fix benchmark

lancedb · Sep 7, 2023 · e9d12b1 · e9d12b1
1 parent f1f651a
commit e9d12b1
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 6 deletions.
diff --git a/python/DEVELOPMENT.md b/python/DEVELOPMENT.md
@@ -110,6 +110,10 @@ Note the parameter `--benchmark-min-time`: this controls how many seconds to run
 the benchmark in each round (default 5 rounds). The default is very low but you
 can increase this so that the profile gets more samples.
 
+You can drop the `--no-inline` to have the program try to identify which functions
+were inlined to get more detail, though this will make the processing take
+considerably longer.
+
 This will only work on Linux.
 
 Note that you'll want to run the benchmarks once prior to profiling, so that

diff --git a/python/python/benchmarks/test_search.py b/python/python/benchmarks/test_search.py
@@ -46,12 +46,13 @@ def test_dataset(data_dir: Path) -> lance.LanceDataset:
 
 @pytest.mark.benchmark(group="query_ann")
 def test_knn_search(test_dataset, benchmark):
-    q = pa.FixedSizeListArray.from_arrays(pc.random(N_DIMS).cast(pa.float32()), N_DIMS)
+    q = pc.random(N_DIMS).cast(pa.float32())
     result = benchmark(
         test_dataset.to_table,
         nearest=dict(
+            column="vector",
             q=q,
-            k=10,
+            k=100,
             nprobes=10,
             use_index=False,
         ),
@@ -61,12 +62,13 @@ def test_knn_search(test_dataset, benchmark):
 
 @pytest.mark.benchmark(group="query_ann")
 def test_flat_index_search(test_dataset, benchmark):
-    q = pa.FixedSizeListArray.from_arrays(pc.random(N_DIMS).cast(pa.float32()), N_DIMS)
+    q = pc.random(N_DIMS).cast(pa.float32())
     result = benchmark(
         test_dataset.to_table,
         nearest=dict(
+            column="vector",
             q=q,
-            k=10,
+            k=100,
             nprobes=10,
         ),
     )
@@ -75,12 +77,13 @@ def test_flat_index_search(test_dataset, benchmark):
 
 @pytest.mark.benchmark(group="query_ann")
 def test_ivf_pq_index_search(test_dataset, benchmark):
-    q = pa.FixedSizeListArray.from_arrays(pc.random(N_DIMS).cast(pa.float32()), N_DIMS)
+    q = pc.random(N_DIMS).cast(pa.float32())
     result = benchmark(
         test_dataset.to_table,
         nearest=dict(
+            column="vector",
             q=q,
-            k=10,
+            k=100,
             nprobes=10,
             refine_factor=2,
         ),