Skip to content

Commit

Permalink
final edit, finishing touches. Python perf test: SPS: 2,045,564.37560…
Browse files Browse the repository at this point in the history 
…13566 on mediocre hardware. C perf test: SPS: 333299.687500. added README. formatted w/ clang-tidy Google for succinctness and readability.
  • Loading branch information
xinpw8 committed Jan 12, 2025
1 parent 791bfe5 commit fcba83c
Show file tree
Hide file tree
Showing 6 changed files with 1,555 additions and 1,550 deletions.
44 changes: 44 additions & 0 deletions pufferlib/ocean/enduro/README.md
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
# README
Please star PufferLib on GitHub: it really makes a difference!
https://github.com/pufferai/pufferlib

by Daniel Addis, 2024
https://github.com/xinpw8

## Puffer Ocean Enduro
This project contains a performant reinforcement-learning environment inspired by the classic Atari 2600 game Enduro. It uses C, Cython, and Python to provide an interactive RL environment, trainable with PufferLib.

## Building & Setup
1. Install dependencies. All commands should be run from the `PufferLib` top directory.
```sh
pip install -e .'[cleanrl]'
```
Instructions on https://puffer.ai/docs.html

2. Compilation is run when pufferlib is pip installed. After making changes to `cy_enduro.pyx`, `enduro.c`, or `enduro.h`, recompile with:
```sh
python setup.py build_ext --inplace
```

3. To locally compile the C environment for testing (without Cython), run:
```sh
scripts/build_ocean.sh enduro local
```
This builds a runnable `enduro` module in the PufferLib top directory, which you can run with:
```sh
./enduro
```
Hold Shift to take control from the agent.

## Training
To train using the demo script with wandb logs, run:
```sh
python demo.py --env puffer_enduro --mode train --track
```
Model files are saved at intervals specified in `config/ocean/enduro.ini` to the `experiments/` directory.

## Evaluation
To evaluate a local checkpoint, run:
```sh
python demo.py --env puffer_enduro --mode eval --eval-model-path your_model.pt
```
7 changes: 1 addition & 6 deletions pufferlib/ocean/enduro/cy_enduro.pyx
View file
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
# cython: language_level=3
cimport numpy as cnp
from libc.stdlib cimport malloc, calloc, free
from libc.stdlib cimport calloc, free
from libc.string cimport memset
from libc.time cimport time
from random import SystemRandom

rng = SystemRandom()
Expand Down Expand Up @@ -38,7 +37,6 @@ cdef extern from "enduro.h":
unsigned char* truncateds
LogBuffer* log_buffer
size_t obs_size
int num_envs

ctypedef struct GameState
GameState* make_client(Enduro* env)
Expand Down Expand Up @@ -69,8 +67,6 @@ cdef class CyEnduro:
self.envs = <Enduro*>calloc(num_envs, sizeof(Enduro))
self.logs = allocate_logbuffer(LOG_BUFFER_SIZE)

from time import time as py_time

for i in range(num_envs):
unique_seed = rng.randint(0, 2**32 - 1) & 0x7FFFFFFF
memset(&self.envs[i], 0, sizeof(Enduro))
Expand Down Expand Up @@ -114,4 +110,3 @@ cdef class CyEnduro:
def log(self):
cdef Log log = aggregate_and_clear(self.logs)
return log

145 changes: 70 additions & 75 deletions pufferlib/ocean/enduro/enduro.c
View file
Original file line number Diff line number Diff line change
@@ -1,97 +1,92 @@
#include "enduro.h"

#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <stddef.h>
#include <time.h>
#include "enduro.h"
#include "raylib.h"
#include "puffernet.h"

#define MAX_ENEMIES 10
#include "puffernet.h"
#include "raylib.h"

