feedback

sbibm/algorithms/sbi/__init__.py

@@ -1,4 +1,5 @@
 from sbibm.algorithms.sbi.mcabc import run as mcabc
+from sbibm.algorithms.sbi.sl import run as sl
 from sbibm.algorithms.sbi.smcabc import run as smcabc
 from sbibm.algorithms.sbi.snle import run as snle
 from sbibm.algorithms.sbi.snpe import run as snpe

sbibm/algorithms/sbi/sl.py

@@ -1,5 +1,5 @@
 import logging
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, Optional
 
 import torch
 import torch.nn as nn
@@ -95,7 +95,7 @@ def run(
     mcmc_method: str = "slice_np",
     mcmc_parameters: Dict[str, Any] = {},
     diag_eps: float = 0.0,
-) -> Tuple[torch.Tensor, int, Optional[torch.Tensor]]:
+) -> tuple[torch.Tensor, int, Optional[torch.Tensor]]:
     """Runs (S)NLE from `sbi`
 
     Args:

sbibm/algorithms/sbi/snle.py

@@ -1,6 +1,6 @@
 import logging
 import math
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, Optional
 
 import torch
 from sbi import inference as inference
@@ -24,14 +24,13 @@ def run(
     neural_net: str = "maf",
     hidden_features: int = 50,
     simulation_batch_size: int = 1000,
-    # NOTE: why is the training batch size so high?
     training_batch_size: int = 10000,
     automatic_transforms_enabled: bool = True,
     mcmc_method: str = "slice_np_vectorized",
     mcmc_parameters: Dict[str, Any] = {
         "num_chains": 100,
         "thin": 10,
-        "warmup_steps": 100,
+        "warmup_steps": 25,
         "init_strategy": "sir",
         # NOTE: sir kwargs changed: num_candidate_samples = num_batches * batch_size
         "init_strategy_parameters": {
@@ -40,8 +39,8 @@ def run(
     },
     z_score_x: bool = True,
     z_score_theta: bool = True,
-    max_num_epochs: Optional[int] = 2**31 -1,
-) -> Tuple[torch.Tensor, int, Optional[torch.Tensor]]:
+    max_num_epochs: Optional[int] = 2**31 - 1,
+) -> tuple[torch.Tensor, int, Optional[torch.Tensor]]:
     """Runs (S)NLE from `sbi`
 
     Args:
@@ -109,8 +108,6 @@ def run(
 
     posteriors = []
     proposal = prior
-    # NOTE: why was this hard coded here?
-    # mcmc_parameters["warmup_steps"] = 25
 
     for r in range(num_rounds):
         theta, x = inference.simulate_for_sbi(

sbibm/algorithms/sbi/snpe.py

@@ -1,6 +1,6 @@
 import logging
 import math
-from typing import Optional, Tuple
+from typing import Optional
 
 import torch
 from sbi import inference as inference
@@ -29,8 +29,8 @@ def run(
     automatic_transforms_enabled: bool = False,
     z_score_x: bool = True,
     z_score_theta: bool = True,
-    max_num_epochs: Optional[int] = 2**31 -1,
-) -> Tuple[torch.Tensor, int, Optional[torch.Tensor]]:
+    max_num_epochs: Optional[int] = 2**31 - 1,
+) -> tuple[torch.Tensor, int, Optional[torch.Tensor]]:
     """Runs (S)NPE from `sbi`
 
     Args:

sbibm/algorithms/sbi/snre.py

@@ -1,6 +1,6 @@
 import logging
 import math
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, Optional
 
 import torch
 from sbi import inference as inference
@@ -31,7 +31,7 @@ def run(
     mcmc_parameters: Dict[str, Any] = {
         "num_chains": 100,
         "thin": 10,
-        "warmup_steps": 100,
+        "warmup_steps": 25,
         "init_strategy": "sir",
         # NOTE: sir kwargs changed: num_candidate_samples = num_batches * batch_size
         "init_strategy_parameters": {
@@ -41,8 +41,8 @@ def run(
     z_score_x: bool = True,
     z_score_theta: bool = True,
     variant: str = "B",
-    max_num_epochs: Optional[int] = 2**31 -1,
-) -> Tuple[torch.Tensor, int, Optional[torch.Tensor]]:
+    max_num_epochs: Optional[int] = 2**31 - 1,
+) -> tuple[torch.Tensor, int, Optional[torch.Tensor]]:
     """Runs (S)NRE from `sbi`
 
     Args:
@@ -118,7 +118,6 @@ def run(
 
     posteriors = []
     proposal = prior
-    # mcmc_parameters["warmup_steps"] = 25
 
     for r in range(num_rounds):
         theta, x = inference.simulate_for_sbi(

setup.py

@@ -29,7 +29,7 @@
     "pandas>=1.0.0",
     "pyabc>=0.10.8",
     "pyabcranger>=0.0.48",
-    "sbi==0.20.0",
+    "sbi>=0.20.0",
     "pyro-ppl",
     "scikit-learn",
     "torch>=1.8.0",

tests/algorithms/sbi/test_sbi_run_methods.py

@@ -1,12 +1,12 @@
 import pytest
 
 import sbibm
-from sbibm.algorithms.sbi import mcabc, smcabc, snle, snpe, snre
+from sbibm.algorithms.sbi import mcabc, smcabc, snle, snpe, snre, sl
 
 
 # a fast test
 @pytest.mark.parametrize(
-    "run_method", (mcabc, smcabc, snle, snpe, snre)
+    "run_method", (mcabc, smcabc, snle, snpe, snre, sl)
 )
 @pytest.mark.parametrize("task_name", ("gaussian_linear",))
 @pytest.mark.parametrize("num_observation", (1, 3))