Resolve conflicts

alexhernandezgarcia · May 28, 2024 · 92d45b7 · 92d45b7
2 parents 4c6efa0 + bcb1443
commit 92d45b7
Show file tree

Hide file tree

Showing 53 changed files with 2,332 additions and 655 deletions.
diff --git a/config/env/base.yaml b/config/env/base.yaml
@@ -1,29 +1,17 @@
 _target_: gflownet.envs.base.GFlowNetEnv
 
-# Reward function: power or boltzmann
-# boltzmann: exp(-1.0 * reward_beta * proxy)
-# power: (-1.0 * proxy / reward_norm) ** self.reward_beta
-# identity: proxy
-reward_func: identity
-# Minimum reward
-reward_min: 1e-8
-# Beta parameter of the reward function
-reward_beta: 1.0
-# Reward normalization for "power" reward function
-reward_norm: 1.0
-# If > 0, reward_norm = reward_norm_std_mult * std(energies)
-reward_norm_std_mult: 0.0
-proxy_state_format: oracle
+env_id: "env"
+# Policy distribution parameters
+fixed_distr_params: null
+random_distr_params: null
 # Check if action valid with mask before step
 skip_mask_check: False
 # Whether the environment has conditioning variables
 conditional: False
 # Whether the environment is continuous
 continuous: False
-# Buffer
+# Buffer: no train and test buffers by default
 buffer:
   replay_capacity: 0
-  train:
-    path: null
-  test:
-    path: null
+  train: null
+  test: null
diff --git a/config/env/tetris.yaml b/config/env/tetris.yaml
@@ -17,5 +17,9 @@ allow_eos_before_full: False
 buffer:
   data_path: null
   train: null
-  test: null
+  test:
+    type: random
+    n: 10
+    output_csv: tetris_test.csv
+    output_pkl: tetris_test.pkl
 
diff --git a/config/experiments/ccube/corners.yaml b/config/experiments/ccube/corners.yaml
@@ -28,7 +28,6 @@ env:
     beta_beta: 10.0
     bernoulli_eos_prob: 0.1
     bernoulli_bts_prob: 0.1
-  reward_func: identity
 
 # GFlowNet hyperparameters
 gflownet:

diff --git a/config/experiments/ccube/uniform.yaml b/config/experiments/ccube/uniform.yaml
@@ -28,7 +28,6 @@ env:
     beta_beta: 10.0
     bernoulli_eos_prob: 0.1
     bernoulli_bts_prob: 0.1
-  reward_func: identity
 
 # GFlowNet hyperparameters
 gflownet:

diff --git a/config/experiments/clatticeparams/clatticeparams_owl.yaml b/config/experiments/clatticeparams/clatticeparams_owl.yaml
@@ -33,7 +33,6 @@ env:
     beta_beta: 0.01
     bernoulli_source_logit: 1.0
     bernoulli_eos_logit: 1.0
-  reward_func: identity
 
 # GFlowNet hyperparameters
 gflownet:

diff --git a/config/experiments/crystals/albatross.yaml b/config/experiments/crystals/albatross.yaml
@@ -42,8 +42,6 @@ env:
       beta_beta: 10.0
       bernoulli_eos_prob: 0.1
       bernoulli_bts_prob: 0.1
-  reward_func: boltzmann
-  reward_beta: 8
   buffer:
     replay_capacity: 0
     test:
@@ -52,6 +50,11 @@ env:
       output_csv: ccrystal_val.csv
       output_pkl: ccrystal_val.pkl
 
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 8
+
 # GFlowNet hyperparameters
 gflownet:
   random_action_prob: 0.1

diff --git a/config/experiments/crystals/albatross_sg_first.yaml b/config/experiments/crystals/albatross_sg_first.yaml
@@ -44,8 +44,6 @@ env:
       beta_beta: 10.0
       bernoulli_eos_prob: 0.1
       bernoulli_bts_prob: 0.1
-  reward_func: boltzmann
-  reward_beta: 8
   buffer:
     replay_capacity: 0
     test:
@@ -54,6 +52,11 @@ env:
       output_csv: ccrystal_val.csv
       output_pkl: ccrystal_val.pkl
 
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 8
+
 # GFlowNet hyperparameters
 gflownet:
   random_action_prob: 0.1

diff --git a/config/experiments/crystals/lattice_parameters.yaml b/config/experiments/crystals/lattice_parameters.yaml
@@ -9,11 +9,14 @@ defaults:
 
