pythonlessons

def getwindowgeometry():
    while True:
        output = subprocess.getstatusoutput(f'xdotool search --name Counter-Strike getwindowgeometry')
        if output[0] == 0:
            t1 = time.time()
            LIST = output[1].split("\n")
            Window = LIST[0][7:]
            Position = LIST[1][12:-12]
            x, y = Position.split(",")
            x, y = int(x), int(y)

pythonlessons

while True:
    t1 = time.time()
    img = np.array(sct.grab({"top": y-30, "left": x, "width": w, "height": h, "mon": -1}))
    img = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)
    image, detection_list, bboxes = detect_enemy(yolo, np.copy(img), input_size=YOLO_INPUT_SIZE, CLASSES=TRAIN_CLASSES, rectangle_colors=(255,0,0))
    cv2.circle(image,(int(w/2),int(h/2)), 3, (255,255,255), -1) # center of weapon sight

    th_list, t_list = [], []
    for detection in detection_list:
        diff_x = (int(w/2) - int(detection[1]))*-1

pythonlessons

import gym
import random

env = gym.make("LunarLander-v2")

def Random_games():
    # Each of this episode is its own game.
    for episode in range(10):
        env.reset()
        # this is each frame, up to 500...but we wont make it that far with random.

pythonlessons

class Actor_Model:
    def __init__(self, input_shape, action_space, lr, optimizer):
        X_input = Input(input_shape)
        self.action_space = action_space

        X = Dense(512, activation="relu", kernel_initializer=tf.random_normal_initializer(stddev=0.01))(X_input)
        X = Dense(256, activation="relu", kernel_initializer=tf.random_normal_initializer(stddev=0.01))(X)
        X = Dense(64, activation="relu", kernel_initializer=tf.random_normal_initializer(stddev=0.01))(X)
        output = Dense(self.action_space, activation="softmax")(X)

pythonlessons

class Critic_Model:
    def __init__(self, input_shape, action_space, lr, optimizer):
        X_input = Input(input_shape)
        old_values = Input(shape=(1,))

        V = Dense(512, activation="relu", kernel_initializer='he_uniform')(X_input)
        V = Dense(256, activation="relu", kernel_initializer='he_uniform')(V)
        V = Dense(64, activation="relu", kernel_initializer='he_uniform')(V)
        value = Dense(1, activation=None)(V)

pythonlessons

def get_gaes(self, rewards, dones, values, next_values, gamma = 0.99, lamda = 0.9, normalize=True):
    deltas = [r + gamma * (1 - d) * nv - v for r, d, nv, v in zip(rewards, dones, next_values, values)]
    deltas = np.stack(deltas)
    gaes = copy.deepcopy(deltas)
    for t in reversed(range(len(deltas) - 1)):
        gaes[t] = gaes[t] + (1 - dones[t]) * gamma * lamda * gaes[t + 1]

    target = gaes + values
    if normalize:
        gaes = (gaes - gaes.mean()) / (gaes.std() + 1e-8)

pythonlessons

def replay(self, states, actions, rewards, predictions, dones, next_states):
    # reshape memory to appropriate shape for training
    states = np.vstack(states)
    next_states = np.vstack(next_states)
    actions = np.vstack(actions)
    predictions = np.vstack(predictions)

    # Get Critic network predictions 
    values = self.Critic.predict(states)
    next_values = self.Critic.predict(next_states)

pythonlessons

def run_batch(self): # train every self.Training_batch episodes
    state = self.env.reset()
    state = np.reshape(state, [1, self.state_size[0]])
    done, score, SAVING = False, 0, ''
    while True:
        # Instantiate or reset games memory
        states, next_states, actions, rewards, predictions, dones = [], [], [], [], [], []
        for t in range(self.Training_batch):
            self.env.render()
            # Actor picks an action

pythonlessons

import gym
import random
import numpy as np

env = gym.make("BipedalWalker-v3")

def Random_games():
    # Each of this episode is its own game.
    action_size = env.action_space.shape[0]
    for episode in range(10):

pythonlessons

class Actor_Model:
    def __init__(self, input_shape, action_space, lr, optimizer):
        X_input = Input(input_shape)
        self.action_space = action_space
        
        X = Dense(512, activation="relu", kernel_initializer=tf.random_normal_initializer(stddev=0.01))(X_input)
        X = Dense(256, activation="relu", kernel_initializer=tf.random_normal_initializer(stddev=0.01))(X)
        X = Dense(64, activation="relu", kernel_initializer=tf.random_normal_initializer(stddev=0.01))(X)
        output = Dense(self.action_space, activation="tanh")(X)