Added better rng, its model file, and GPS noise simulation #2492

kanemasa1987 · 2016-08-03T15:19:52Z

"Updated"
Short Description:

Added somewhat better rng to replicate a human player.
Added Config class so that the behavior is customizable by user.
#1980, merging branch was wrong.

New Features

GPS noise replication
Ball throw fail, wait, precisions can be configured now

Fixes:

Added better rngs (pokemongo_bot/human_behavior.py)
Added singleton Config class and get_config utility (pokemongo_bot/config.py)
Changed api.set_position to replicate gps errors (pokemongo_bot/api_wrapper.py, init.py).
api.get_position and bot.position returns the actual position the bot is meant to be.

…t...

douglascamata · 2016-08-03T18:06:13Z

@kanemasa1987 we need to be able to support not using these randomisations at all.... for example I like to use only sleep for 1 or 2 seconds max whenever a sleep is needed and force excellent throw all the time.

… rng everytime we call them...

… aswell...

…ingies

kanemasa1987 · 2016-08-07T12:06:55Z

It seems to be working fine as far as I tested.
Please review the code.

Nihisil · 2016-08-11T06:20:20Z

pokemongo_bot/human_behaviour.py

+    return 1.0
+
+"""
+def _visualize():


Why there is a big piece of commented code?

_visualize was used to draw graphs to check the distribution of rng (which I showed in #1980,) to verify that there are no anomalies at the bounds etc. Suppose you use a gaussian of mu=0.5, with big std. If you clip the min and max value between [0, 1], then you will see pillars at 0 and 1, which is not nice as a rng to replicate human behaviour.
It should be useful for future rng development, however, the CI failed due to "import seaborn" line in the function (which is never used or called in the actual app, but detected by CI,) so I decided to comment them out rather than erasing.

The other function (_precalc_lognormal_ping_param) was used to calculate parameter for the log-normal. It uses theano, and since it is somewhat heavy task, I precalculated some of them and commented out for same reason, just incase we want to calculate other parameter for log-normal that generates our desired random value.

peter-bonanni · 2016-08-11T15:54:06Z

I've been testing this locally merged with the latest dev @ 138b664. Everything works with a simple manual merge of the new config options.

I'm seeing a lot of drift and reflex with both replicate_reflex and replicate_gps_noise set to true. It makes it very difficult for the path worker to follow a path when they're rubber-banding. I also noticed that with the default values and the above two flags set to true I received a soft-ban pretty regularly. I was able to fix this using the following config options:

"replicator": { "max_sleep" : 60, "replicate_reticle_fail_delay": true, "reticle_fail_chance" : 0.05, "reticle_fail_max_trial" : 10, "reticle_fail_max_sleep" : 10, "jitter_ping" : 80, "jitter_range" : 15, "jitter_max_sleep" : 1.0, "replicate_reflex" : false, "reflex_time" : 270, "reflex_range" : 100, "reflex_max_sleep" : 1.0, "replicate_gps_noise" : false, "ball_throw_hit_success_chance": 0.80, "gps_noise_radius" : 0.00032, "replicate_ball_throw_reticle" : "human", "replicate_ball_throw_spin" : "human" },

peter-bonanni · 2016-08-11T17:02:49Z

The changes related to throwing pokeballs seem to working well. I'm missing a decent amount of throws and my stats are starting to look more 'normal' however I'd really like it if there was a VIP flag that would let me do a perfect toss for pokemon in that group.

I just missed a snorlax with 3 ultraballs and then he fled.

douglascamata · 2016-08-11T20:03:37Z

Just saying that if you think this will protect you from bans, it won't. There are much easier ways for Niantic to detect and ban all bots user base (not just ours).

peter-bonanni · 2016-08-11T21:48:36Z

@douglascamata yeah the stuff getting sent in the signature is pretty telling.

kanemasa1987 · 2016-08-11T23:06:43Z

@douglascamata
That's true. What I'm trying here is at least not get banned statistically from the user's action log.
We don't know from what features they ban users in the future, but at least people with 100% hit chance are easier to ban than those does not.

