feature-selection
This repository contains implementation of permutation importance based feature selection. To run the pipeline for training an evaluation of the framework run:
$ python3 main.py --test_no=1It needs a file located in ./LIME/data/constituents.csv with the stock set it will build the feature selection on.
Outputs
A metric file found in ./LIME/data/LOOC_metrics_cr_{test_no}.csv that has the following format:
| MSE_baseline | MAE_baseline | walk | ticker | MSE_pi | MAE_pi | model | selection_error | removed_FI | removed_error | removed_column |
|---|---|---|---|---|---|---|---|---|---|---|
| 0.00019 | 0.01085 | 1.0 | FP.PA | 0.00027 | 0.01210 | pi | -1.8409e-05 | 0.00827 | Returns_3 |
with one line for each tuple (walk,ticker,data_type)
Example command
$ python3 main.py
--data_type pi --start_date 2007-01-01 --end_date 2017-01-01 --no_walks 7
--no_features 1 --train_length 4 --test_length 1
--num_rounds 50 --test_no 11