Fatal Python error: (pygame prachute) Segmentation Fault

[KornbergFresnel]

I've tried to run your given example in README, then the script thrown an exception at g.update():

pygame 1.9.4
Hello from the pygame community. https://www.pygame.org/contribute.html
[FPS=0, UPS=0]
Fatal Python error: (pygame parachute) Segmentation Fault

Current thread 0x00007feadc869740 (most recent call first):
  File "test.py", line 25 in <module>
[1]    23004 abort (core dumped)  python3 test.py

System information

ubuntu 16.0.4 python 3.6 pygame 1.9.4 DeepRTS latest version

[dhineshkumar-r]

Were you able to resolve this issue?

[perara]

Hi, could you try the lastest version? I am not experiencing this error

[zdx3578]

@perara https://arxiv.org/pdf/1907.11971.pdf this paper rl code will open? or have other deep-rts rl code?

[perara]

@perara https://arxiv.org/pdf/1907.11971.pdf this paper rl code will open? or have other deep-rts rl code?

I'm not sure if I understand the question.

[perara]

@zdx3578 Perhaps you are looking for https://github.com/cair/deep-warehouse ?

[zdx3578]

I am Very interested Tsetlin machine ,I Want to ask are there any RL code integrate with Tsetlin machine code? like Gym env ; or integrate with deep-rts;

[perara]

Hi,

Not that I am aware of TM put into an RL setting. But I'll tag @olegranmo which might be able to guide you further.

[zdx3578]

@perara thanks very much!

[olegranmo]

Hi @zdx3578! Thanks for your interest in the TM.

We are currently working on RL based on the TM, however, no integration with deep-rts yet. This would be very interesting, though. One of our main activities right now is building an AI for the board game Hex, based on Monte Carlo Tree Search using the TM to evaluate board positions. Will publish code here as we proceed: https://github.com/cair/hex-ai Integrating with Gym env is definitively on the TODO list.

[zdx3578]

@olegranmo thanks very much! let me see this;

[zdx3578]

@olegranmo now i am see https://github.com/cair/pyTsetlinMachine/blob/master/pyTsetlinMachine/Tools.c file;but in paper no number_of_ta_chunks number_of_clause_chunks; now id dont know the mnist data is how to into the encodex;

[olegranmo]

The chunks are there because both the TA states and input are represented as bits, so there are e.g. 32 features in one 32-bit int. I guess it is best to use the Python-wrapper, with MNISTDemo.py as an example. Note that I have just released 0.4.0, making classification 6 times faster and learning 3 times faster for MNIST.

[zdx3578]

@olegranmo thanks very much !

[zdx3578]

@olegranmo now conv operate is one bit step ？ may be conv stride can increase？so number_of_patches
can rapid descend；so less compute. i test Reduced accuracy so you say future work Optimize convolution code ?

[olegranmo]

Good point, one-bit steps, so increasing stride would absolutely reduce cost. Have put it on the roadmap. Thanks!

[perara]

Hi, I'm closing this. If there is any future questions regard DeepRTS, feel free to open a new issue :) Best Regards Per-Arne