Very dumb change. Prefix c functions with c_ to prevent Conda's garba…

…ge compiler from breaking

config/default.ini

@@ -1,6 +1,7 @@
 [base]
 package = None
 env_name = None
+vec = native
 policy_name = Policy
 rnn_name = None
 max_suggestion_cost = 3600

config/ocean/connect4.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
 env_name = puffer_connect4
+vec = multiprocessing
 policy_name = Policy
 rnn_name = Recurrent
 

config/ocean/go.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
 env_name = puffer_go
+vec = multiprocessing
 policy_name = Go
 rnn_name = Recurrent
 

config/ocean/grid.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
 env_name = puffer_grid
+vec = multiprocessing
 policy_name = Policy
 rnn_name = Recurrent
 

config/ocean/moba.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
 env_name = puffer_moba
+vec = multiprocessing
 policy_name = MOBA
 rnn_name = Recurrent
 

config/ocean/nmmo3.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
-env_name = nmmo3
+env_name = puffer_nmmo3
+vec = multiprocessing
 policy_name = NMMO3
 rnn_name = NMMO3LSTM
 

config/ocean/snake.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
 env_name = puffer_snake
+vec = multiprocessing
 rnn_name = Recurrent
 
 [env]

config/ocean/trash_pickup.ini

@@ -1,6 +1,7 @@
 [base]
 package = ocean
 env_name = trash_pickup puffer_trash_pickup 
+vec = multiprocessing
 policy_name = TrashPickup
 rnn_name = Recurrent
 

demo.py

@@ -199,13 +199,13 @@ def carbs_param(group, name, space, wandb_params, mmin=None, mmax=None,
         is_wandb_logging_enabled=False,
         resample_frequency=5,
         num_random_samples=len(param_spaces),
-        max_suggestion_cost=args['base']['max_suggestion_cost'],
+        max_suggestion_cost=args['max_suggestion_cost'],
         is_saved_on_every_observation=False,
     )
     carbs = CARBS(carbs_params, param_spaces)
 
     # GPUDrive doesn't let you reinit the vecenv, so we have to cache it
-    cache_vecenv = args['base']['env_name'] == 'gpudrive'
+    cache_vecenv = args['env_name'] == 'gpudrive'
 
     elos = {'model_random.pt': 1000}
     vecenv = {'vecenv': None} # can't reassign otherwise
@@ -293,7 +293,7 @@ def train(args, make_env, policy_cls, rnn_cls, wandb,
     elif args['vec'] == 'native':
         vec = pufferlib.environment.PufferEnv
     else:
-        raise ValueError(f'Invalid --vector (serial/multiprocessing/ray/native).')
+        raise ValueError(f'Invalid --vec (serial/multiprocessing/ray/native).')
 
     if vecenv is None:
         vecenv = pufferlib.vector.make(
@@ -360,8 +360,6 @@ def train(args, make_env, policy_cls, rnn_cls, wandb,
         default='puffer_squared', help='Name of specific environment to run')
     parser.add_argument('--mode', type=str, default='train',
         choices='train eval evaluate sweep sweep-carbs autotune profile'.split())
-    parser.add_argument('--vec', '--vector', '--vectorization', type=str,
-        default='native', choices=['serial', 'multiprocessing', 'ray', 'native'])
     parser.add_argument('--vec-overwork', action='store_true',
         help='Allow vectorization to use >1 worker/core. Not recommended.')
     parser.add_argument('--eval-model-path', type=str, default=None,
@@ -377,6 +375,7 @@ def train(args, make_env, policy_cls, rnn_cls, wandb,
     parser.add_argument('--wandb-group', type=str, default='debug')
     args = parser.parse_known_args()[0]
 
+
     file_paths = glob.glob('config/**/*.ini', recursive=True)
     for path in file_paths:
         p = configparser.ConfigParser()
@@ -394,7 +393,10 @@ def train(args, make_env, policy_cls, rnn_cls, wandb,
 
     for section in p.sections():
         for key in p[section]:
-            argparse_key = f'--{section}.{key}'.replace('_', '-')
+            if section == 'base':
+                argparse_key = f'--{key}'.replace('_', '-')
+            else:
+                argparse_key = f'--{section}.{key}'.replace('_', '-')
             parser.add_argument(argparse_key, default=p[section][key])
 
     # Late add help so you get a dynamic menu based on the env
@@ -416,7 +418,7 @@ def train(args, make_env, policy_cls, rnn_cls, wandb,
         except:
             prev[subkey] = value
 
