Refactored agents-key in configuration files (#396)

* changed jsons to new format

* changed test-data

* Adapted to new agent format in json

* Adapted to new agent-structure in json

* Fixed tests

* Adapted to new agents-config format in json

* Adapted to new agents-format in json

* Adapted to new agents-format, upload still not working

* Fixed typo

* Fixed merge for agents

* Fixed tests

* Small refactoring

* Updated docstring

* Added comment

* Added migration

* Recommerce :)

* Updated comment

* Changed interval for updating API indicator

Co-authored-by: Judith <[email protected]>

recommerce/configuration/config_validation.py

@@ -1,3 +1,5 @@
+# This file contains logic used by the webserver to validate configuration files
+
 from recommerce.configuration.environment_config import EnvironmentConfig
 from recommerce.configuration.hyperparameter_config import HyperparameterConfig
 
@@ -11,11 +13,10 @@ def validate_config(config: dict, config_is_final: bool) -> tuple:
 		config_is_final (bool): Whether or not the config must contain all required keys.
 
 	Returns:
-		triple: success: A status (True) and the split hyperparameter_config and environment_config dictionaries as a tuple.
+		tuple: success: A status (True) and the split hyperparameter_config and environment_config dictionaries as a tuple.
 				failure: A status (False) and the errormessage as a string.
 	"""
 	try:
-		# validate the config using `recommerce` validation logic
 		# first check if the environment and hyperparameter parts are already split up
 		if 'environment' in config and 'hyperparameter' in config:
 			assert len(config) == 2, 'Your config should not contain keys other than "environment" and "hyperparameter"'
@@ -54,11 +55,11 @@ def validate_sub_keys(config_class: HyperparameterConfig or EnvironmentConfig, c
 		AssertionError: If the given config contains a key that is invalid.
 	"""
 	for key, _ in config.items():
-		# TODO: Remove this workaround with the agent-rework in the config files
+		# we need to separately check agents, since it is a list of dictionaries
 		if key == 'agents':
-			for agent_key in config['agents'].keys():
-				assert all(this_key in {'agent_class', 'argument'} for this_key in config['agents'][agent_key]), \
-					f'an invalid key for agents was provided: {config["agents"][agent_key].keys()}'
+			for agent in config['agents']:
+				assert all(agent_key in {'name', 'agent_class', 'argument'} for agent_key in agent.keys()), \
+					f'an invalid key for agents was provided: {agent.keys()}'
 		# the key is key of a dictionary in the config
 		elif top_level_keys[key]:
 			assert isinstance(config[key], dict), f'The value of this key must be of type dict: {key}, but was {type(config[key])}'
@@ -128,7 +129,8 @@ def check_config_types(hyperparameter_config: dict, environment_config: dict, mu
 
 	# check types for environment_config
 	task = environment_config['task'] if must_contain else 'None'
-	EnvironmentConfig.check_types(environment_config, task, must_contain)
+	single_agent = task in ['training', 'exampleprinter']
+	EnvironmentConfig.check_types(environment_config, task, single_agent, must_contain)
 
 
 if __name__ == '__main__':  # pragma: no cover
@@ -152,16 +154,18 @@ def check_config_types(hyperparameter_config: dict, environment_config: dict, mu
 			'storage_cost_per_product': 0.1
 		},
 		'episodes': 5,
-		'agents': {
-			'CE Rebuy Agent (QLearning)': {
+		'agents': [
+			{
+				'name': 'CE Rebuy Agent (QLearning)',
 				'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
 				'argument': ''
 			},
-			'CE Rebuy Agent (QLearaning)': {
+			{
+				'name': 'CE Rebuy Agent (QLearaning)',
 				'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
 				'argument': ''
 			}
-		}
+		]
 	}
 	hyper, env = split_combined_config(test_config)
 	check_config_types(hyper, env)

recommerce/default_data/configuration_files/environment_config_agent_monitoring.json

