RL4AD - Reinforcement Learning for Anomaly Detection

This needs to be used in combination with gym-network_intrusion library

Installation

Step 1:

Install all the required python packages described in requirements.txt

Step 2:

Install gym from OpenAI using the following steps

1. git clone https://github.com/openai/gym.git
2. cd gym
3. pip install -e .

Step 3:

Install gym-network_intrusion using the following steps

1. git clone https://github.com/harik68/gym-network_intrusion.git
2. cd gym-network_intrusion
3. pip install -e .

Step 4:

Install RL4AD using the following steps

1. git clone https://github.com/harik68/RL4AD.git
2. Copy the following file into the directory datasets https://www.dropbox.com/s/e2n5ow6b117ub80/kdd_nsl_train_onehot_string.pkl?dl=0
3. In your code create an instance of gym_network_intrusion environment using the following commands
4. Run the code NSL_KDD_DNN_QLearning.py

There are 3 directories

1. datasets - where you need to keep all the inputdata
2. results - where the program will output the results
3. temp - where the program will store intermediate results such as configuration of DNN after training

Running Configuration

I have set default values of running configuration as follows n_iterations = 10 # number of training iterations n_max_steps = 100 # max steps per episode This is for testing the code. It should complete in about 10 minutes. However this will not produce good results

You need to edit the code and use the following configuration to get better results once you finish the testing. n_iterations = 250 # number of training iterations n_max_steps = 1000 # max steps per episode

Remember to clear tmp folder after each run if you want the DNN to learn from scratch