pulsatrix/api/gflownet__trajectory_8hpp_source.html

#pragma once


#include <cstdint>

#include <vector>


#include "pulsatrix/gflownet_forward_policy.hpp"

#include "pulsatrix/hypergrid_env.hpp"

#include "pulsatrix/tensor.hpp"


namespace pulsatrix {


struct GFlowNetTrajectory {

    std::vector<Tensor> states;

    std::vector<int64_t> actions;

    float sum_log_pf = 0.0f;

    float sum_log_pb = 0.0f;

    float terminal_reward = 0.0f;

};


[[nodiscard]] GFlowNetTrajectory sample_gflownet_trajectory(HyperGridEnv& env, GFlowNetForwardPolicy& forward_policy);


}  // namespace pulsatrix

pulsatrix::GFlowNetForwardPolicy
Samples from softmax(mask(policy_network(observation))) – a categorical policy restricted to a caller...
Definition gflownet_forward_policy.hpp:47

pulsatrix::HyperGridEnv
An n-dimensional grid: state is an integer coordinate in [0, H-1]^ndim, actions increment one coordin...
Definition hypergrid_env.hpp:31

gflownet_forward_policy.hpp
Masked stochastic categorical policy over a logit-producing Module – P_F for GFlowNet training object...

hypergrid_env.hpp
HyperGrid – the standard minimal GFlowNet correctness-check environment.

pulsatrix
Definition acquisition_functions.hpp:16

pulsatrix::sample_gflownet_trajectory
GFlowNetTrajectory sample_gflownet_trajectory(HyperGridEnv &env, GFlowNetForwardPolicy &forward_policy)
Samples one full trajectory: resets env, then repeatedly samples a masked action from forward_policy ...

pulsatrix::GFlowNetTrajectory
One full sampled trajectory: every state the forward policy acted from, the action taken at each,...
Definition gflownet_trajectory.hpp:26

pulsatrix::GFlowNetTrajectory::actions
std::vector< int64_t > actions
The action index sampled at each corresponding entry of states.
Definition gflownet_trajectory.hpp:30

pulsatrix::GFlowNetTrajectory::sum_log_pf
float sum_log_pf
Σ_t log P_F(s_{t+1}|s_t) over the whole trajectory, including the final stop decision.
Definition gflownet_trajectory.hpp:33

pulsatrix::GFlowNetTrajectory::terminal_reward
float terminal_reward
R(x) at the trajectory's terminal state.
Definition gflownet_trajectory.hpp:38

pulsatrix::GFlowNetTrajectory::sum_log_pb
float sum_log_pb
Σ_t log P_B(s_t|s_{t+1}) over every real move (the stop action does not move the state,...
Definition gflownet_trajectory.hpp:36

pulsatrix::GFlowNetTrajectory::states
std::vector< Tensor > states
The state the policy acted from at each decision point, in order.
Definition gflownet_trajectory.hpp:28

tensor.hpp
N-dimensional tensor – owns a buffer via DeviceBackend*, RAII (Rule of Five).