@@ -4,14 +4,16 @@
 	"episodes": 500,
 	"plot_interval": 100,
 	"marketplace": "recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario",
-	"agents": {
-		"Rule_Based Agent": {
+	"agents": [
+		{
+			"name": "Rule_Based Agent",
 			"agent_class": "recommerce.market.circular.circular_vendors.RuleBasedCERebuyAgent",
 			"argument": ""
 		},
-		"CE Rebuy Agent (QLearning)": {
+		{
+			"name": "CE Rebuy Agent (QLearning)",
 			"agent_class": "recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent",
 			"argument": "CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat"
 		}
-	}
+	]
 }

recommerce/default_data/configuration_files/environment_config_exampleprinter.json

@@ -1,10 +1,11 @@
 {
 	"task": "exampleprinter",
 	"marketplace": "recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceOneCompetitor",
-	"agents": {
-		"CE Rebuy Agent (QLearning)": {
+	"agents": [
+		{
+			"name": "CE Rebuy Agent (QLearning)",
 			"agent_class": "recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent",
 			"argument": "CircularEconomyRebuyPriceOneCompetitor_QLearningCERebuyAgent.dat"
 		}
-	}
+	]
 }

recommerce/default_data/configuration_files/environment_config_training.json

@@ -1,10 +1,11 @@
 {
 	"task": "training",
 	"marketplace": "recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario",
-	"agents": {
-		"CE Rebuy Agent (QLearning)": {
+	"agents": [
+		{
+			"name": "CE Rebuy Agent (QLearning)",
 			"agent_class": "recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent",
 			"argument": ""
 		}
-	}
+	]
 }

recommerce/monitoring/exampleprinter.py

@@ -101,17 +101,18 @@ def main():  # pragma: no cover
 	printer = ExamplePrinter()
 
 	config: ExampleprinterEnvironmentConfig = EnvironmentConfigLoader.load('environment_config_exampleprinter')
+	# TODO: Theoretically, the name of the agent is saved in config['name'], but we don't use it yet.
 	marketplace = config.marketplace()
 
 	# QLearningAgents need more initialization
-	if issubclass(config.agent[0], QLearningAgent):
+	if issubclass(config.agent['agent_class'], QLearningAgent):
 		printer.setup_exampleprinter(marketplace=marketplace,
-			agent=config.agent[0](
+			agent=config.agent['agent_class'](
 				n_observations=marketplace.observation_space.shape[0],
 				n_actions=marketplace.get_n_actions(),
-				load_path=os.path.abspath(os.path.join(PathManager.data_path, config.agent[1]))))
+				load_path=os.path.abspath(os.path.join(PathManager.data_path, config.agent['argument']))))
 	else:
-		printer.setup_exampleprinter(marketplace=marketplace, agent=config.agent[0]())
+		printer.setup_exampleprinter(marketplace=marketplace, agent=config.agent['agent_class']())
 
 	print(f'The final profit was: {printer.run_example()}')
 

recommerce/rl/training_scenario.py

@@ -57,9 +57,8 @@ def train_from_config():
 	Use the `environment_config_training.json` file to decide on the training parameters.
 	"""
 	config: TrainingEnvironmentConfig = EnvironmentConfigLoader.load('environment_config_training')
-	# Since we store a tuple (class, None) in config.agent, we just want the first component
-	# Since we store an one-element list [class] in config.marketplace, we just want the stored element
-	run_training_session(config.marketplace, config.agent)
+	# TODO: Theoretically, the name of the agent is saved in config['name'], but we don't use it yet.
+	run_training_session(config.marketplace, config.agent['agent_class'])
 
 
 def main():

tests/test_data/environment_config_agent_monitoring.json

@@ -4,14 +4,16 @@
 	"episodes": 50,
 	"plot_interval": 25,
 	"marketplace": "recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario",
-	"agents": {
-		"Rule_Based Agent": {
+	"agents": [
+		{
+			"name": "Rule_Based Agent",
 			"agent_class": "recommerce.market.circular.circular_vendors.RuleBasedCERebuyAgent",
 			"argument": ""
 		},
-		"CE Rebuy Agent (QLearning)": {
+		{
+			"name": "CE Rebuy Agent (QLearning)",
 			"agent_class": "recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent",
 			"argument": "CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat"
 		}
-	}
+	]
 }

