feat: Add new layers and models for enhanced recommendation capabilities #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed

piotrlaczkowski wants to merge 12 commits into main from feat/feat_adding_reco_engines

README.md

-Original file line number
+Diff line change
@@ Expand Up / @@ -279,12 +279,11 @@ KMR is optimized for performance with: @@
     - **Memory Optimization**: Careful memory management in complex layers
     - **Batch Processing**: Optimized for batch operations
-    ## 🔮 Roadmap
+    ## 💬 Join Our Community
-    - [ ] **v0.3.0**: Additional model architectures and pre-trained models
-    - [ ] **v0.4.0**: Integration with popular ML frameworks
-    - [ ] **v0.5.0**: Model zoo with pre-trained weights
-    - [ ] **v1.0.0**: Production-ready with comprehensive benchmarks
+    Have questions or want to connect with other KDP users? Join us on Discord:
+    [![Discord](https://img.shields.io/badge/Discord-Join%20Us-7289DA?logo=discord&logoColor=white)](https://discord.gg/bhvGunkF)
     ## 📄 License
@@ Expand All @@
     - **Issues**: [GitHub Issues](https://github.com/UnicoLab/keras-model-registry/issues)
     - **Discussions**: [GitHub Discussions](https://github.com/UnicoLab/keras-model-registry/discussions)
     - **Documentation**: [Online Docs](https://unicolab.github.io/keras-model-registry/)
+    - **Discort**: [![Discord](https://img.shields.io/badge/Discord-Join%20Us-7289DA?logo=discord&logoColor=white)](https://discord.gg/bhvGunkF)
     ---
@@ Expand Down @@

docs/api/layers.md

-Original file line number
+Diff line change
@@ Expand Up @@
     Detects anomalies in categorical features.
     ::: kmr.layers.CategoricalAnomalyDetectionLayer
+    ## 📊 Recommendation Systems
+    ### 👥 CollaborativeUserItemEmbedding
+    Dual embedding lookup layer for collaborative filtering with separate user and item embedding tables.
+    ::: kmr.layers.CollaborativeUserItemEmbedding
+    ### 🏢 DeepFeatureTower
+    Dense neural network tower for processing user or item features in two-tower recommendation architectures.
+    ::: kmr.layers.DeepFeatureTower
+    ### 📏 NormalizedDotProductSimilarity
+    Compute normalized dot product (cosine) similarity between user and item representations.
+    ::: kmr.layers.NormalizedDotProductSimilarity
+    ### 🏆 TopKRecommendationSelector
+    Select top-K recommendation items based on scores with efficient heap-based selection.
+    ::: kmr.layers.TopKRecommendationSelector
+    ### 🔢 DynamicBatchIndexGenerator
+    Generate dynamic batch indices for grouping and indexing operations in recommendation systems.
+    ::: kmr.layers.DynamicBatchIndexGenerator
+    ### 📐 TensorDimensionExpander
+    Expand tensor dimensions for broadcasting and reshaping operations in recommendation pipelines.
+    ::: kmr.layers.TensorDimensionExpander
+    ### 🎭 ThresholdBasedMasking
+    Apply threshold-based masking to filter values in geospatial and recommendation systems.
+    ::: kmr.layers.ThresholdBasedMasking
+    ### 🌍 HaversineGeospatialDistance
+    Compute Haversine great-circle distance between geographic coordinates for location-based recommendations.
+    ::: kmr.layers.HaversineGeospatialDistance
+    ### 🗺️ SpatialFeatureClustering
+    Cluster spatial features into geographic regions for location-aware recommendation filtering.
+    ::: kmr.layers.SpatialFeatureClustering
+    ### 📍 GeospatialScoreRanking
+    Rank recommendations based on geospatial clustering features for location-aware recommendations.
+    ::: kmr.layers.GeospatialScoreRanking
+    ### 🧠 DeepFeatureRanking
+    Deep neural network tower for feature-based ranking in learning-to-rank models.
+    ::: kmr.layers.DeepFeatureRanking
+    ### ⚖️ LearnableWeightedCombination
+    Combine multiple recommendation scores with learnable softmax-normalized weights for hybrid recommendations.
+    ::: kmr.layers.LearnableWeightedCombination
+    ### 🔍 CosineSimilarityExplainer
+    Compute and explain cosine similarity between embeddings for interpretable recommendations.
+    ::: kmr.layers.CosineSimilarityExplainer
+    ### 💬 FeedbackAdjustmentLayer
+    Adjust recommendation scores based on user feedback signals for adaptive recommendations.
+    ::: kmr.layers.FeedbackAdjustmentLayer

