its working

[vpetrea71]

i make few modification en is ok @ import tensorflow as tf from tf_agents.agents.dqn import dqn_agent from tf_agents.environments import wrappers from tf_agents.environments import tf_py_environment from environment import LiveBinanceEnvironment import pandas as pd from tf_agents.networks import q_network import pickle import os from dotenv import load_dotenv from binance.client import Client import sched, time from cryptography.hazmat.primitives.serialization import load_pem_private_key import base64 import requests import pytz

load_dotenv()

AGENT_MODEL_PATH = "policy_1min_1000"

Check for GPU

physical_devices = tf.config.experimental.list_physical_devices('GPU') if physical_devices: tf.config.experimental.set_memory_growth(physical_devices[0], True)

Initialize Binance client

api_key = os.getenv("CLIENT_KEY") api_secret = os.getenv("SECRET_KEY") client = Client(api_key, api_secret)

Initialize Live Environment

live_env = LiveBinanceEnvironment("BTC", "USDT", .002, 15, 15, 10, 12, 26) live_env = tf_py_environment.TFPyEnvironment(live_env)

Initialize Q Network

fc_layer_params = (100, 23, 33) q_net = q_network.QNetwork( live_env.observation_spec(), live_env.action_spec(), fc_layer_params=fc_layer_params )

Initialize DQN Agent

optimizer = tf.compat.v1.train.AdamOptimizer(learning_rate=1e-4) train_step_counter = tf.compat.v2.Variable(0) tf_agent = dqn_agent.DqnAgent( live_env.time_step_spec(), live_env.action_spec(), q_network=q_net, optimizer=optimizer, train_step_counter=train_step_counter )

Load the agent policy

try: policy = tf_agent.policy policy_state = policy.get_initial_state(batch_size=3) print("Policy loaded from: {}".format(AGENT_MODEL_PATH)) except: raise Exception("Model needed")

Schedule the agent to run at every T time interval

The agent will make a step depending on the current state

s = sched.scheduler(time.time, time.sleep)

time_step = live_env.current_time_step()

def run_step(step, sc): action_step = policy.action(step) next_time_step = live_env.step(action_step.action) time_step = next_time_step print(action_step.action) print(time_step.observation) print(time_step.reward)

s.enter(60, 1, run_step, (step, sc,))

s.enter(60, 1, run_step, (time_step, s, )) s.run()