-    package = args['base']['package']
+    package = args['package']
     module_name = f'pufferlib.environments.{package}'
     if package == 'ocean':
         module_name = 'pufferlib.ocean'
@@ -425,12 +427,12 @@ def train(args, make_env, policy_cls, rnn_cls, wandb,
     env_module = importlib.import_module(module_name)
 
     make_env = env_module.env_creator(env_name)
-    policy_cls = getattr(env_module.torch, args['base']['policy_name'])
+    policy_cls = getattr(env_module.torch, args['policy_name'])
 
-    rnn_name = args['base']['rnn_name']
+    rnn_name = args['rnn_name']
     rnn_cls = None
     if rnn_name is not None:
-        rnn_cls = getattr(env_module.torch, args['base']['rnn_name'])
+        rnn_cls = getattr(env_module.torch, args['rnn_name'])
 
     if args['baseline']:
         assert args['mode'] in ('train', 'eval', 'evaluate')

pufferlib/ocean/breakout/breakout.h

@@ -437,7 +437,7 @@ void reset_round(Breakout* env) {
     env->ball_vx = 0.0;
     env->ball_vy = 0.0;
 }
-void reset(Breakout* env) {
+void c_reset(Breakout* env) {
     env->log = (Log){0};
     env->score = 0;
     env->num_balls = 5;
@@ -482,11 +482,11 @@ void step_frame(Breakout* env, int action) {
         env->dones[0] = 1;
         env->log.score = env->score;
         add_log(env->log_buffer, &env->log);
-        reset(env);
+        c_reset(env);
     }
 }
 
-void step(Breakout* env) {
+void c_step(Breakout* env) {
     env->dones[0] = 0;
     env->log.episode_length += 1;
     env->rewards[0] = 0.0;
@@ -523,7 +523,7 @@ Client* make_client(Breakout* env) {
     return client;
 }
 
-void render(Client* client, Breakout* env) {
+void c_render(Client* client, Breakout* env) {
     if (IsKeyDown(KEY_ESCAPE)) {
         exit(0);
     }

pufferlib/ocean/breakout/cy_breakout.pyx

@@ -55,9 +55,9 @@ cdef extern from "breakout.h":
 
     Client* make_client(Breakout* env)
     void close_client(Client* client)
-    void render(Client* client, Breakout* env)
-    void reset(Breakout* env)
-    void step(Breakout* env)
+    void c_render(Client* client, Breakout* env)
+    void c_reset(Breakout* env)
+    void c_step(Breakout* env)
 
 cdef class CyBreakout:
     cdef:
@@ -103,12 +103,12 @@ cdef class CyBreakout:
     def reset(self):
         cdef int i
         for i in range(self.num_envs):
-            reset(&self.envs[i])
+            c_reset(&self.envs[i])
 
     def step(self):
         cdef int i
         for i in range(self.num_envs):
-            step(&self.envs[i])
+            c_step(&self.envs[i])
 
     def render(self):
         cdef Breakout* env = &self.envs[0]
@@ -119,7 +119,7 @@ cdef class CyBreakout:
             self.client = make_client(env)
             os.chdir(cwd)
 
-        render(self.client, env)
+        c_render(self.client, env)
 
     def close(self):
         if self.client != NULL:

pufferlib/ocean/connect4/connect4.h

@@ -276,7 +276,7 @@ void compute_observation(CConnect4* env) {
     }
 }
 
-void reset(CConnect4* env) {
+void c_reset(CConnect4* env) {
     env->log = (Log){0};
     env->dones[0] = NOT_DONE;
     env->player_pieces = 0;
@@ -294,13 +294,13 @@ void finish_game(CConnect4* env, float reward) {
     compute_observation(env);
 }
 
-void step(CConnect4* env) {
+void c_step(CConnect4* env) {
     env->log.episode_length += 1;
     env->rewards[0] = 0.0;
 
     if (env->dones[0] == DONE) {
         add_log(env->log_buffer, &env->log);
-        reset(env);
+        c_reset(env);
         return;
     }
 
@@ -359,7 +359,7 @@ Client* make_client(int width, int height) {
     return client;
 }
 
