# Questions tagged [policy-gradients]

For questions related to reinforcement learning algorithms often referred to as "policy gradients" (or "policy gradient algorithms"), which attempt to directly optimise a parameterised policy (without first attempting to estimate value functions) using gradients of an objective function with respect to the policy's parameters.

50 questions
Filter by
Sorted by
Tagged with
1answer
35 views

1answer
2k views

### Do off-policy policy gradient methods exist?

Do off-policy policy gradient methods exist? I know that policy gradient methods themselves using the policy function for sampling rollouts. But can't we easily have a model for sampling from the ...
1answer
109 views

### How is gradient being calculated in Andrej Karpathy's pong code?

I was going through the code by Andrej Karpathy on reinforcement learning using a policy gradient. I have some questions from the code. Where is the logarithm of the probability being calculated? ...
1answer
46 views

### Purpose of using actor-critic algorithms under deterministic MDP dynamics?

One of the main disadvantages of the MC Policy Gradient algorithm (REINFORCE) as described say here is the fact that it has high variance (returns, which we sample, will significantly vary from ...
1answer
291 views

### Reinforcement learning with PPO: rewards decreasing

I'm trying to train a PPO agent and my average rewards graph looks like this. Could this indicate that it's stuck at a local maximum? Do I need to promote exploring by increasing the entropy or does ...
0answers
25 views

### What is the complexity of policy gradient algorithms compared to discrete action space algorithms?

I am using a policy gradient algorithm (actor-critic) for wireless networks. The policy gradient-based algorithm helps because it considers continuous action space. But how much does a policy ...
1answer
70 views

### Solving a Multi-Armed, “Multi-Bandit” Problem

This is the problem: I have 66 slot-machines and for each of them I have 7 possible actions/arms to choose from. At each trial, I have to choose one of 7 actions for each and every one of the 66 slots....
1answer
35 views

### Why is image classification tasks are dominated by minimizing cost function instead of maximizing ones?

I was watching a video of policy gradient by Andrej Karpathy at 10:00 there shows an equation for supervised learning for image classification. $max\sum _{i}log \:p(y_i|x_i)$ I have worked with ...
1answer
64 views

### Model-based Reinforcement Learning algorithm for real-time robotics task

I'm quite a newbie when it comes to practically working with Deep Learning techniques, although I studied them quite a lot theoretically in the last months. However, now I'm facing my first practical ...
1answer
50 views

### How do policy gradients compute an infinite probability distribution from a neural network

Do neural networks compute the probability distribution for policy gradient methods. If so, how do they compute an infinite probability distribution? How do you represent a continuous action policy ...
1answer
87 views

### How can the derivative of a neural network be calculated, given no mathematical expression?

Neural networks (NNs) are used as approximators in reinforcement learning (RL). To update the policy in RL, the actor network's gradients w.r.t its weights are needed. Since NN doesn't have a ...
0answers
83 views

### PPO: action std or entropy for exploration?

When trying to implement my own PPO (Proximal Policy Optimizer), I came accross two different implementations : Exploration with action std : Collect trajectories on ...
3answers
2k views

### Why does is make sense to normalize rewards per episode in reinforcement learning?

In Open AI's actor-critic and in Open AI's REINFORCE, the rewards are being normalized like so rewards = (rewards - rewards.mean()) / (rewards.std() + eps) on ...
0answers
131 views

### Policy gradient methods for continuous action space

I have a problem I would like to tackle with RL but I am not sure if it is even doable. My agent has to figure out how to fill a very large vector (let's say from 600 to 4000 in the most complex ...
0answers
65 views

### Understanding policy update in PPO2

I have a question regarding the functionality of the PPO2 algorithm together with the Stable Baselines implementation: From the original paper I know that the policy parameters $\theta$ are updated K-...
0answers
55 views

### Encoding real valued inputs

UPDATE: After reading more about the topic, I've tried implementing the DDPG algorithm instead of using a variation of Q-Learning and still have the same issue. I have the following issue: I want ...
2answers
860 views