docs/api/models.md

-Original file line number
+Diff line change
@@ Expand Up @@
     ::: kmr.models.autoencoder.Autoencoder
+    ## 📊 Recommendation Systems
+    ### 🗺️ GeospatialClusteringModel
+    Unsupervised geospatial clustering recommendation model using distance-based clustering and spatial ranking.
+    ::: kmr.models.GeospatialClusteringModel
+    **Key Features:**
+    - Haversine distance calculation for geographic coordinates
+    - Spatial feature clustering into geographic regions
+    - Geospatial score ranking based on proximity
+    - Unsupervised learning with entropy and variance losses
+    - Configurable training mode (supervised/unsupervised)
+    ### 📈 MatrixFactorizationModel
+    Matrix factorization recommendation model using collaborative filtering with user and item embeddings.
+    ::: kmr.models.MatrixFactorizationModel
+    **Key Features:**
+    - Dual user-item embedding lookups
+    - Normalized dot product similarity computation
+    - Top-K recommendation selection
+    - L2 regularization on embeddings
+    - Scalable to millions of users/items
+    ### 🏗️ TwoTowerModel
+    Two-tower recommendation model with separate towers for user and item features.
+    ::: kmr.models.TwoTowerModel
+    **Key Features:**
+    - Separate deep feature towers for users and items
+    - Normalized dot product similarity between towers
+    - Content-based feature processing
+    - Batch normalization and dropout for regularization
+    - Efficient similarity computation
+    ### 🧠 DeepRankingModel
+    Deep neural network ranking model for learning-to-rank recommendations.
+    ::: kmr.models.DeepRankingModel
+    **Key Features:**
+    - Deep feature ranking with multiple dense layers
+    - Combined user-item feature processing
+    - Batch normalization and dropout
+    - Learning-to-rank optimization
+    - Complex non-linear ranking functions
+    ### 🤝 UnifiedRecommendationModel
+    Unified recommendation model combining collaborative filtering, content-based, and hybrid approaches.
+    ::: kmr.models.UnifiedRecommendationModel
+    **Key Features:**
+    - Multiple recommendation components (CF, CB, Hybrid)
+    - Score combination with learnable weights
+    - Flexible architecture for different data types
+    - End-to-end learning of optimal combination
+    - Production-ready hybrid system
+    ### 🔍 ExplainableRecommendationModel
+    Explainable recommendation model with similarity explanations and feedback adjustment.
+    ::: kmr.models.ExplainableRecommendationModel
+    **Key Features:**
+    - Cosine similarity explanations for transparency
+    - User feedback integration
+    - Interpretable similarity scores
+    - Feedback-aware score adjustment
+    - Transparent recommendation reasoning
+    ### 🎯 ExplainableUnifiedRecommendationModel
+    Explainable unified recommendation model combining multiple approaches with transparency features.
+    ::: kmr.models.ExplainableUnifiedRecommendationModel
+    **Key Features:**
+    - Multiple recommendation components with explanations
+    - Component-level similarity scores
+    - Transparent weight learning
+    - Explainable hybrid recommendations
+    - Full interpretability across all components
     ## 🔧 Base Classes
     ### 🏛️ BaseModel
@@ Expand Down @@

docs/index.md

            
                      Original file line number
                      Diff line number
                      Diff line change
                  
    @@ -9,7 +9,7 @@
  
    </div>

    !!! success "🎯 Production-Ready Tabular AI"

        Build sophisticated tabular models with **38+ specialized layers**, **smart preprocessing**, and **intelligent feature engineering** - all designed exclusively for Keras 3.

        Build sophisticated tabular models with **50+ specialized layers**, **smart preprocessing**, **intelligent feature engineering**, and **recommendation systems** - all designed exclusively for Keras 3.

    ---

    @@ -20,7 +20,8 @@ KMR (Keras Model Registry) is a comprehensive collection of **production-ready l
  
    - 🧠 **Advanced Attention Mechanisms** for tabular data

    - 🔧 **Feature Engineering Layers** for data preprocessing  

    - 🏗️ **Pre-built Models** for common ML tasks

    - 📊 **Data Analysis Tools** for intelligent layer recommendations

    - 📊 **Recommendation Systems** with collaborative filtering, content-based, and geospatial models

    - 📈 **Data Analysis Tools** for intelligent layer recommendations

    - ⚡ **Keras 3 Native** - No TensorFlow dependencies in production code

    !!! tip "Why KMR?"

    @@ -33,9 +34,9 @@ KMR (Keras Model Registry) is a comprehensive collection of **production-ready l
  
    <div class="grid cards" markdown>

    - **38+ Production Layers**

    - **50+ Production Layers**

        Advanced attention mechanisms, feature processing, and specialized architectures ready for production use.

        Advanced attention mechanisms, feature processing, recommendation systems, and specialized architectures ready for production use.

        [Explore All Layers →](api/layers.md){ .md-button .md-button--primary }

    @@ -198,29 +199,31 @@ KMR (Keras Model Registry) is a comprehensive collection of **production-ready l
  
    === "🛒 E-commerce Recommendations"

        Build user-item interaction models:

        Build recommendation systems with collaborative filtering and content-based features:

        ```python

        from kmr.layers import TabularAttention, GatedFeatureFusion

        user_features = keras.Input(shape=(20,))

        item_features = keras.Input(shape=(15,))

        user_repr = TabularAttention(

            num_heads=4, head_dim=16

        )(keras.layers.Concatenate()(

            [user_features, item_features]

        ))

        fused = GatedFeatureFusion()([user_repr, item_features])

        compatibility = keras.layers.Dense(1)(fused)

        model = keras.Model(

            inputs=[user_features, item_features],

            outputs=compatibility

        from kmr.models import MatrixFactorizationModel, TwoTowerModel

        # Option 1: Collaborative Filtering

        model = MatrixFactorizationModel(

            num_users=10000,

            num_items=5000,

            embedding_dim=64,

            top_k=10

        )

        # Option 2: Content-Based (Two-Tower)

        model = TwoTowerModel(

            user_feature_dim=20,

            item_feature_dim=15,

            output_dim=64,

            top_k=10

        )

        model.compile(optimizer='adam', loss='binary_crossentropy')

        ```

        **Use case:** Product recommendations, CTR prediction, customer lifetime value

        **Use case:** Product recommendations, CTR prediction, customer lifetime value, personalized search

    ---

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: Add new layers and models for enhanced recommendation capabilities #5

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!

Uh oh!