Add mean/std calculation
Browse files
README.md
CHANGED
@@ -25,5 +25,24 @@ model-index:
|
|
25 |
This is a trained model of a **Reinforce** agent playing **Pixelcopter-PLE-v0** .
|
26 |
To learn to use this model and train yours check Unit 4 of the Deep Reinforcement Learning Course: https://huggingface.co/deep-rl-course/unit4/introduction
|
27 |
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
|
|
25 |
This is a trained model of a **Reinforce** agent playing **Pixelcopter-PLE-v0** .
|
26 |
To learn to use this model and train yours check Unit 4 of the Deep Reinforcement Learning Course: https://huggingface.co/deep-rl-course/unit4/introduction
|
27 |
|
28 |
+
Some math about 'Pixelcopter' training.
|
29 |
+
The game is to fly in a passage and avoid blocks. Let we have trained our agent so that the probability to crash at block is _p_ (low enogh, I hope).
|
30 |
+
The probability that the copter crashes exactly at _n_-th block is product of probabilities it doesn't crash at previous _(n-1)_ blocks and probability it crashes at current block:
|
31 |
+
$$P = {p \cdot (1-p)^{n-1}}$$
|
32 |
+
The mathematical expectation of number of the block it crashes at is:
|
33 |
+
$$<n> = \sum_{n=1}^\infty{n \cdot p \cdot (1-p)^{n-1}} = \frac{1}{p}$$
|
34 |
+
The std is:
|
35 |
+
$$std(n) = \sqrt{<n^2>-<n>^2}$$
|
36 |
+
$$<n^2> = \sum_{n=1}^\infty{n^2 \cdot p \cdot (1-p)^{n-1}} = \frac{2-p}{p^2}$$
|
37 |
+
$$std(n) = \sqrt{\frac{2-p}{p^2}-\left( \frac{1}{p} \right) ^2} = \frac{\sqrt{1-p}}{p}$$
|
38 |
+
So difference is:
|
39 |
+
$$<n> - std(n) = \frac{1 - \sqrt{1-p}}{p}$$
|
40 |
+
As long as
|
41 |
+
$$ 0 \le p \le 1 $$
|
42 |
+
the following is true:
|
43 |
+
$$\sqrt{1-p} \ge 1-p$$
|
44 |
+
$$<n> - std(n) = \frac{1 - \sqrt{1-p}}{p} \le \frac{1 - (1-p)}{p} = 1$$
|
45 |
+
The scores _s_ in 'Pixelcopter' are the number of blocks passed decreased by 5 (for crash). So the average is lower by 5 and the std is the same. No matter how small _p_ is, our 'least score' is:
|
46 |
+
$$ (<n> - 5) - std(n) = <n> - std(n) - 5 \le - 4$$
|
47 |
+
But as we use only 10 episodes to calculate statistics and episode duration is limited, we can still achieve the goal, better agent, more chances. But understanding this is disappointing
|
48 |
|