tests/test_data/environment_config_exampleprinter.json

@@ -1,10 +1,11 @@
 {
 	"task": "exampleprinter",
 	"marketplace": "recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceOneCompetitor",
-	"agents": {
-		"CE Rebuy Agent (QLearning)": {
+	"agents": [
+		{
+			"name": "CE Rebuy Agent (QLearning)",
 			"agent_class": "recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent",
 			"argument": "CircularEconomyRebuyPriceOneCompetitor_QLearningCERebuyAgent.dat"
 		}
-	}
+	]
 }

tests/test_data/environment_config_training.json

@@ -1,10 +1,11 @@
 {
 	"task": "training",
 	"marketplace": "recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario",
-	"agents": {
-		"CE Rebuy Agent (QLearning)": {
+	"agents": [
+		{
+			"name": "CE Rebuy Agent (QLearning)",
 			"agent_class": "recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent",
 			"argument": ""
 		}
-	}
+	]
 }

tests/test_environment_config.py

@@ -9,68 +9,6 @@
 from recommerce.market.circular.circular_vendors import RuleBasedCERebuyAgent
 from recommerce.rl.q_learning.q_learning_agent import QLearningCERebuyAgent
 
-valid_training_dict = {
-	'task': 'training',
-	'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
-	'agents': {
-		'CE Rebuy Agent (QLearning)': {
-			'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
-			'argument': ''
-		}
-	}
-}
-
-valid_agent_monitoring_dict = {
-	'task': 'agent_monitoring',
-	'enable_live_draw': False,
-	'episodes': 10,
-	'plot_interval': 5,
-	'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
-	'agents': {
-		'Rule_Based Agent': {
-			'agent_class': 'recommerce.market.circular.circular_vendors.RuleBasedCERebuyAgent',
-			'argument': ''
-		},
-		'CE Rebuy Agent (QLearning)': {
-			'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
-			'argument': 'CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat'
-		}
-	}
-}
-
-valid_exampleprinter_dict = {
-	'task': 'exampleprinter',
-	'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
-	'agents': {
-		'CE Rebuy Agent (QLearning)': {
-			'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
-			'argument': 'CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat'
-		}
-	}
-}
-
-invalid_agent_dict = {
-	'task': 'exampleprinter',
-	'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
-	'agents': {
-		'Agent_name': {
-			'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
-			'argument': ''
-		}
-	}
-}
-
-invalid_task_dict = {
-	'task': 'not_existing_test_task',
-	'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
-	'agents': {
-		'Agent_name': {
-			'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
-			'argument': 'CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat'
-		}
-	}
-}
-
 
 def test_abstract_parent_class():
 	with pytest.raises(TypeError) as error_message:
@@ -79,10 +17,22 @@ def test_abstract_parent_class():
 
 
 def test_str_representation():
-	config = env_config.TrainingEnvironmentConfig(valid_training_dict)
+	test_dict = {
+		'task': 'training',
+		'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
+		'agents': [
+			{
+				'name': 'CE Rebuy Agent (QLearning)',
+				'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
+				'argument': ''
+			}
+		]
+	}
+	config = env_config.TrainingEnvironmentConfig(test_dict)
 	assert str(config) == ("TrainingEnvironmentConfig: {'task': 'training', "
-		"'agent': <class 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent'>, "
-		"'marketplace': <class 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario'>}")
+		"'marketplace': <class 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario'>, "
+		"'agent': {'name': 'CE Rebuy Agent (QLearning)', "
+		"'agent_class': <class 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent'>, 'argument': ''}}")
 
 
 get_class_testcases = [
@@ -138,9 +88,47 @@ def test_get_task(tested_class: env_config.EnvironmentConfig, expected_task):
 
