Weaviate embeddings rework #77

DrejcPesjak · 2024-02-07T01:02:49Z

This rework contains:

added Weaviate vector database support
added an abstract class VectorDB_API
moved from required uuid as file name, to storing image file paths alongside in the vector DB
removed pytorch dependencies from the library (still present in examples)
reworked the examples jupyter notebooks

Note: currently supports only cosine similarity as a distance score, as it is nicely limited between 0 and 1.

…mb. processing methods

kozlov721 · 2024-02-07T02:33:54Z

examples/utils/torch_utils.py

+    return model
+
+
+def load_model_resnet50() -> nn.Module:


Might be nicer to merge these two functions together:

def load_model_resnet50(discard_last_layer: bool): ...

conorsim · 2024-02-07T03:56:26Z

I'll review later but it would be nice to add type hints to all the function definitions - I noticed them occasionally but not consistently.

klemen1999 · 2024-02-07T10:10:06Z

luxonis_ml/embeddings/methods/duplicate.py

@@ -83,7 +78,7 @@ def _plot_kde(xs, s, density, maxima, minima):
    plt.show()


-def kde_peaks(data, bandwidth="scott", plot=False):
+def kde_peaks(data: np.ndarray, bandwidth: Union[str, float] = "scott", plot: bool = False) -> tuple[np.ndarray, np.ndarray, int, float]:


AFAIK luxonis-ml has python3.8>= requirement and for python3.8 tuple type annotation should be switched with Tuple or it will throw an error

In addition to this, it would be nice to create at least a very dummy test case for each added module that at least imports everything, so it catches errors like this.

conorsim

Thanks - I like the processing example!

Requesting minor change due to a typing error which breaks the import in the notebooks. And then not sure if my Weaviate issue is unique to me or not.

conorsim · 2024-02-07T16:48:09Z

luxonis_ml/embeddings/methods/duplicate.py

-    plot=False,
-):
+def find_similar(
+    reference_embeddings: Union[str, List[str], List[List[float], np.ndarray]],


Suggested change

reference_embeddings: Union[str, List[str], List[List[float], np.ndarray]],

reference_embeddings: Union[str, List[str], List[List[float]], np.ndarray]

I'm not sure if this should be the type or not, but the current typing throws an error at import in the example notebook.

conorsim · 2024-02-07T16:49:26Z

examples/Embeddings_LDF_Weaviate_Example.ipynb

I get this error once I get to the weaviate section of the notebook. Not sure if it would be a simple fix or that something could be wrong in my environment. If there is additional weaviate setup besides the embeddings requirements we should try to document it somewhere.

WeaviateStartUpError: Weaviate did not start up in 5 seconds. Either the Weaviate URL http://localhost:8080/ is wrong or Weaviate did not start up in the interval given in 'startup_period'.

conorsim · 2024-02-07T16:53:55Z

luxonis_ml/embeddings/utils/embedding.py

+
+    return img_transposed
+
+def get_image_tensors_from_LFS(


(Minor): Maybe get_image_tensors_from_remote or even get_image_tensors_from_lfs would be a better function name

kozlov721 · 2024-02-07T17:14:28Z

luxonis_ml/embeddings/methods/duplicate.py

-within a given set of embeddings. The removal process uses Kernel Density
-Estimation (KDE) on embeddings cosine similarity for optimal split, making
-this approach particularly suited for embeddings in high dimensional spaces.
+**Overview:**


Use B{Overview:} for bold text instead of asterisks, see epytext documentation for more details.

kozlov721 · 2024-02-07T17:15:54Z

luxonis_ml/embeddings/methods/duplicate.py


+- For Weaviate-specific examples, refer to the provided code examples.
+- The `search_vectordb` function can be used with either Qdrant or Weaviate, depending on the provided `vectordb_api` object.


Similarly, C{search_vectordb} instead of backticks.

DrejcPesjak · 2024-02-10T10:12:24Z

Pre-commit hooks passed, but pre-commit CI test still fails

Ran the pre-commits as described here. And got the following output when commiting:

git commit -m "Pre-commit hook changes"

ruff.....................................................................Passed
ruff-format..............................................................Passed
docformatter.............................................................Passed
don't commit to branch...................................................Passed
mdformat.............................................(no files to check)Skipped
[weaviate_emb 25427a1] Pre-commit hook changes
5 files changed, 267 insertions(+), 235 deletions(-)

But the CI/ call-reusable-workflow/pre-commit (pull_request) test still fails.

kozlov721 · 2024-02-12T17:47:31Z

The pre-commit command does not pass on my end with the same errors as the ones in the CI. I see where's the issue probably and will update the CONTRIBUTING.md. When commiting it only runs on the commited files, which might pass, but it doesn't mean it passes on all files. This would be caused by previously pushing without the pre-commit checks, so the problematic files are already upstream and no longer checked when git commit is ran. Run pre-commit run --all-files to get the errors on all files.

github-actions · 2024-02-14T01:49:36Z

Test Results

17 tests 17 ✅ 0s ⏱️
1 suites 0 💤
1 files 0 ❌

Results for commit 7b63170.

