The policy acts greedily with probability :math:`1-\epsilon` and acts randomly otherwise. It is now used as a default policy for the neural agent.

The NeuralAgent class wraps a learning algorithm (such as a deep Q-network) for training and testing in a given environment. Attach controlle

Deep Q-learning network using Keras (with any backend) Parameters ----------- environment : object from class Environment

Exception raised for errors when calling the various Agent methods at wrong times. Attributes: expr -- input expression in which the error

Exception raised for errors when getting slices from CircularBuffers. Attributes: expr -- input expression in which the error occurred

Combined Reinforcement learning via Abstract Representations (CRAR) using Keras Parameters ----------- environment : object from

A replay memory consisting of circular buffers for observations, actions, rewards and terminals.

Simple alternative to :math:`\epsilon`-greedy that can explore more efficiently for a broad class of realistic problems. Parameters -----

Actor-critic learning (using Keras) with Deep Deterministic Policy Gradient (DDPG) for the continuous action domain Parameters -

Warning issued of the various Agent methods. Attributes: expr -- input expression in which the error occurred msg -- explanation

A base controller that does nothing when receiving the various signals emitted by an agent. This class should be the base class of any controller

A controller that modifies the q-network discount periodically (only in training mode, i.e., agent.mode() == -1). More informations in : Francois-

All your Environment classes should inherit this interface. The environment defines the dynamics and the reward signal that the agent observe

A controller that modifies the probability "epsilon" of taking a random action periodically (only in training mode, i.e., agent.mode() == -1).

A controller that finds the neural net performing at best in validation mode (i.e. for mode = [validationID]) and computes the associated general

A controller that interleaves a valid/test epoch between training epochs of the agent (only in training mode, i.e., agent.mode() == -1). Para

All the Q-networks, actor-critic networks, etc. should inherit this interface. Parameters ----------- environment : object from class En

A controller that modifies the learning rate periodically upon epochs end (only in training mode, i.e., agent.mode() == -1). Parameters -

Deep Q-learning network using Keras Parameters ----------- batch_size : int Number of tuples taken into account for each

Deep Q-learning network with LSTM's using Keras Parameters ----------- batch_size : int Number of tuples taken into acco

Deep Q-learning network using Keras Parameters ----------- batch_size : int Number of tuples taken into account for each

Abstract class for all policies. A policy takes observations as input, and outputs an action. Parameters ----------- learning_algo :

A controller that makes the agent train on its current database periodically (only in training mode, i.e., agent.mode() == -1). Parameters --

A controller that print various agent information periodically: * Count of passed [evaluateOn] * Agent current learning rate * Agent