void get_input(Enduro* env) {
if ((IsKeyDown(KEY_DOWN) && IsKeyDown(KEY_RIGHT)) || (IsKeyDown(KEY_S) && IsKeyDown(KEY_D))) {
env->actions[0] = ACTION_DOWNRIGHT; // Decelerate and move right
} else if ((IsKeyDown(KEY_DOWN) && IsKeyDown(KEY_LEFT)) || (IsKeyDown(KEY_S) && IsKeyDown(KEY_A))) {
env->actions[0] = ACTION_DOWNLEFT; // Decelerate and move left
} else if (IsKeyDown(KEY_SPACE) && (IsKeyDown(KEY_RIGHT) || IsKeyDown(KEY_D))) {
env->actions[0] = ACTION_RIGHTFIRE; // Accelerate and move right
} else if (IsKeyDown(KEY_SPACE) && (IsKeyDown(KEY_LEFT) || IsKeyDown(KEY_A))) {
env->actions[0] = ACTION_LEFTFIRE; // Accelerate and move left
} else if (IsKeyDown(KEY_SPACE)) {
env->actions[0] = ACTION_FIRE; // Accelerate
} else if (IsKeyDown(KEY_DOWN) || IsKeyDown(KEY_S)) {
env->actions[0] = ACTION_DOWN; // Decelerate
} else if (IsKeyDown(KEY_LEFT) || IsKeyDown(KEY_A)) {
env->actions[0] = ACTION_LEFT; // Move left
} else if (IsKeyDown(KEY_RIGHT) || IsKeyDown(KEY_D)) {
env->actions[0] = ACTION_RIGHT; // Move right
} else {
env->actions[0] = ACTION_NOOP; // No action
}
void get_input(Enduro *env) {
if ((IsKeyDown(KEY_DOWN) && IsKeyDown(KEY_RIGHT)) ||
(IsKeyDown(KEY_S) && IsKeyDown(KEY_D))) {
env->actions[0] = ACTION_DOWNRIGHT; // Decelerate and move right
} else if ((IsKeyDown(KEY_DOWN) && IsKeyDown(KEY_LEFT)) ||
(IsKeyDown(KEY_S) && IsKeyDown(KEY_A))) {
env->actions[0] = ACTION_DOWNLEFT; // Decelerate and move left
} else if (IsKeyDown(KEY_SPACE) &&
(IsKeyDown(KEY_RIGHT) || IsKeyDown(KEY_D))) {
env->actions[0] = ACTION_RIGHTFIRE; // Accelerate and move right
} else if (IsKeyDown(KEY_SPACE) &&
(IsKeyDown(KEY_LEFT) || IsKeyDown(KEY_A))) {
env->actions[0] = ACTION_LEFTFIRE; // Accelerate and move left
} else if (IsKeyDown(KEY_SPACE)) {
env->actions[0] = ACTION_FIRE; // Accelerate
} else if (IsKeyDown(KEY_DOWN) || IsKeyDown(KEY_S)) {
env->actions[0] = ACTION_DOWN; // Decelerate
} else if (IsKeyDown(KEY_LEFT) || IsKeyDown(KEY_A)) {
env->actions[0] = ACTION_LEFT; // Move left
} else if (IsKeyDown(KEY_RIGHT) || IsKeyDown(KEY_D)) {
env->actions[0] = ACTION_RIGHT; // Move right
} else {
env->actions[0] = ACTION_NOOP; // No action
}
}

int demo() {
Weights* weights = load_weights("resources/enduro/0105enduro_weights.bin", 142218);
LinearLSTM* net = make_linearlstm(weights, 1, 68, 9);

Enduro env = {
.num_envs = 1,
.max_enemies = MAX_ENEMIES,
.obs_size = OBSERVATIONS_MAX_SIZE
};

allocate(&env);
GameState* client = make_client(&env);
unsigned int seed = 0;
init(&env, seed, 0);
reset(&env);
Weights *weights =
load_weights("resources/enduro/enduro_weights.bin", 142218);
LinearLSTM *net = make_linearlstm(weights, 1, 68, 9);

while (!WindowShouldClose()) {
if (IsKeyDown(KEY_LEFT_SHIFT)) {
get_input(&env);
} else {
forward_linearlstm(net, env.observations, env.actions);
}
Enduro env = {.obs_size = OBSERVATIONS_MAX_SIZE};
allocate(&env);
GameState *client = make_client(&env);
unsigned int seed = 0;
init(&env, seed, 0);
reset(&env);

c_step(&env);
render(client, &env);
while (!WindowShouldClose()) {
if (IsKeyDown(KEY_LEFT_SHIFT)) {
get_input(&env);
} else {
forward_linearlstm(net, env.observations, env.actions);
}

free_linearlstm(net);
free(weights);
close_client(client);
free_allocated(&env);
return 0;
c_step(&env);
render(client, &env);
}

free_linearlstm(net);
free(weights);
close_client(client);
free_allocated(&env);
return 0;
}

void perftest(float test_time) {
Enduro env = {
.num_envs = 1,
.max_enemies = MAX_ENEMIES,
.obs_size = OBSERVATIONS_MAX_SIZE
};

allocate(&env);
Enduro env = {.obs_size = OBSERVATIONS_MAX_SIZE};
allocate(&env);

unsigned int seed = 12345;
init(&env, seed, 0);
reset(&env);
unsigned int seed = 12345;
init(&env, seed, 0);
reset(&env);

int start = time(NULL);
int i = 0;
while (time(NULL) - start < test_time) {
env.actions[0] = rand()%9;
c_step(&env);
i++;
}
int start = time(NULL);
int i = 0;
while (time(NULL) - start < test_time) {
env.actions[0] = rand() % 9;
c_step(&env);
i++;
}

int end = time(NULL);
printf("SPS: %f\n", i / (float)(end - start));
free_allocated(&env);
int end = time(NULL);
printf("SPS: %f\n", i / (float)(end - start));
free_allocated(&env);
}

int main() {
demo();
// perftest(20.0f);
return 0;
demo();
// perftest(20.0f);
return 0;
}
Loading

0 comments on commit fcba83c

Please sign in to comment.