refactor: improve warning clarity and formatting

Author: janfb

sbi/inference/posteriors/mcmc_posterior.py

@@ -426,9 +426,9 @@ def sample_batched(
         # warn if num_chains is larger than num requested samples
         if num_chains > torch.Size(sample_shape).numel():
             warnings.warn(
-                f"""Passed num_chains {num_chains} is larger than the number of
-                requested samples {torch.Size(sample_shape).numel()}, resetting
-                it to {torch.Size(sample_shape).numel()}.""",
+                "The passed number of MCMC chains is larger than the number of "
+                f"requested samples: {num_chains} > {torch.Size(sample_shape).numel()},"
+                f" resetting it to {torch.Size(sample_shape).numel()}.",
                 stacklevel=2,
             )
             num_chains = torch.Size(sample_shape).numel()
@@ -453,12 +453,11 @@ def sample_batched(
         num_chains_extended = batch_size * num_chains
         if num_chains_extended > 100:
             warnings.warn(
-                f"""Note that for batched sampling, we use {num_chains} for each
-                 x in the batch. With the given settings, this results in a
-                 large number of chains ({num_chains_extended}),  This can be
-                 large number of chains ({num_chains_extended}), which can be
-                 slow and memory-intensive. Consider reducing the number of
-                 chains.""",
+                "Note that for batched sampling, we use num_chains many chains for each"
+                " x in the batch. With the given settings, this results in a large "
+                f"number large number of chains ({num_chains_extended}), which can be "
+                "slow and memory-intensive for vectorized MCMC. Consider reducing the "
+                "number of chains.",
                 stacklevel=2,
             )
         init_strategy_parameters["num_return_samples"] = num_chains_extended

sbi/utils/sbiutils.py

@@ -406,11 +406,11 @@ def warn_on_batched_x(batch_size):
     if batch_size > 1:
         warnings.warn(
             f"An x with a batch size of {batch_size} was passed. "
-            + """Unless you are using `sample_batched` or `log_prob_batched`, this will
-            be interpreted as a batch of independent and identically distributed data
-            X={x_1, ..., x_n}, i.e., data generated based on the same underlying
-            (unknown) parameter. The resulting posterior will be with respect to entire
-            batch, i.e,. p(theta | X).""",
+            "Unless you are using `sample_batched` or `log_prob_batched`, this will"
+            "be interpreted as a batch of independent and identically distributed data"
+            "X={x_1, ..., x_n}, i.e., data generated based on the same underlying"
+            "(unknown) parameter. The resulting posterior will be with respect to"
+            " the entire batch, i.e,. p(theta | X).",
             stacklevel=2,
         )