Description
Final Policy Evaluation
ECR
`python3 MDP_policy.py -input Training_data.csv` will run the MDP program and get the best ECR value along with its policy.
IS
`python3 MDP_policy.py -input Training_data_is.csv` will run the MDP program and get the best IS value along with its policy.
Necessary packages
“`
numpy
pandas
pymdptoolbox
seaborn
matplotlib
pyswarms
sklearn
“`
All Data Used
`binned_{2,3,4,5}_reorder.csv`
All Logs
All the CSV files in `bin_{2,3,4,5}_history`
Feature Discretization
` python3 discretize.py` will discretize the columns. If you got the error of `ImportError: cannot import name ‘KBinsDiscretizer’`, please upgrade your `sklearn` to the latest version (at least >= 0.20).
Feature Selection
Genetic Algorithm
`python3 repeat.py` will call `genetic.py` to run some iterations of genetic algorithm to select up to 8 features. The log is saved in `bin_NUM_history` folder where NUM is the number of bins from 2 to 5. When it runs, a CSV file named `temp.csv` will be created as input to `MDP_policy.py` to evaluate the selected features.
Please change the parameters in `repeat.py` to do feature selection using other number of bins and number of features.
PSO Algorithm
`python3 pso_pyswarms_control.py` calls `pso_pyswarms.py` to further run the algorithm with the features that were output in the previous run.
The parameters are clearly defined in `python3 pso_pyswarms_control.py`.
(However, we abandoned our use of this algorithm due to poor preliminary results).
Correlation and Heat Map
“`
cd correlation
python3 correlation.py
“`
will generate heat map and do correlation calculation
