Revert csv parsing empty data to return empty arrays

cmdstanpy/utils/stancsv.py

@@ -60,13 +60,13 @@ def csv_bytes_list_to_numpy(
     """Efficiently converts a list of bytes representing whose concatenation
     represents a CSV file into a numpy array. Includes header specifies
     whether the bytes contains an initial header line."""
+    if not csv_bytes_list:
+        return np.empty((0,))
+    num_cols = csv_bytes_list[0].count(b",") + 1
     try:
         import polars as pl
 
         try:
-            if not csv_bytes_list:
-                raise ValueError("No data found to parse")
-            num_cols = csv_bytes_list[0].count(b",") + 1
             out: npt.NDArray[np.float64] = (
                 pl.read_csv(
                     io.BytesIO(b"".join(csv_bytes_list)),
@@ -77,10 +77,8 @@ def csv_bytes_list_to_numpy(
                 .to_numpy()
                 .astype(np.float64)
             )
-            if out.shape[0] == 0:
-                raise ValueError("No data found to parse")
-        except pl.exceptions.NoDataError as exc:
-            raise ValueError("No data found to parse") from exc
+        except pl.exceptions.NoDataError:
+            return np.empty((0, num_cols))
     except ImportError:
         with warnings.catch_warnings():
             warnings.filterwarnings("ignore")
@@ -91,10 +89,11 @@ def csv_bytes_list_to_numpy(
                 dtype=np.float64,
                 ndmin=1,
             )
-        if out.shape == (0,):
-            raise ValueError("No data found to parse")  # pylint: disable=W0707
         if len(out.shape) == 1:
-            out = out.reshape(1, -1)
+            if out.shape[0] == 0:  # No data read
+                out = np.empty((0, num_cols))
+            else:
+                out = out.reshape(1, -1)
 
     return out
 

test/test_sample.py

@@ -2138,3 +2138,13 @@ def test_sample_dense_mass_matrix():
     fit = linear_model.sample(data=jdata, metric="dense_e", chains=2)
     assert fit.metric is not None
     assert fit.metric.shape == (2, 3, 3)
+
+
+def test_no_output_draws():
+    stan = os.path.join(DATAFILES_PATH, 'bernoulli.stan')
+    model = cmdstanpy.CmdStanModel(stan_file=stan)
+    data = os.path.join(DATAFILES_PATH, 'bernoulli.data.json')
+
+    mcmc = model.sample(data=data, iter_sampling=0, save_warmup=False, chains=2)
+    draws = mcmc.draws()
+    assert np.array_equal(draws, np.empty((0, 2, len(mcmc.column_names))))

test/test_stancsv.py

@@ -136,17 +136,10 @@ def test_csv_bytes_to_numpy_with_header_no_polars():
         assert np.array_equal(arr_out, expected)
 
 
-def test_csv_bytes_to_numpy_empty():
-    lines = [b""]
-    with pytest.raises(ValueError):
-        stancsv.csv_bytes_list_to_numpy(lines)
-
-
-def test_csv_bytes_to_numpy_empty_no_polars():
-    lines = [b""]
-    with without_import("polars", cmdstanpy.utils.stancsv):
-        with pytest.raises(ValueError):
-            stancsv.csv_bytes_list_to_numpy(lines)
+def test_csv_bytes_empty():
+    lines = []
+    arr = stancsv.csv_bytes_list_to_numpy(lines)
+    assert np.array_equal(arr, np.empty((0,)))
 
 
 def test_csv_bytes_to_numpy_header_no_draws():
@@ -156,8 +149,8 @@ def test_csv_bytes_to_numpy_header_no_draws():
             b"n_leapfrog__,divergent__,energy__,theta\n"
         ),
     ]
-    with pytest.raises(ValueError):
-        stancsv.csv_bytes_list_to_numpy(lines)
+    arr = stancsv.csv_bytes_list_to_numpy(lines)
+    assert arr.shape == (0, 8)
 
 
 def test_csv_bytes_to_numpy_header_no_draws_no_polars():
@@ -168,8 +161,8 @@ def test_csv_bytes_to_numpy_header_no_draws_no_polars():
         ),
     ]
     with without_import("polars", cmdstanpy.utils.stancsv):
-        with pytest.raises(ValueError):
-            stancsv.csv_bytes_list_to_numpy(lines)
+        arr = stancsv.csv_bytes_list_to_numpy(lines)
+        assert arr.shape == (0, 8)
 
 
 def test_parse_comments_and_draws():