 # Environment
 env:
-  reward_func: boltzmann
-  reward_beta: 0.3
   buffer:
     replay_capacity: 1000
 
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 0.3
+
 # GFlowNet hyperparameters
 gflownet:
   random_action_prob: 0.1

diff --git a/config/experiments/crystals/pigeon.yaml b/config/experiments/crystals/pigeon.yaml
@@ -46,8 +46,6 @@ env:
       beta_beta: 10.0
       bernoulli_eos_prob: 0.1
       bernoulli_bts_prob: 0.1
-  reward_func: boltzmann
-  reward_beta: 8
   buffer:
     replay_capacity: 0
     test:
@@ -56,6 +54,11 @@ env:
       output_csv: ccrystal_val.csv
       output_pkl: ccrystal_val.pkl
 
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 8
+
 # GFlowNet hyperparameters
 gflownet:
   random_action_prob: 0.1

diff --git a/config/experiments/icml23/ctorus.yaml b/config/experiments/icml23/ctorus.yaml
@@ -12,7 +12,6 @@ env:
   n_dim: 2
   length_traj: 10
   n_comp: 5
-  reward_func: identity
   vonmises_min_concentration: 4
   policy_encoding_dim_per_angle: 10
 

diff --git a/config/experiments/icml23/dtorus.yaml b/config/experiments/icml23/dtorus.yaml
@@ -12,7 +12,6 @@ env:
   n_dim: 2
   n_angles: 20
   length_traj: 20
-  reward_func: identity
 
 # Proxy
 proxy:

diff --git a/config/experiments/icml23/htorus.dryrun.yaml b/config/experiments/icml23/htorus.dryrun.yaml
@@ -11,7 +11,6 @@ defaults:
 env:
   n_dim: 2
   length_traj: 5
-  reward_func: identity
 
 # Proxy
 proxy:

diff --git a/config/experiments/icml23/htorus.yaml b/config/experiments/icml23/htorus.yaml
@@ -11,7 +11,6 @@ defaults:
 env:
   n_dim: 2
   length_traj: 20
-  reward_func: identity
 
 # Proxy
 proxy:

diff --git a/config/experiments/neurips23/crystal-comp-sg-lp.yaml b/config/experiments/neurips23/crystal-comp-sg-lp.yaml
@@ -16,8 +16,11 @@ env:
     grid_size: 10
   composition_kwargs:
     elements: 89
-  reward_func: boltzmann
-  reward_beta: 1
+
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 1
 
 # GFlowNet hyperparameters
 gflownet:

diff --git a/config/experiments/scrabble/jay.yaml b/config/experiments/scrabble/jay.yaml
@@ -20,7 +20,6 @@ env:
       n: 1000
       output_csv: scrabble_test.csv
       output_pkl: scrabble_test.pkl
-  reward_func: identity
 
 # Proxy
 proxy:

diff --git a/config/experiments/scrabble/penguin.yaml b/config/experiments/scrabble/penguin.yaml
@@ -20,7 +20,6 @@ env:
       n: 1000
       output_csv: scrabble_test.csv
       output_pkl: scrabble_test.pkl
-  reward_func: identity
 
 # Proxy
 proxy:

diff --git a/config/experiments/simple_tetris.yaml b/config/experiments/simple_tetris.yaml
@@ -8,8 +8,6 @@ defaults:
   - override /logger: wandb
 
 env:
-  reward_func: boltzmann
-  reward_beta: 10.0
   width: 4
   height: 4
   pieces: ["I", "O", "J", "L", "T"]
@@ -22,6 +20,10 @@ env:
       output_pkl: simple_tetris_val.pkl
       n: 100
 
+proxy:
+  reward_function: exponential
+  beta: 10
+
 gflownet:
   random_action_prob: 0.3
   optimizer:
@@ -44,4 +46,4 @@ device: cpu
 logger:
   do:
     online: True
-  project_name: simple_tetris
+  project_name: simple_tetris
diff --git a/config/experiments/tree.yaml b/config/experiments/tree.yaml
@@ -14,13 +14,16 @@ env:
   continuous: False
   policy_format: mlp
   threshold_components: 3
-  reward_func: boltzmann
-  reward_beta: 32
   test_args:
     top_k_trees: 100
   buffer:
     replay_capacity: 100
 
