Skip to content

Commit

Permalink
fix broken link
Browse files Browse the repository at this point in the history
  • Loading branch information
saleml committed Aug 3, 2023
1 parent 2c5c869 commit 8c157d9
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion tutorials/ENV.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ The logic of the environment is handled by the methods `maskless_step` and `mask
- The `log_reward` function that assigns the logarithm of a nonnegative reward to every terminating state (i.e. state with all $s_f$ as a child in the DAG). If `log_reward` is not implemented, `reward` needs to be.


For `DiscreteEnv`s, the user can define a`get_states_indices` method that assigns a unique integer number to each state, and a `n_states` property that returns an integer representing the number of states (excluding $s_f$) in the environment. The function `get_terminating_states_indices` can also be implemented and serves the purpose of uniquely identifying terminating states of the environment, which is useful for [tabular `GFNModule`s](https://github.com/saleml/torchgfn/tree/master/src/gfn/estimators.py). Other properties and functions can be implemented as well, such as the `log_partition` or the `true_dist_pmf` properties.
For `DiscreteEnv`s, the user can define a`get_states_indices` method that assigns a unique integer number to each state, and a `n_states` property that returns an integer representing the number of states (excluding $s_f$) in the environment. The function `get_terminating_states_indices` can also be implemented and serves the purpose of uniquely identifying terminating states of the environment, which is useful for [tabular `GFNModule`s](https://github.com/saleml/torchgfn/tree/master/src/gfn/utils/modules.py). Other properties and functions can be implemented as well, such as the `log_partition` or the `true_dist_pmf` properties.

For reference, it might be useful to look at one of the following provided environments:
- [HyperGrid](https://github.com/saleml/torchgfn/tree/master/src/gfn/gym/hypergrid.py) is an example of a discrete environment where all states are terminating states. - [DiscreteEBM](https://github.com/saleml/torchgfn/tree/master/src/gfn/gym/discrete_ebm.py) is an example of a discrete environment where all trajectories are of the same length but only some states are terminating.
Expand Down

0 comments on commit 8c157d9

Please sign in to comment.