-void render(Client* client, CConnect4* env) {
+void c_render(Client* client, CConnect4* env) {
     if (IsKeyDown(KEY_ESCAPE)) {
         exit(0);
     }

pufferlib/ocean/connect4/cy_connect4.pyx

@@ -36,9 +36,9 @@ cdef extern from "connect4.h":
     void free_cconnect4(CConnect4* env)
     Client* make_client(float width, float height)
     void close_client(Client* client)
-    void render(Client* client, CConnect4* env)
-    void reset(CConnect4* env)
-    void step(CConnect4* env)
+    void c_render(Client* client, CConnect4* env)
+    void c_reset(CConnect4* env)
+    void c_step(CConnect4* env)
 
 cdef class CyConnect4:
     cdef:
@@ -75,12 +75,12 @@ cdef class CyConnect4:
     def reset(self):
         cdef int i
         for i in range(self.num_envs):
-            reset(&self.envs[i])
+            c_reset(&self.envs[i])
 
     def step(self):
         cdef int i
         for i in range(self.num_envs):
-            step(&self.envs[i])
+            c_step(&self.envs[i])
 
     def render(self):
         cdef CConnect4* env = &self.envs[0]
@@ -91,7 +91,7 @@ cdef class CyConnect4:
             self.client = make_client(env.width, env.height)
             os.chdir(cwd)
 
-        render(self.client, env)
+        c_render(self.client, env)
 
     def close(self):
         if self.client != NULL:

pufferlib/ocean/enduro/cy_enduro.pyx

@@ -52,7 +52,7 @@ cdef extern from "enduro.h":
     void free_logbuffer(LogBuffer* buffer)
     Log aggregate_and_clear(LogBuffer* logs)
     void init(Enduro* env, int seed, int env_index)
-    void reset(Enduro* env)
+    void c_reset(Enduro* env)
     void c_step(Enduro* env)
     void c_render(Client* client, Enduro* env)
     Client* make_client(Enduro* env)
@@ -103,15 +103,15 @@ cdef class CyEnduro:
             self.envs[i].log_buffer = self.logs
             self.envs[i].obs_size = observations.shape[1]
 
-            if i % 100 == 0:
-                print(f"Initializing environment #{i} with seed {unique_seed}")
+            #if i % 100 == 0:
+            #    print(f"Initializing environment #{i} with seed {unique_seed}")
 
             init(&self.envs[i], unique_seed, i)
 
     def reset(self):
         cdef int i
         for i in range(self.num_envs):
-            reset(&self.envs[i])
+            c_reset(&self.envs[i])
 
     def step(self):
         cdef int i

pufferlib/ocean/enduro/enduro.h

@@ -475,7 +475,7 @@ void allocate(Enduro* env);
 void init(Enduro* env, int seed, int env_index);
 void free_allocated(Enduro* env);
 void reset_round(Enduro* env);
-void reset(Enduro* env);
+void c_reset(Enduro* env);
 unsigned char check_collision(Enduro* env, Car* car);
 int get_player_lane(Enduro* env);
 float get_car_scale(float y);
@@ -865,7 +865,7 @@ void reset_round(Enduro* env) {
 }
 
 // Reset all init vars; only called once after init
-void reset(Enduro* env) {
+void c_reset(Enduro* env) {
     // No random after first reset
     int reset_seed = (env->reset_count == 0) ? xorshift32(&env->rng_state) : 0;
 

pufferlib/ocean/go/cy_go.pyx

@@ -30,8 +30,6 @@ cdef extern from "go.h":
     int find(Group*)
     void union_groups(Group*, int, int)
 
-
-
     ctypedef struct CGo:
         float* observations
         int* actions
@@ -68,12 +66,12 @@ cdef extern from "go.h":
 
     void init(CGo* env)
     void free_initialized(CGo* env)
-    void reset(CGo* env)
-    void step(CGo* env)
+    void c_reset(CGo* env)
+    void c_step(CGo* env)
 
     Client* make_client(float width, float height)
     void close_client(Client* client)
-    void render(Client* client, CGo* env)
+    void c_render(Client* client, CGo* env)
 
 
 cdef class CyGo:
@@ -122,19 +120,19 @@ cdef class CyGo:
     def reset(self):
         cdef int i
         for i in range(self.num_envs):
-            reset(&self.envs[i])
+            c_reset(&self.envs[i])
 
     def step(self):
         cdef int i
         for i in range(self.num_envs):
-            step(&self.envs[i])
+            c_step(&self.envs[i])
 
     def render(self):
         cdef CGo* env = &self.envs[0]
         if self.client == NULL:
             self.client = make_client(env.width,env.height)
 
-        render(self.client, &self.envs[0])
+        c_render(self.client, &self.envs[0])
 
     def close(self):
         if self.client != NULL:
@@ -144,4 +142,4 @@ cdef class CyGo:
 
     def log(self):
         cdef Log log = aggregate_and_clear(self.logs)
-        return log
+        return log