Comments (6)
JoanFM
commented on May 27, 2024
Hello @tommykoctur , the docarray version used was the same?
from annlite.
tommykoctur
commented on May 27, 2024
hi,
yes for both tests they used docarray==0.21.0
from annlite.
numb3r3
commented on May 27, 2024
@tommykoctur Thanks for your reports. We will perform some experiments to identify the cause, and try to fix this issue ASAP.
from annlite.
numb3r3
commented on May 27, 2024
Storing 1 document (with embedding d 768) took 0.35s vs 1.15s.
@tommykoctur What's the index size in your comparison? The insert performance is very sensitive with the indexer built already. Inserting a new point in a large indexer usually takes a longer time.
from annlite.
jemmyshin
commented on May 27, 2024
from annlite import AnnLite
from docarray import Document, DocumentArray
import numpy as np
import time
import shutil
N = 10000
REPEAT = 10
# directly using Annlite
time_elapse_list = []
indexer = AnnLite(n_dim=768, data_path='./data')
for _ in range(REPEAT):
da = DocumentArray()
for i in range(N):
da.extend([Document(id=str(i), embedding=np.random.randn(768))])
indexer.index(da)
time.sleep(1) # wait for indexing
da = DocumentArray([Document(embedding=np.random.randn(768))])
start_time = time.time()
indexer.index(da)
print(indexer.stat) # make sure that the insertion happends, you are supposed to see index_size=10001 and total_docs=10001
time_elapse_list.append(time.time() - start_time)
indexer.clear()
time.sleep(1)
print(f"===> insert using Annlite: {sum(time_elapse_list) / len(time_elapse_list)}")
indexer.close()
shutil.rmtree('./data')
# using Annlite in docarray
time_elapse_list = []
indexer = DocumentArray(storage='annlite',
config={'n_dim': 768, 'data_path': './data'})
for _ in range(REPEAT):
with indexer:
da = DocumentArray()
for i in range(N):
da.extend([Document(id=str(i), embedding=np.random.randn(768))])
indexer.extend(da)
time.sleep(1)
d = DocumentArray([Document(embedding=np.random.randn(768))])
start_time = time.time()
indexer.extend(d)
print(indexer._annlite.stat)
time_elapse_list.append(time.time() - start_time)
indexer.clear()
time.sleep(1)
print(f"===> insert using docarray: {sum(time_elapse_list) / len(time_elapse_list)}")
shutil.rmtree('./data')can you try this script and print out the time elape? @tommykoctur
I run this script on my local machine using annlite==0.5.4 and 0.5.8, actually 0.5.8 is even faster than 0.5.4 since we move part of index logic into memory instead of disk(10ms for 0.5.4 and 8ms for 0.5.8). Could you double-check your index size? The time for insertion will have huge difference when index size is different.
from annlite.
tommykoctur
commented on May 27, 2024
Hi,
your script doesn't seems problematic. I will investigate more why this is happening to our project. Thank you for now.
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
===> insert using Annlite: 0.12578973770141602
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
{'total_docs': 10001, 'index_size': 10001, 'n_cells': 1, 'n_dim': 768, 'n_components': None, 'metric': 'COSINE', 'is_trained': True}
===> insert using docarray: 0.13696033954620362```
from annlite.
Related Issues (20)
- Support for 16 bit quantization HOT 2
- Support Lucene backend via PyLucene HOT 1
- fix: links to documentation are broken HOT 2
- RuntimeError: wrong dimensionality of the vectors HOT 5
- RuntimeError: cannot return results
- add dump/backup endpoints
- Support for Mac with Apple Silicon HOT 1
- Can annlite be accelerated? HOT 4
- AttributeError: 'builtins.WriteOptions' object has no attribute 'set_sync' HOT 2
- annlite wrong filter name bug HOT 1
- Delete in executor does not works HOT 11
- Update does not work in annlite executor HOT 31
- Link missing in README.md HOT 2
- (bug)ID mismatch between hnsw and sqlite HOT 1
- ImportError in tests directory HOT 2
- 支持gpu? HOT 1
- Annliteindexer results change every bootup within a jina flow HOT 9
- AttributeError: 'builtins.WriteOptions' object has no attribute 'set_sync' HOT 1
- snapshot's index_hash has wrong value when deleting only HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from annlite.