benwtrent commented on issue #12342:
URL: https://github.com/apache/lucene/issues/12342#issuecomment-1668237698
Here are my results for the en-ja mixed dataset.
max inner product in baseline, euclidean in transformed
# EN-Ja rerun
Reversed | Ordered | Random
:-------------------------:|:-------------------------:|:-------------------------:
||
<details>
<summary>Updated knnPerfTest.py </summary>
My testing run I do the following:
- Run all files & tests with just `fanout:0` but with `-reindex` passed as
an arg. This builds the indices
- Then I remove the `-reindex` the param and run with all the fanout
parameters.
I found this to be the quickest way to test.
```python
i#!/usr/bin/env/python
import subprocess
import benchUtil
import constants
LUCENE_CHECKOUT = 'lucene_candidate'
# test parameters. This script will run KnnGraphTester on every combination
of these parameters
VALUES = {
'ndoc': (100000,),
'maxConn': (48, ),
'beamWidthIndex': (200,),
'fanout': (0, 10, 50, 90, 190, 490, 590, 990),
'topK': (10,),
}
def advance(ix, values):
for i in reversed(range(len(ix))):
param = list(values.keys())[i]
#print("advance " + param)
if ix[i] == len(values[param]) - 1:
ix[i] = 0
else:
ix[i] += 1
return True
return False
def run_knn_benchmark(checkout, values, training_file, testing_file, dims,
metric):
indexes = [0] * len(values.keys())
indexes[-1] = -1
args = []
print(f"\n\n\nNow running{training_file}\n\n\n")
dim = dims #768
doc_vectors = training_file
query_vectors = testing_file
cp = benchUtil.classPathToString(benchUtil.getClassPath(checkout))
JAVA_EXE =
'/Users/benjamintrent/Library/Java/JavaVirtualMachines/jdk-20.0.1.jdk/Contents/Home/bin/java'
cmd = [JAVA_EXE,
'-cp', cp,
'--add-modules', 'jdk.incubator.vector',
'-Dorg.apache.lucene.store.MMapDirectory.enableMemorySegments=false',
'KnnGraphTester']
print("recall\tlatency\tnDoc\tfanout\tmaxConn\tbeamWidth\tvisited\tindex
ms")
while advance(indexes, values):
pv = {}
args = []
for (i, p) in enumerate(list(values.keys())):
if p in values:
if values[p]:
value = values[p][indexes[i]]
pv[p] = value
else:
args += ['-' + p]
args += [a for (k, v) in pv.items() for a in ('-' + k, str(v)) if a]
this_cmd = cmd + args + [
'-dim', str(dim),
'-docs', doc_vectors,
#'-reindex',
'-metric', metric,
'-search', query_vectors,
'-forceMerge',
'-quiet',
]
subprocess.run(this_cmd)
tests = [
('%s/util/en_ja.random.train' % constants.BASE_DIR, '%s/util/en_ja.test'
% constants.BASE_DIR, 768, "angular"),
('%s/util/en_ja.ordered.train' % constants.BASE_DIR,
'%s/util/en_ja.test' % constants.BASE_DIR, 768, "angular"),
('%s/util/en_ja.reversed.train' % constants.BASE_DIR,
'%s/util/en_ja.test' % constants.BASE_DIR, 768, "angular"),
('%s/util/en_ja.random-transform.train' % constants.BASE_DIR,
'%s/util/en_ja-transform.test' % constants.BASE_DIR, 769, "euclidean"),
('%s/util/en_ja.ordered-transform.train' % constants.BASE_DIR,
'%s/util/en_ja-transform.test' % constants.BASE_DIR, 769, "euclidean"),
('%s/util/en_ja.reversed-transform.train' % constants.BASE_DIR,
'%s/util/en_ja-transform.test' % constants.BASE_DIR, 769, "euclidean"),
]
for (training_file, testing_file, dims, metric) in tests:
run_knn_benchmark(LUCENE_CHECKOUT, VALUES, training_file, testing_file,
dims, metric)
```
</details>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