@peter-bonanni
Thankyou for the test! The human reflex is not currently used, so it might be the gps that triggered softban. The value was set to 0.00075 according to the gps error of smartphone (somewhere I read,) but it might have been too high. I'm currently setting it at 0.00025 with PolylineWalker, seemingly not having softban issue (we need to change codes use PolylineWalker atm..., I may work on this later...)
For the vips, there can be a change in the future.

kanemasa1987 · 2016-08-13T03:21:41Z

At this point, the issue is whether to centralize the human_behavior config (this PR,) or split each config to each task's config (like current dev catch_pokemon.) ...
Any ideas?

douglascamata · 2016-08-13T16:22:08Z

@kanemasa1987 I think split in each task's config is better, imho.

kanemasa1987 · 2016-08-14T00:22:35Z

I will close this PR and resend GPS part later.
The human_behavior part has overextended.

M.K added 4 commits August 4, 2016 00:08

added better rng

9945ef3

forgot adding

b467da3

added better rng and model file

c05fbcf

CI failing due to the imports used in unused code, commenting them ou…

c0566ed

…t...

M.K added 14 commits August 4, 2016 07:51

Merge remote-tracking branch 'fork_master/dev' into random

e449178

Added Singleton Config class, so that we don't have to pass config to…

d692e2c

… rng everytime we call them...

quick fix on wrong syntax etc, waiting api update before further test...

53091d7

Worked around with nested config

35a815e

Added to conf examples and fixed config bug, added altitude gps noise…

1eaede7

… aswell...

fixed conflict, and bypassed bot.position()

7f53a74

fixed conflict correctly

c840e2b

CI failing due to wrong import path, fixed

4cfa87f

Merge remote-tracking branch 'fork_master/dev' into random

ab577cb

Merge remote-tracking branch 'fork_master/dev' into random

da1326d

fixed broken json and remaining conf issue, seems to be working now...

95e8669

Set default to not using random features, and added ballthrow fail th…

9b1de99

…ingies

minor fix

7dec8e9

merged with newest dev...

0d32662

fixed conflict

fe54a09

kanemasa1987 changed the title ~~added better rng and model file~~ Added better rng, its model file, and GPS noise simulation Aug 9, 2016

fixed conflict

3176b06

Nihisil reviewed Aug 11, 2016
View reviewed changes

fixed conflicts

a6c3e11

kanemasa1987 mentioned this pull request Aug 14, 2016

Gps noise #3912

Merged

kanemasa1987 closed this Aug 14, 2016

kanemasa1987 deleted the random branch August 20, 2016 11:46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Added better rng, its model file, and GPS noise simulation #2492

Added better rng, its model file, and GPS noise simulation #2492

kanemasa1987 commented Aug 3, 2016 •

edited

Loading

douglascamata commented Aug 3, 2016 •

edited

Loading

kanemasa1987 commented Aug 7, 2016

Nihisil Aug 11, 2016

kanemasa1987 Aug 11, 2016 •

edited

Loading

peter-bonanni commented Aug 11, 2016

peter-bonanni commented Aug 11, 2016

douglascamata commented Aug 11, 2016

peter-bonanni commented Aug 11, 2016

kanemasa1987 commented Aug 11, 2016 •

edited

Loading

kanemasa1987 commented Aug 13, 2016 •

edited

Loading

douglascamata commented Aug 13, 2016

kanemasa1987 commented Aug 14, 2016 •

edited

Loading

Added better rng, its model file, and GPS noise simulation #2492

Added better rng, its model file, and GPS noise simulation #2492

Conversation

kanemasa1987 commented Aug 3, 2016 • edited Loading

douglascamata commented Aug 3, 2016 • edited Loading

kanemasa1987 commented Aug 7, 2016

Nihisil Aug 11, 2016

Choose a reason for hiding this comment

kanemasa1987 Aug 11, 2016 • edited Loading

Choose a reason for hiding this comment

peter-bonanni commented Aug 11, 2016

peter-bonanni commented Aug 11, 2016

douglascamata commented Aug 11, 2016

peter-bonanni commented Aug 11, 2016

kanemasa1987 commented Aug 11, 2016 • edited Loading

kanemasa1987 commented Aug 13, 2016 • edited Loading

douglascamata commented Aug 13, 2016

kanemasa1987 commented Aug 14, 2016 • edited Loading

kanemasa1987 commented Aug 3, 2016 •

edited

Loading

douglascamata commented Aug 3, 2016 •

edited

Loading

kanemasa1987 Aug 11, 2016 •

edited

Loading

kanemasa1987 commented Aug 11, 2016 •

edited

Loading

kanemasa1987 commented Aug 13, 2016 •

edited

Loading

kanemasa1987 commented Aug 14, 2016 •

edited

Loading