Google KerasRS

Keras Recommenders (KerasRS) is a new library from Google for building advanced recommender systems with ease and speed

It is built on Keras 3 and supports multi-backend operation, working seamlessly with TensorFlow, JAX, or PyTorch

[{"selector":"#anim-c23fc27a-aba2-451e-9f6f-1cc78394311a [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-0c0879b1-7e5b-4290-b3da-4164f2bbd8ab","keyframes":{"transform":["translate3d(-118.79194%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-345c1234-de8a-479b-a7e2-3a1a4727f014","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-f6b6212a-2122-48d8-bc4b-0820a9e1c954","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

KerasRS provides specialized layers, losses, and metrics tailored for retrieval and ranking tasks in recommender systems

[{"selector":"#anim-d29d21f5-8754-4c32-8c4c-a8d1412fdeaa [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-801fbde9-48cb-46ad-bf38-d92dc0ccba6c","keyframes":{"transform":["translate3d(115.2381%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1f1bcdf1-5533-4570-bd43-b4af57f6d1d2","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-a311fa7e-4185-4397-af18-f5f39f81a01b","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Models built with KerasRS can be serialized and transferred between supported frameworks without costly migrations

[{"selector":"#anim-c5d9f48a-1a27-48b3-bb07-47f64663b6bd [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-eacb863a-2a4e-4cd8-bf21-aafc3aaa1a58","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-caf9e147-e0c3-4cb8-919e-2efdebc89d73","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

The library integrates naturally with standard Keras APIs, allowing use of familiar methods like model.compile() and model.fit()

[{"selector":"#anim-44befe1c-3ff6-434e-a1b6-39890cf2f627 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-f5f92fec-2c33-433e-abb2-19cd71b7794a","keyframes":{"transform":["translate3d(-115.55555%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-edf49dac-75f1-4b3c-82b3-3f4656e4b032","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-a4fe3e3f-7850-4ff8-8d60-d104e2267906","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

Installation is simple via pip, and the backend is selected by setting the KERAS_BACKEND environment variable before importing Keras

[{"selector":"#anim-428b2c21-afcd-4240-968e-e4e65f6a3df7 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-1714f656-7832-469d-8145-3477d9e391d8","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-c98f8077-a17c-4e5b-ad70-fa30725b79d5","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

KerasRS includes components such as BruteForceRetrieval for candidate retrieval and FeatureCross for feature engineering

[{"selector":"#anim-4a299282-865d-4fae-9398-63a59c008988 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-50e247b6-1e81-4af5-9d9c-e10c1485ed35","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-ed4b6f92-d973-4940-bf07-a5fd83c43d0c","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Example models include retrieval architectures and ranking models, with code samples provided in the documentation

[{"selector":"#anim-0ff2cc94-3f65-435b-aaae-e0f8ed2add44 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-de5f2615-3569-4d9b-8fb2-dd5d80e605cb","keyframes":{"transform":["translate3d(-119.34426%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-447e36b8-3baf-4fb1-b311-ba824b7b95ec","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-6afc30ba-2aca-48a7-ae72-11c2ba69ef93","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]

The library supports distributed embedding lookups using SparseCore chips on TPUs for large-scale recommendation tasks

[{"selector":"#anim-18239186-afe3-40a1-9a4f-313359f89efc [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-0919113c-1511-4058-9f2f-d2d9a39a004c","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-3bee1fb0-f372-4212-9b24-0fcbb11c186e","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

Regular updates will introduce popular model implementations, such as Deep & Cross Network (DCN) and two-tower models

[{"selector":"#anim-787a79f1-eb21-41af-89bb-417537e4c87d [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-d6a7f161-3a1f-400b-97a0-2f7bc409fe61","keyframes":{"opacity":[0,1]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-6d8925ea-dbd8-429c-8242-d762823d1e62","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

For more details visit Govindhtech.com

[{"selector":"#anim-de36e05f-5b1e-421d-8752-dd7b47470ea9 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]