 
 valid_ConfigLoader_validate_testcases = [
-	valid_training_dict,
-	valid_agent_monitoring_dict,
-	valid_exampleprinter_dict
+	{
+		'task': 'training',
+		'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
+		'agents': [
+			{
+				'name': 'CE Rebuy Agent (QLearning)',
+				'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
+				'argument': ''
+			}
+		]
+	},
+	{
+		'task': 'agent_monitoring',
+		'enable_live_draw': False,
+		'episodes': 10,
+		'plot_interval': 5,
+		'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
+		'agents': [
+			{
+				'name': 'Rule_Based Agent',
+				'agent_class': 'recommerce.market.circular.circular_vendors.RuleBasedCERebuyAgent',
+				'argument': ''
+			},
+			{
+				'name': 'CE Rebuy Agent (QLearning)',
+				'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
+				'argument': 'CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat'
+			}
+		]
+	},
+	{
+		'task': 'exampleprinter',
+		'marketplace': 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
+		'agents': [
+			{
+				'name': 'CE Rebuy Agent (QLearning)',
+				'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
+				'argument': 'CircularEconomyRebuyPriceMonopolyScenario_QLearningCERebuyAgent.dat'
+			}
+		]
+	}
 ]
 
 
@@ -152,9 +140,10 @@ def test_valid_ConfigLoader_validate(config):
 valid_ConfigLoader_load_training_testcases = [
 	# TODO: Currently no testcases for ActorCriticAgents
 	('training', 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceMonopolyScenario',
-		{'CE Rebuy Agent (QLearning)': {'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent', 'argument': ''}}),
+		[{'name': 'CE Rebuy Agent (QLearning)', 'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCERebuyAgent',
+			'argument': ''}]),
 	('training', 'recommerce.market.circular.circular_sim_market.CircularEconomyRebuyPriceOneCompetitor',
-		{'CE Rebuy Agent (QLearning)': {'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCEAgent', 'argument': ''}})
+		[{'name': 'CE Rebuy Agent (QLearning)', 'agent_class': 'recommerce.rl.q_learning.q_learning_agent.QLearningCEAgent', 'argument': ''}])
 ]
 
 

webserver/alpha_business_app/buttons.py

@@ -292,13 +292,9 @@ def _start(self) -> HttpResponse:
 		# convert post request to normal dict
 		post_request = dict(self.request.POST.lists())
 
-		try:
-			config_dict = ConfigFlatDictParser().flat_dict_to_hierarchical_config_dict(post_request)
-		except AssertionError:
-			self.message = ['error', 'Could not create config: Please eliminate identical Agent names']
-			return self._decide_rendering()
+		config_dict = ConfigFlatDictParser().flat_dict_to_hierarchical_config_dict(post_request)
 
-		validate_status, validate_data = validate_config(config_dict, True)
+		validate_status, validate_data = validate_config(config=config_dict, config_is_final=True)
 		if not validate_status:
 			self.message = ['error', validate_data]
 			return self._decide_rendering()
@@ -311,16 +307,15 @@ def _start(self) -> HttpResponse:
 			if Container.objects.filter(id=response['id']).exists():
 				# we will kindly ask the user to try it again and stop the container
 				# TODO insert better handling here
-				print('the new container has the same id, as another container')
-				self.message = ['error', 'please try again']
+				self.message = ['error', 'The new container has the same id as an already existing container, please try again.']
 				return self._remove()
 			# get all necessary parameters for container object
 			container_name = self.request.POST['experiment_name']
 			container_name = container_name if container_name != '' else response['id'][:10]
 			config_object = ConfigModelParser().parse_config(copy.deepcopy(config_dict))
 			command = config_object.environment.task
 			Container.objects.create(id=response['id'], config=config_object, name=container_name, command=command)
-			config_object.name = f'used for {container_name}'
+			config_object.name = f'Config for {container_name}'
 			config_object.save()
 			return redirect('/observe', {'success': 'You successfully launched an experiment'})
 		else: