RANDOM' or 'GRID
- output_layer_dim_reduction: if none, the output layer is as in the paper otherwise, you can specify the dimension to reduce it too
- max_para: the maximum number of independent threads in the TF while loops
- verbose: define whether or not to print progress in training