RPLH-Vanilla Weird Action Issue

[KevinBian107]

RE empty extraction issue, please use the updated Re prompt that Jack fixed

Example:

----------EXECUTION AVAILABILITY CHECKER---------- JUDGE MODIFIED: {} -------###-------###-------###-------LOCAL_ROW_0_COL_1-------###-------###-------###------- -------###-------###-------###-------JUDGE_ON_ROW_0_COL_1-------###-------###-------###------- Match response:{'0.5_0.5': ['target_blue', 'target_red', 'target_purple', 'target_purple', 'box_orange'], '0.5_1.5': ['box_red', 'target_green', 'box_purple', 'target_orange', 'box_blue'], '1.5_0.5': ['target_green', 'box_purple'], '1.5_1.5': ['box_purple', 'target_purple']} Processed response:{}

----------EXECUTION AVAILABILITY CHECKER---------- JUDGE MODIFIED: {} ORIGINAL PLAN: {} -------###-------###-------###-------LOCAL_ROW_1_COL_0-------###-------###-------###------- -------###-------###-------###-------JUDGE_ON_ROW_1_COL_0-------###-------###-------###------- Match response:{'0.5_0.5': ['target_blue', 'target_red', 'target_purple', 'target_purple', 'box_orange'], '0.5_1.5': ['box_red', 'target_green', 'box_purple', 'target_orange', 'box_blue'], '1.5_0.5': ['target_green', 'box_purple'], '1.5_1.5': ['box_purple', 'target_purple']} Processed response:{}

----------EXECUTION AVAILABILITY CHECKER---------- JUDGE MODIFIED: {} -------###-------###-------###-------LOCAL_ROW_1_COL_1-------###-------###-------###------- -------###-------###-------###-------JUDGE_ON_ROW_1_COL_1-------###-------###-------###------- Match response:{'0.5_0.5': ['target_blue', 'target_red', 'target_purple', 'target_purple', 'box_orange'], '0.5_1.5': ['box_red', 'target_green', 'box_purple', 'target_orange', 'box_blue'], '1.5_0.5': ['target_green', 'box_purple'], '1.5_1.5': ['box_purple', 'target_purple']} Processed response:{}

----------EXECUTION AVAILABILITY CHECKER---------- JUDGE MODIFIED: {} ORIGINAL PLAN: {}

[KevinBian107]

• Fixe agent passed in lcoation issue, this may be a issue coming from the environment representation?
• When no history option is given, this solves the issue easily, need more testing though.

[KevinBian107]

Match response:{ "Agent[0.5, 0.5]": "move(box_red, target_red)", "Agent[0.5, 1.5]": null, "Agent[1.5, 0.5]": null, "Agent[1.5, 1.5]": null } Agent-action pair is inconsistent: { "Agent[0.5, 0.5]": "move(box_red, target_red)", "Agent[0.5, 1.5]": null, "Agent[1.5, 0.5]": null, "Agent[1.5, 1.5]": null } Processed response:{ "Agent[0.5, 0.5]": "move(box_red, target_red)", "Agent[0.5, 1.5]": null, "Agent[1.5, 0.5]": null, "Agent[1.5, 1.5]": null }

need to directly throw away no response agent stuff

[KevinBian107]

HISTORY METHOD: _w_no_history Match response:{ "action": "move(box_blue, square[1.5, 0.5])" } Agent-action pair is inconsistent: { "action": "move(box_blue, square[1.5, 0.5])" } Processed response:{ "action": "move(box_blue, square[1.5, 0.5])" }

[KevinBian107]

----------RETAKE ACTION---------- Traceback (most recent call last): File "/Users/kevinb/Desktop/dsc190/RPLH/rplh_vanilla/rplh_inference.py", line 584, in ) = run_exp( File "/Users/kevinb/Desktop/dsc190/RPLH/rplh_vanilla/rplh_inference.py", line 197, in run_exp response, token_num_count_list_add = with_action_syntactic_check_func( File "/Users/kevinb/Desktop/dsc190/RPLH/rplh_vanilla/execution_checker.py", line 271, in with_action_syntactic_check_func response, token_num_count_list_add = retake_action( File "/Users/kevinb/Desktop/dsc190/RPLH/rplh_vanilla/execution_checker.py", line 172, in retake_action retake_action_prompt_1 = prompt_func(feedback) TypeError: rplh_prompt_func() got multiple values for argument 'state_update_prompt' (rplh) (base) kevinb@Kevins-MacBook-Pro-2 RPLH %

[KevinBian107]

Fixed issue, wouldn't error out, but stuck in RETAKE ACTION while loop