### How large should the replay buffer be?

I'm learning DDPG algorithm by following the following link: Open AI Spinning Up document on DDPG, where it is written In order for the algorithm to have stable behavior, the replay buffer should ...
1answer
97 views

### Reinforcement Learning without state space

I want to use Reinforcement Learning to optimize the distribution of energy for a peak shaving problem given by a thermodynamical simulation. However, I am not sure how to proceed as the action space ...
2answers
182 views

### Is REINFORCE the same as 'vanilla policy gradient'?

I don't know what people mean by 'vanilla policy gradient', but what comes to mind is REINFORCE, which is the simplest policy gradient algorithm I can think of. Is this an accurate statement? By ...
0answers
36 views

### How does the TRPO surrogate loss account for the error in the policy?

In the Trust Region Policy Optimization (TRPO) paper, on page 10, it is stated An informal overview is as follows. Our proof relies on the notion of coupling, where we jointly define the ...
1answer
196 views

### How is the policy gradient calculated in REINFORCE?

Reading Sutton and Barto, I see the following in describing policy gradients: How is the gradient calculated with respect to an action (taken at time t)? I've read implementations of the algorithm, ...
0answers
126 views

### Policy gradient loss for neural network training

Say i want to train a neural network with 10 classes as outputs and use categorical_cross_entropy as a loss function in keras. This will try to fit the training ...
0answers
82 views

### Can I use deterministic policy gradient methods for stochastic policy learning?

Can I treat a stochastic policy (over a finite action space of size $n$) as a deterministic policy (in the set of probability distribution in $\mathbb{R}^n$)? It seems to me that nothing is broken ...
0answers
172 views

### Policy gradient in keras predicts only one action

I have trouble with the REINFORCE algorithm in keras with Atari games. After round about 30 episodes the network converges to one action. But the same algorithm is working with CartPole-v1 and ...
1answer
88 views

### Reinforcement Learning with more actions than states

I have read a lot about RL recently. As far as I understood, most RL applications have much more states than there are actions to choose from. I am thinking about using RL for a problem where I have ...
1answer
5k views

### What is the relation between Q-learning and policy gradients methods?

As far as I understand, Q-learning and policy gradients (PG) are the two major approaches used to solve RL problems. While Q-learning aims to predict the reward of a certain action taken in a certain ...
1answer
81 views

### Calculating gradient for log policy when variance is not constant

I've noticed that when modelling a continuous action space, the default thing to do is to estimate a mean and a variance where each is parameterized by a neural network or some other model. I also ...
1answer
270 views

### Meaning of Actor Output in Actor Critic Reinforcement Learning

In actor critic, The equations for calculating the loss in actor critic are an actor loss (parameterized by $\theta$) $$log[\pi_\theta(s_t,a_t)]Q_w(s_t,a_t)$$ and a critic loss (parameterized by ...
1answer
182 views

### What information should be cached in experience replay for actor-critic?

Experience replay is buffer (or a "memory") of transactions $e_t = (s_t, a_t, r_t, s_{t+1})$. The equations for calculating the loss in actor critic are an actor loss (parameterized by $\theta$) ...
1answer
31 views

### Neural network with logical hidden layer - how to train it? Is it policy gradient problem? Chaining NNs?

I am doing neural machine translation task from language S to language T via interlingua L. So - there is the structure: ...
0answers
30 views

### How to include exploration in Gaussian policy

When dealing with continuous action spaces, a common choice when designing a policy in policy gradient methods is to learn mean and variance of actions for a specific state and then simply sample from ...
0answers
29 views

### Impact of Varying Length Trajectories on Policy Gradient Optimization

As the question states, I am wondering how, if at all, a varying length of a trajectory (series of state,action pairs) will impact training/performance of policy gradient algorithms such as PPO, TRPO ...
1answer
244 views

### Can gradient descent training be used for nonsmooth loss functions?

I have non-smooth loss function - e.g. loss(x)=min(x, 0.5). Can gradient descent be used for training neural networks with such functions. Can gradient descent be used for fairly general, ...