This repository provides code for the DQNimb model proposed in paper DRL for imbalanced classification:
-
train_model: Training the agent with DQN algorithm.
-
ICMDP_Env: The simulated environment for imabalanced classification.
-
get_model: Define the network structure for image or text.
-
data_pre: Loading the balanced datasets and constructing the imbalanced datasets according to the imbalanced rate.
- tensorflow
- keras
- keras-rl
- gym
python train_model.py --model image --data famnist --imb-rate 0.04 --min-class 456 --maj-class 789 --training-steps 120000
python train_model.py --model text --data imdb --imb-rate 0.1 --min-class 0 --maj-class 1 --training-steps 150000