Removed depcrated simformer unit test in posterior_nn tests

Author: nMaax

tests/posterior_nn_test.py

@@ -342,94 +342,6 @@ def test_batched_score_sample_with_different_x(
         )
 
 
-@pytest.mark.slow
-@pytest.mark.parametrize("x_o_batch_dim", (0, 1, 2))
-@pytest.mark.parametrize("sampling_method", ["sde", "ode"])
-@pytest.mark.parametrize(
-    "sample_shape",
-    (
-        (5,),  # less than num_chains
-        (4, 2),  # 2D batch
-    ),
-)
-def test_batched_score_simformer_sample_with_different_x(
-    x_o_batch_dim: bool,
-    sampling_method: str,
-    sample_shape: torch.Size,
-):
-    num_dim = 2
-    num_simulations = 100
-
-    prior = MultivariateNormal(loc=zeros(num_dim), covariance_matrix=eye(num_dim))
-    simulator = diagonal_linear_gaussian
-
-    inference = Simformer(prior=prior)
-
-    thetas = prior.sample((num_simulations,))
-    xs = simulator(thetas)
-    inputs = torch.stack([thetas, xs], dim=1)
-
-    inference.append_simulations(
-        inputs=inputs,
-    ).train(max_num_epochs=100)
-
-    x_o = ones(num_dim) if x_o_batch_dim == 0 else ones(x_o_batch_dim, num_dim)
-
-    # Build conditional for the specific task: infer theta (node 0) given x (node 1).
-    inference_condition_mask = torch.tensor([False, True])
-
-    posterior = inference.build_conditional(
-        condition_mask=inference_condition_mask,
-        sample_with=sampling_method,  # type: ignore
-    )
-
-    samples = posterior.sample_batched(
-        sample_shape,
-        x_o,
-    )
-
-    assert (
-        samples.shape == (*sample_shape, x_o_batch_dim, num_dim)
-        if x_o_batch_dim > 0
-        else (*sample_shape, num_dim)
-    ), "Sample shape wrong"
-
-    # test only for 1 sample_shape case to avoid repeating this test.
-    if x_o_batch_dim > 1 and sample_shape == (5,):
-        assert samples.shape[1] == x_o_batch_dim, "Batch dimension wrong"
-        inference = Simformer(prior=prior)
-
-        inference.append_simulations(
-            inputs=inputs,
-        ).train(max_num_epochs=100)
-
-        inference_condition_mask = torch.tensor([False, True])
-
-        posterior = inference.build_conditional(
-            condition_mask=inference_condition_mask,
-            sample_with=sampling_method,  # type: ignore
-        )
-
-        x_o = torch.stack([0.5 * ones(num_dim), -0.5 * ones(num_dim)], dim=0)
-        # test with multiple chains to test whether correct chains are
-        # concatenated.
-        sample_shape = torch.Size([1000])  # use enough samples for accuracy comparison
-        samples = posterior.sample_batched(sample_shape, x_o)
-
-        samples_separate1 = posterior.sample(sample_shape, x_o[0])
-        samples_separate2 = posterior.sample(sample_shape, x_o[1])
-
-        # Check if means are approx. same
-        samples_m = torch.mean(samples, dim=0, dtype=torch.float32)
-        samples_separate1_m = torch.mean(samples_separate1, dim=0, dtype=torch.float32)
-        samples_separate2_m = torch.mean(samples_separate2, dim=0, dtype=torch.float32)
-        samples_sep_m = torch.stack([samples_separate1_m, samples_separate2_m], dim=0)
-
-        assert torch.allclose(samples_m, samples_sep_m, atol=0.2, rtol=0.2), (
-            "Batched sampling is not consistent with separate sampling."
-        )
-
-
 @pytest.mark.slow
 @pytest.mark.parametrize("density_estimator", ["mdn", "maf", "zuko_nsf"])
 def test_batched_sampling_and_logprob_accuracy(density_estimator: str):