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 32
+
 # GFlowNet hyperparameters
 gflownet:
   random_action_prob: 0.1

diff --git a/config/experiments/workshop23/discrete-matbench.yaml b/config/experiments/workshop23/discrete-matbench.yaml
@@ -18,11 +18,14 @@ env:
     grid_size: 10
   composition_kwargs:
     elements: [1,3,4,5,6,7,8,9,11,12,13,14,15,16,17,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,89,90,91,92,93,94]
-  reward_func: boltzmann
-  reward_beta: 1
   buffer:
     replay_capacity: 0
 
+# Proxy
+proxy:
+  reward_function: exponential
+  beta: 1
+
 # GFlowNet hyperparameters
 gflownet:
   random_action_prob: 0.1

diff --git a/config/logger/base.yaml b/config/logger/base.yaml
@@ -26,6 +26,8 @@ test:
   logprobs_bootstrap_size: 10000
   # Maximum number of test data points to compute log likelihood probs.
   max_data_logprobs: 1e5
+  # Number of points to obtain a grid to estimate the reward density
+  n_grid: 40000
 # Oracle metrics
 oracle:
   period: 100000

diff --git a/config/proxy/base.yaml b/config/proxy/base.yaml
@@ -0,0 +1,24 @@
+_target_: gflownet.proxy.base.Proxy
+
+# Reward function: string identifier of the proxy-to-reward function:
+# - identity (default)
+# - absolute
+# - power
+# - exponential
+# - shift
+# - product
+# Alternatively, it can be a callable of the function itself.
+reward_function: identity
+# A callable of the proxy-to-logreward function. 
+# None by default, which takes the log of the proxy-to-reward function
+logreward_function: null
+# Arguments of the proxy-to-reward function (beta):
+# - power: R(x) = x ** beta
+# - exponential: R(x) = exp(x * beta)
+# - shift: R(x) = x + beta
+# - product: R(x) = x * beta
+reward_function_kwargs: {}
+# Minimum reward. Used to clip the rewards.
+reward_min: 0.0
+# Flag to control whether rewards are clipped
+do_clip_rewards: False
diff --git a/config/proxy/corners.yaml b/config/proxy/corners.yaml
@@ -0,0 +1,8 @@
+defaults:
+  - base
+
+_target_: gflownet.proxy.corners.Corners
+
+mu: 0.75
+sigma: 0.05
+higher_is_better: False
diff --git a/config/proxy/crystals/dave.yaml b/config/proxy/crystals/dave.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.crystals.dave.DAVE
 
 release: 0.3.4

diff --git a/config/proxy/crystals/lattice_parameters.yaml b/config/proxy/crystals/lattice_parameters.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.crystals.lattice_parameters.LatticeParameters
 
 min_value: -100
diff --git a/config/proxy/crystals/spacegroup.yaml b/config/proxy/crystals/spacegroup.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.crystals.spacegroup.SpaceGroup
 
 normalize: True
diff --git a/config/proxy/length.yaml b/config/proxy/length.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.aptamers.Aptamers
 
 oracle_id: length

diff --git a/config/proxy/molecule.yaml b/config/proxy/molecule.yaml
@@ -1,4 +1,7 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.molecule.RFMoleculeEnergy
 
 path_to_model: './data/random_forest_reward_100.pkl'
-url_to_model: 'https://drive.google.com/uc?id=1OpQNC8WWIsMh8K4olfSaQRFlj3emYThF'
+url_to_model: 'https://drive.google.com/uc?id=1OpQNC8WWIsMh8K4olfSaQRFlj3emYThF'
diff --git a/config/proxy/scrabble.yaml b/config/proxy/scrabble.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.scrabble.ScrabbleScorer
 
 vocabulary_check: False
diff --git a/config/proxy/tetris.yaml b/config/proxy/tetris.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.tetris.Tetris
 
 normalize: True

diff --git a/config/proxy/torus.yaml b/config/proxy/torus.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.torus.Torus
 
 normalize: True

diff --git a/config/proxy/tree.yaml b/config/proxy/tree.yaml
@@ -1,3 +1,6 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.tree.TreeProxy
 
 use_prior: False
diff --git a/config/proxy/uniform.yaml b/config/proxy/uniform.yaml
@@ -1 +1,4 @@
+defaults:
+  - base
+
 _target_: gflownet.proxy.uniform.Uniform
-Original file line number
+Diff line change
@@ Expand Up / @@ -11,7 +11,6 @@ defaults: @@
     env:
       n_dim: 2
       length_traj: 5
-      reward_func: identity
     # Proxy
     proxy:
@@ Expand Down @@