Removed more old scripts. Added header comments to all scripts.

gui.py

@@ -1,5 +1,9 @@
 #!/usr/bin/env python3
 
+"""
+Sample GUI application to interact with BabyAI levels
+"""
+
 import time
 import sys
 import threading

run_tests.py

@@ -1,5 +1,10 @@
 #!/usr/bin/env python3
 
+"""
+Run basic BabyAI level tests
+Note: there are other automated tests in .circleci/config.yml
+"""
+
 import babyai
 from babyai import levels
 from babyai import agents

scripts/compute_possible_instructions.py

@@ -1,5 +1,9 @@
 #!/usr/bin/env python3
 
+"""
+Compute the number of possible instructions in the BabyAI grammar.
+"""
+
 from gym_minigrid.minigrid import COLOR_NAMES
 
 def count_Sent():

scripts/enjoy.py

@@ -3,6 +3,7 @@
 """
 Visualize the performance of a model on a given environment.
 """
+
 import argparse
 import gym
 import time

scripts/eval_bot.py

@@ -1,5 +1,10 @@
 #!/usr/bin/env python3
 
+"""
+Evaluate the success rate of the bot
+This script is used for testing/debugging purposes
+"""
+
 import random
 import time
 from optparse import OptionParser

scripts/evaluate.py

@@ -1,5 +1,9 @@
 #!/usr/bin/env python3
 
+"""
+Evaluate a trained model or bot
+"""
+
 import argparse
 import gym
 import time

scripts/evaluate_all_demos.py

@@ -1,7 +1,10 @@
-'''
-Script to evaluate all available demos. It assumes all demos (human and agent, except the "valid" ones)
+"""
+Script to evaluate all available demos.
+
+Assumes all demos (human and agent, except the "valid" ones)
 are generated with seed 1
-'''
+"""
+
 import os
 from subprocess import call
 import sys
@@ -19,4 +22,4 @@
         print("> Env: {} - {}".format(env, demos))
         seed = 0 if demos.endswith('valid') else 1
         command = ["python evaluate.py --env {} --demos {} --seed {}".format(env, demos, seed)] + sys.argv[1:]
-        call(" ".join(command), shell=True)
+        call(" ".join(command), shell=True)

scripts/evaluate_all_models.py

@@ -1,9 +1,11 @@
-"""Evaluate all models in a storage directory.
+"""
+Evaluate all models in a storage directory.
 
 In order to use this script make sure to add baby-ai-game/scripts to the $PATH
 environment variable.
-Example of usage: evaluate_all_models.py --episodes 200 --argmax
 
+Sample usage:
+evaluate_all_models.py --episodes 200 --argmax
 """
 
 import os

scripts/make_agent_demos.py

@@ -1,5 +1,11 @@
 #!/usr/bin/env python3
 
+"""
+Generate a set of agent demonstrations.
+
+The agent can either be a trained model or the heuristic expert (bot)
+"""
+
 import argparse
 import gym
 import logging
@@ -194,4 +200,4 @@ def generate_demos(n_episodes, valid, seed, shift=0):
 
     # Validation demos
     if args.valid_episodes:
-        generate_demos(args.valid_episodes, True, 0)
+        generate_demos(args.valid_episodes, True, 0)

scripts/make_human_demos.py

@@ -1,5 +1,9 @@
 #!/usr/bin/env python3
 
+"""
+Generate a set of human demonstrations
+"""
+
 import sys
 import copy
 import random

scripts/show_level_instructions.py

@@ -1,4 +1,6 @@
-"""Randomly sample instructions from a level."""
+"""
+Randomly sample and print out instructions from a level.
+"""
 
 import argparse
 

scripts/train_il.py

@@ -3,6 +3,7 @@
 """
 Script to train agent through imitation learning using demonstrations.
 """
+
 import os
 import argparse
 import csv

scripts/train_intelligent_expert.py

@@ -1,8 +1,14 @@
 #!/usr/bin/env python3
 
 """
-Train an agent using an intelligent expert
+Train an agent using an intelligent expert.
 
+The procedure starts with a small set of training demonstrations, and
+iteratively grows the training set by some percentage. At every step, the new
+demos used to grow the training set are demos the agent is currently failing
+on. A new model is trained from scratch at every step.
+
+Sample usage:
 scripts/train_intelligent_expert.py --env BabyAI-GoToObj-v0 --demos GoToObj-bot-100k --validation-interval 5
 
 Vanilla imitation learning:

test_mission_gen.py

@@ -1,5 +1,13 @@
 #!/usr/bin/env python3
 
+"""
+Test the level/mission generation code.
+
+This script allows users to visualize generated missions for a given
+level and control the agent manually. The instruction string for the
+current mission is printed in the console.
+"""
+
 import random
 import time
 from optparse import OptionParser