🤖 Crypto_RL 🤖

ALGO:

• vpg - Vanilla Policy Gradient
• a2c - Advantage Actor Critic
• ppo - Proximal Policy Optimization
• dqn - Deep Q-Netowork
• ddqn - Dueling Deep Q-Netowork

FEATURE_EXTRACTOR:

• mlp
• lstm
• transformer

REWARD_TYPE:

• profit
• sr (!not yet added)

DEVICE:

• cuda

HYPERPARAMS:

• lr - actor and critic lr
• gamma - discount factor
• epochs

Before running code :

conda activate rl
mkdir CHECKPOINT
mkdir dataset

Training :

python project_code/drl_algo/train.py --algo ALGO --feature_extractor FEATURE_EXTRACTOR --reward_type REWARD_TYPE --lr LR --gamma GAMMA --device DEVICE --epochs EPOCH

Testing :

python project_code/drl_algo/test.py --algo ALGO --feature_extractor FEATURE_EXTRACTOR --reward_type REWARD_TYPE --device DEVICE

Ploting Action Plots :

#if buy_Sell
python project_code/utils/plot_buy_sell.py --algo ALGO --feature_extractor FEATURE_EXTRACTOR --reward_type REWARD_TYPE
#if hold_Sell
python project_code/utils/plot_hold_sell.py --algo ALGO --feature_extractor FEATURE_EXTRACTOR --reward_type REWARD_TYPE
#if hold_buy_sell
python project_code/utils/plot_hold_buy_sell.py --algo ALGO --feature_extractor FEATURE_EXTRACTOR --reward_type REWARD_TYPE

Ploting Close Values :

MODE: train/test PROJECT_NAME: wandb project name used for testing

python project_code/utils/plot_states.py --mode MODE --project_name PROJECT_NAME