File size: 1,439 Bytes
5def754
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cb3a260
 
5def754
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
<!DOCTYPE html>
<html>

<head>
    <meta charset="utf-8">
    <title>Frozen Lake</title>
    <meta name="viewport" content="width=device-width, initial-scale=1">
    <style>
        body {
            background-color: #000;
        }

        #container {
            margin: auto;
            max-width: 800px;
            text-align: center;
        }

        #container>img {
            width: 100%
        }

        #container>a,
        #container>h2,
        #container>p {
            color: #fff;
        }

        #container>a {
            margin-top: 16px;
        }
    </style>
</head>

<body>
    <div id="container">
        <h2>RL - Slippery Frozen Lake Q-Learning</h2>
        <p>I trained a Q-Learning model on the OpenAI Gym Slippery Frozen Lake environment for 20,000 iterations, and
            evaluated for 1,000 iterations.
            The trained model had a success rate of about 73%. Action for the 3 column on the second row is especially
            interesting,
            it correctly learns that the best policy is to try move toward one of the holes because there is a 1/3
            chance of slipping and slips are orthogonal to the desired direction.</p>
        <img src="assets/eval_screenshot_train_20k.png" alt="Q-Learning Agent">
        <a href="https://www.youtube.com/watch?v=b1oh3TK6Jhg" target="_blank">Training and Evaluation Video
            (youtube)</a>
    </div>
</body>

</html>