Conversation
| for t_idx in self.t_idxes: | ||
| tm = self.get_teammate_from_idx(t_idx) | ||
| if type(tm) != CustomAgent: | ||
| # if not isinstance(tm, CustomAgent): |
There was a problem hiding this comment.
I think we are good to not have this comment.
| @@ -6,7 +6,7 @@ HOW_LONG=20 | |||
| NUM_OF_CKPOINTS=40 | |||
| LAYOUT_NAMES="c4_v2" | |||
| EXP_DIR=${LAYOUT_NAMES} | |||
There was a problem hiding this comment.
Should EXP_DIR be ${LAYOUT_NAMES}/${NUM_PLAYER}, instead?
There was a problem hiding this comment.
This is just an experiment that I decided to run this way. For the final experiments we can set them be layout_names/num_player
| HOW_LONG=20 | ||
| NUM_OF_CKPOINTS=40 | ||
| LAYOUT_NAMES="counter_circuit,coordination_ring,cramped_room,asymmetric_advantages,forced_coordination" | ||
| EXP_DIR="Classic/$NUM_PLAYERS" # When quick_test=True this will be overwritten to "Test/$EXP_DIR" |
There was a problem hiding this comment.
I suppose we want to save agents under a directory, mentioning $NUM_PLAYERS, like this one. Although we have it in this scripts, many scripts do not have EXP_DIR mentioning $NUM_PLAYERS.
There was a problem hiding this comment.
The directory of best_baseline_experiment is already executed so I am not going to change their scripts. Whenever we add a new script we can follow this template
|
|
||
|
|
||
| def best_EGO(args, add_adv=False) -> None: | ||
| '''for a very specifric experimetn: only for 2 players:: ignore this''' |
There was a problem hiding this comment.
a little typo here. experimetn
Jmiceli17
left a comment
Jmiceli17
left a comment
There was a problem hiding this comment.
Awesome! Just a few small questions
| self.agent = agent | ||
| self.p_idx = p_idx | ||
| self.env.set_teammates(teammates) | ||
| self.env.set_teammates(teamtype='run_type') |
There was a problem hiding this comment.
set_teammates just checks if teamtype is set right? I think it'd be better to have teamtype=True here then so we don't think there's something special about the 'run_type' string
There was a problem hiding this comment.
what's the difference between this and sandbox/profile_analyze.py? do we need both?
| print("Training agent: " + self.name + ", for experiment: " + experiment_name) | ||
| self.print_tc_helper(self.teammates_collection, "Train TC") | ||
| self.print_tc_helper(self.eval_teammates_collection, "Eval TC") | ||
| self.print_tc_helper(self.eval_envs[0].teammates_collection[TeammatesCollection.EVAL], "Eval TC") |
There was a problem hiding this comment.
Are the teammate collections are always going to be uniform?
4 participants
This PR sets the default computation device to CPU and enables compatibility with SubProcEnv, which allows the code to leverage multiple processes for parallel execution. To support this, the logic for teammate_collection has to be moved to base_overcooked_env.
Currently, the default is configured not to use multiple processes because of memory issues (the higher the n_envs the more memory). Until that is resolved, we continue to use DummyVecEnv by default.