Fix cosine normalization

[kacperlukawski]

This PR enables the normalization of vectors if cosine distance is used as reported in #212. It also enables vector comparison in tests, as the previous scroll call did not return vectors. In addition to that, the vectors are compared with atol=1e-3 as random vectors are rounded upon generation:

qdrant-client/tests/fixtures/points.py

Line 15 in bc08733

return np.random.random(vector_sizes).round(3).tolist()

[netlify]

✅ Deploy Preview for poetic-froyo-8baba7 ready!

Name	Link
🔨 Latest commit	66f8510
🔍 Latest deploy log	https://app.netlify.com/sites/poetic-froyo-8baba7/deploys/64bc31a1e9250c0008451f4f
😎 Deploy Preview	https://deploy-preview-213--poetic-froyo-8baba7.netlify.app/qdrant_client.conversions.conversion
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[poutyface]

eps is needed if all vectors are zero

[joein]

fair point

but why would anyone like to store zeroed vectors?

[joein]

#236

[poutyface]

That's true. I found it when I was inputting the raw sensor data. But it might happen when you want to use high dimensional learned latent vectors (Black images, or vq-vae encoder vector)