An alias for Adam optimizer. .. code-block:: python from d3rlpy.optimizers import AdamFactory factory = AdamFactory(weight_deca

Represents a fully-connected layer in a network.

mini-batch of Transition objects. This class is designed to hold :class:`d3rlpy.dataset.Transition` objects for being passed to

Represents a fully-connected layer in a network.

Transition class. This class is designed to hold data between two time steps, which is usually used as inputs of loss calculatio

Markov-Decision Process Dataset class. MDPDataset is deisnged for reinforcement learning datasets to use them like supervised le

GPU Device class. This class manages GPU id. The purpose of this device class instead of PyTorch device class is to assign GPU ids when t

StackedObservation class. This class is used to stack images to handle temporal features. References: * `Mnih et al., Human-level co

Vector encoder factory class. This is the default encoder factory for vector observation. Args: hidden_units (list): list of hidden

Simple history container used inside the search. Only used to keep track of the actions executed.

Channel-first wrapper for image observation environments. d3rlpy expects channel-first images since it's built with PyTorch. You can transfor

Standard Replay Buffer. Args: maxlen (int): the maximum number of data length. env (gym.Env): gym-like environment to extract sha

Neural network models which model aleatoric uncertainty (and possibly epistemic uncertainty with ensembling).

Simple FIFO queue implementation. Random access of this queue object is O(1).

Store only useful information of a self-play game.

Core Monte Carlo Tree Search algorithm. To decide on an action, we run N simulations, always starting at the root of the search tree and

Pixel encoder factory class. This is the default encoder factory for image observation. Args: filters (list): list of tuples consist

r"""Invertible transformation (bijector) that computes :math:`Y = tanh(X)`, therefore :math:`Y \in (-1, 1)`. This can be achieved by an affin

Mock state space model. The transition function is to add the action to the observation. The posterior function is to return the ground truth obs

Dataclass keeping all data for one agent/lumberjack in environment. In order to keep the support for Python3.6 we are not using `dataclasses` modu

r"""Atari 2600 preprocessings. This class follows the guidelines in Machado et al. (2018), "Revisiting the Arcade Learning Environment: Ev

Neural network models which model aleatoric uncertainty (and possibly epistemic uncertainty with ensembling).

r"""Conservative Offline Model-Based Optimization. COMBO is a model-based RL approach for offline policy optimization. COMBO is similar to MO