问题
Just for learning I wanted to test this code. But there is a problem in it. I do not understand the problem. It says: ValueError: Shapes (20, 1) and (20, 2) are incompatiblefrom the line loss = network.train_on_batch(states, discounted_rewards)
Maybe there is something new in Tensorflow that was not there, the it was implemented.
The code from the website: https://adventuresinmachinelearning.com/policy-gradient-tensorflow-2/
import gym
import tensorflow as tf
from tensorflow import keras
import numpy as np
import datetime as dt
#STORE_PATH = '/Users/andrewthomas/Adventures in ML/TensorFlowBook/TensorBoard/PolicyGradientCartPole'
GAMMA = 0.95
env = gym.make("CartPole-v0")
state_size = 4
num_actions = env.action_space.n
tf.keras.backend.set_floatx('float64')
network = keras.Sequential([
keras.layers.Dense(30, activation='relu', kernel_initializer=keras.initializers.he_normal()),
keras.layers.Dense(30, activation='relu', kernel_initializer=keras.initializers.he_normal()),
keras.layers.Dense(num_actions, activation='softmax')
])
network.compile(loss='categorical_crossentropy',optimizer=keras.optimizers.Adam())
def get_action(network, state, num_actions):
print(state.reshape((1, -1)))
softmax_out = network(state.reshape((1, -1)))
print(softmax_out)
selected_action = np.random.choice(num_actions, p=softmax_out.numpy()[0])
return selected_action
def update_network(network, rewards, states, actions, num_actions):
reward_sum = 0
discounted_rewards = []
for reward in rewards[::-1]: # reverse buffer r
reward_sum = reward + GAMMA * reward_sum
discounted_rewards.append(reward_sum)
discounted_rewards.reverse()
discounted_rewards = np.array(discounted_rewards)
# standardise the rewards
discounted_rewards -= np.mean(discounted_rewards)
discounted_rewards /= np.std(discounted_rewards)
states = np.vstack(states)
print("States", states.shape)
print(states)
print("Rewards", discounted_rewards.shape)
print(discounted_rewards)
loss = network.train_on_batch(states, discounted_rewards)
return loss
num_episodes = 10000000
#train_writer = tf.summary.create_file_writer(STORE_PATH + f"/PGCartPole_{dt.datetime.now().strftime('%d%m%Y%H%M')}")
for episode in range(num_episodes):
state = env.reset()
rewards = []
states = []
actions = []
while True:
action = get_action(network, state, num_actions)
new_state, reward, done, _ = env.step(action)
states.append(state)
rewards.append(reward)
actions.append(action)
if done:
loss = update_network(network, rewards, states, actions, num_actions)
tot_reward = sum(rewards)
print(f"Episode: {episode}, Reward: {tot_reward}, avg loss: {loss:.5f}")
with train_writer.as_default():
tf.summary.scalar('reward', tot_reward, step=episode)
tf.summary.scalar('avg loss', loss, step=episode)
break
state = new_state
来源：https://stackoverflow.com/questions/61932773/tensorflow-2-valueerror-shapes-20-1-and-20-2-are-incompatible-in-gym-envi