gflownet.proxy.tree

Classes

TreeProxy

Module Contents

class gflownet.proxy.tree.TreeProxy(use_prior=True, beta=1.0, **kwargs)[source]

Bases: gflownet.proxy.base.Proxy

Parameters:

  • use_prior (bool) – Whether to use -likelihood * prior for energy computation or just the -likelihood.

  • beta (float) – Beta coefficient in prior = np.exp(-self.beta * n_nodes). Note that this is temporary prior implementation that was used for debugging, in combination with reward_func=”boltzmann” it doesn’t make much sense.

use_prior = True[source]
beta = 1.0[source]
X = None[source]
y = None[source]
setup(env=None)[source]
Parameters:

env (Optional[gflownet.envs.tree.Tree])

__call__(states)[source]

Implement this function to call the get_reward method of the appropriate Proxy Class (EI, UCB, Proxy, Oracle etc).

Parameters:

states (ndarray)

Return type:

torchtyping.TensorType[batch]