♻️ This comment has been updated with latest results.

github-actions · 2024-02-14T01:49:40Z

☂️ Python Coverage

current status: ✅

Overall Coverage

Lines	Covered	Coverage	Threshold	Status
3247	284	9%	0%	🟢

New Files

File	Coverage	Status
luxonis_ml/embeddings/utils/vectordb.py	0%	🟢
luxonis_ml/embeddings/utils/weaviate.py	0%	🟢
TOTAL	0%	🟢

Modified Files

File	Coverage	Status
luxonis_ml/embeddings/init.py	0%	🟢
luxonis_ml/embeddings/methods/OOD.py	0%	🟢
luxonis_ml/embeddings/methods/init.py	0%	🟢
luxonis_ml/embeddings/methods/duplicate.py	0%	🟢
luxonis_ml/embeddings/methods/mistakes.py	0%	🟢
luxonis_ml/embeddings/methods/representative.py	0%	🟢
luxonis_ml/embeddings/utils/init.py	0%	🟢
luxonis_ml/embeddings/utils/embedding.py	0%	🟢
luxonis_ml/embeddings/utils/ldf.py	0%	🟢
luxonis_ml/embeddings/utils/model.py	0%	🟢
luxonis_ml/embeddings/utils/qdrant.py	0%	🟢
TOTAL	0%	🟢

updated for commit: 7b63170 by action🐍

conorsim

Looks good now - the import issue is fixed

DrejcPesjak and others added 11 commits January 9, 2024 17:45

Renamed jupyter notebooks, WeaviateAPI added

4910797

WeaviateAPI tested, LDF new version code update

6338b33

Slight Weaviate emb. adjustments, written some examples, started on e…

f1df959

…mb. processing methods

Weaviate embeddings small bug fixes

adcec0f

Added abstract class VectorDBAPI, WeaviateAPI and QdrantAPI reworked

881c53f

Reworked for VectorDB, testing needed.

7f2b499

Reworked LDF-EMB examples and tested

6b7dbf5

Tmp data folder, string uuids in WeaviateAPI

fa03ec5

Removed PyTorch dependencies

c3fdac7

General properties added (for image_path), similarity metric bug fix.

3faf7a2

Added comments and documentation

5ece227

DrejcPesjak requested review from kozlov721, klemen1999 and conorsim February 7, 2024 01:02

kozlov721 reviewed Feb 7, 2024

View reviewed changes

Added type hints, merged some functions, and reworked requirements.txt

542e570

klemen1999 reviewed Feb 7, 2024

View reviewed changes

conorsim requested changes Feb 7, 2024

View reviewed changes

kozlov721 reviewed Feb 7, 2024

View reviewed changes

Pre-commit hook changes

25427a1

fixed pre-commit errors

9cc1902

add weaviate info to notebook

7b63170

conorsim approved these changes Mar 4, 2024

View reviewed changes

conorsim merged commit 7d44719 into dev Mar 4, 2024
12 checks passed

conorsim deleted the weaviate_emb branch March 4, 2024 15:59

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Weaviate embeddings rework #77

Weaviate embeddings rework #77

DrejcPesjak commented Feb 7, 2024

kozlov721 Feb 7, 2024

conorsim commented Feb 7, 2024

klemen1999 Feb 7, 2024

kozlov721 Feb 7, 2024

conorsim left a comment

conorsim Feb 7, 2024

conorsim Feb 7, 2024

conorsim Feb 7, 2024

kozlov721 Feb 7, 2024

kozlov721 Feb 7, 2024

DrejcPesjak commented Feb 10, 2024

kozlov721 commented Feb 12, 2024 •

edited

Loading

github-actions bot commented Feb 14, 2024 •

edited

Loading

github-actions bot commented Feb 14, 2024 •

edited

Loading

conorsim left a comment

	reference_embeddings: Union[str, List[str], List[List[float], np.ndarray]],
	reference_embeddings: Union[str, List[str], List[List[float]], np.ndarray]


		- For Weaviate-specific examples, refer to the provided code examples.
		- The `search_vectordb` function can be used with either Qdrant or Weaviate, depending on the provided `vectordb_api` object.

Weaviate embeddings rework #77

Weaviate embeddings rework #77

Conversation

DrejcPesjak commented Feb 7, 2024

Choose a reason for hiding this comment

conorsim commented Feb 7, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

conorsim left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

DrejcPesjak commented Feb 10, 2024

Pre-commit hooks passed, but pre-commit CI test still fails

kozlov721 commented Feb 12, 2024 • edited Loading

github-actions bot commented Feb 14, 2024 • edited Loading

Test Results

github-actions bot commented Feb 14, 2024 • edited Loading

☂️ Python Coverage

Overall Coverage

New Files

Modified Files

conorsim left a comment

Choose a reason for hiding this comment

kozlov721 commented Feb 12, 2024 •

edited

Loading

github-actions bot commented Feb 14, 2024 •

edited

Loading

github-actions bot commented Feb 14, 2024 •

edited

Loading