Update config

Author: stefanradev93

.github/workflows/tests.yml

@@ -10,7 +10,7 @@ jobs:
     strategy:
       matrix:
         os: [ubuntu-latest, windows-latest]
-        python-version: ['3.8']
+        python-version: ['3.9']
 
     steps:
     - uses: actions/checkout@v2

bayesflow/amortizers.py

@@ -513,6 +513,10 @@ def sample(self, input_dict, n_samples, to_numpy=True, **kwargs):
         # Obtain random draws from the surrogate likelihood given conditioning variables
         lik_samples = self.surrogate_net.inverse(z_samples, conditions, training=False, **kwargs)
 
+        # Only return 2D array, if first dimensions is 1
+        if lik_samples.shape[0] == 1:
+            lik_samples = lik_samples[0]
+
         if to_numpy:
             return lik_samples.numpy()
         return lik_samples

environment.yaml

@@ -2,7 +2,7 @@
 
 name: bf
 dependencies:
-  - python==3.8
+  - python==3.9
   - pip
   - pip:
     - -r requirements.txt

requirements.txt

@@ -1,6 +1,6 @@
 h5py==3.5
 numba===0.55
-numpy==1.19
+numpy==1.21
 pandas==1.4
 scikit-learn==1.1
 scipy==1.9
@@ -13,4 +13,4 @@ matplotlib==3.5
 m2r2==0.3
 ipykernel==6.7
 nbsphinx==0.8
-ipywidgets==7.6.5
+ipywidgets==7.6.5

setup.cfg

@@ -30,7 +30,7 @@ keywords =
     model comparison
 
 [options]
-python_requires = >=3.8
+python_requires = >=3.9
 packages = find:
 install_requires =
     h5py >= 3.6

tests/test_amortizers.py

@@ -121,7 +121,7 @@ def test_amortized_posterior(cond_shape, summary_loss, soft):
     # Log posterior and lpdf should be the same, unless using softflow
     # which will introduce some noise in the untrained version
     if not soft:
-        assert np.allclose(log_post, lpdf, atol=1e-6)
+        assert np.allclose(log_post, lpdf, atol=1e-5)
 
     # Test shape of samples
     if batch_size == 1:
@@ -223,7 +223,7 @@ def test_amortized_likelihood(inp_shape, soft):
     # Log posterior and lpdf should be the same, unless using softflow
     # which will introduce some noise in the untrained version
     if not soft:
-        assert np.allclose(log_lik, lpdf, atol=1e-6)
+        assert np.allclose(log_lik, lpdf, atol=1e-5)
 
     # Test shape of samples
     if batch_size == 1:

tests/test_coupling_networks.py

@@ -90,7 +90,7 @@ def test_coupling_layer(condition, spec_norm, use_perm, use_act_norm, input_shap
     else:
         assert network.act_norm is None
     # Test invertibility
-    assert np.allclose(inp, inp_rec, atol=1e-6)
+    assert np.allclose(inp, inp_rec, atol=1e-5)
     # Test shapes (bijectivity)
     assert z.shape == inp.shape
     if input_shape == '2d':

tests/test_helper_networks.py

@@ -138,7 +138,7 @@ def test_invariant_module(n_dense_s1, n_dense_s2, output_dim):
     out_perm = inv_module(x_perm).numpy()
 
     # Assert outputs equal
-    assert np.allclose(out, out_perm, atol=1e-6)
+    assert np.allclose(out, out_perm, atol=1e-5)
     # Assert shape 2d
     assert len(out.shape) == 2 and len(out_perm.shape) == 2
     # Assert first and last dimension equals output dimension
@@ -172,7 +172,7 @@ def test_equivariant_module(n_dense_s3, output_dim):
     out_perm = equiv_module(x_perm).numpy()
 
     # Assert outputs equal
-    assert np.allclose(out[:, perm, :], out_perm, atol=1e-6)
+    assert np.allclose(out[:, perm, :], out_perm, atol=1e-5)
     # Assert shape 3d
     assert len(out.shape) == 3 and len(out_perm.shape) == 3
     # Assert first and last dimension equals output dimension

tests/test_inference_networks.py

@@ -97,7 +97,7 @@ def test_invertible_network(input_shape, condition, use_act_norm, use_soft_flow,
         assert network.soft_flow is False
     # Test invertibility (in case no soft flow)
     if not use_soft_flow:
-        assert np.allclose(inp, inp_rec, atol=1e-6)
+        assert np.allclose(inp, inp_rec, atol=1e-5)
     # Test shapes (bijectivity)
     assert z.shape == inp.shape
     assert z.shape[-1] == input_dim

tests/test_summary_networks.py

@@ -65,7 +65,7 @@ def test_invariant_network(num_equiv, summary_dim):
     # Assert numebr of equivariant layers correct
     assert len(inv_net.equiv_layers.layers) == num_equiv
     # Assert outputs equal
-    assert np.allclose(out, out_perm, atol=1e-6)
+    assert np.allclose(out, out_perm, atol=1e-5)
     # Assert shape 2d
     assert len(out.shape) == 2 and len(out_perm.shape) == 2
     # Assert batch and last dimension equals output dimension