from rl_coach.agents.n_step_q_agent import NStepQAgentParameters
from rl_coach.architectures.layers import Conv2d, Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection
from rl_coach.environments.gym_environment import Atari, atari_deterministic_v4
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters

####################
# Graph Scheduling #
####################
schedule_params = ScheduleParameters()
schedule_params.improve_steps = TrainingSteps(10000000000)
schedule_params.steps_between_evaluation_periods = EnvironmentEpisodes(100)
schedule_params.evaluation_steps = EnvironmentEpisodes(3)
schedule_params.heatup_steps = EnvironmentSteps(0)

#########
# Agent #
#########
agent_params = NStepQAgentParameters()

agent_params.network_wrappers['main'].learning_rate = 0.0001
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Conv2d(16, 8, 4),
                                                                                          Conv2d(32, 4, 2)]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense(256)]

###############
# Environment #
###############
env_params = Atari(level=SingleLevelSelection(atari_deterministic_v4))

########
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.trace_test_levels = ['breakout', 'pong', 'space_invaders']

graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
                                    schedule_params=schedule_params, vis_params=VisualizationParameters(),
                                    preset_validation_params=preset_validation_params)