🔍 Passthrough Features

Passthrough Features in KDP

Include pre-processed data and custom vectors in your model without additional transformations.

📋 Overview

Passthrough features allow you to include data in your model without any preprocessing modifications. They're perfect for pre-processed data, custom vectors, and scenarios where you need to preserve exact values.

⚡

Direct Integration

Include pre-processed data without modifications

🔢

Custom Vectors

Use pre-computed embeddings and vectors

📊

Raw Values

Preserve exact original values in your model

🔄

Flexible Integration

Seamlessly combine with other feature types

🚀 When to Use Passthrough Features

🔄

Pre-processed Data

You have already processed the data externally

🔢

Custom Vectors

You want to include pre-computed embeddings or vectors

📊

Raw Values

You need the exact original values in your model

🔍

Feature Testing

You want to compare raw vs processed feature performance

🔄

Gradual Migration

You're moving from a legacy system and need compatibility

💡 Defining Passthrough Features

from kdp import PreprocessingModel, FeatureType
from kdp.features import PassthroughFeature
import tensorflow as tf

# Simple approach using enum
features = {
    "embedding_vector": FeatureType.PASSTHROUGH,
    "age": FeatureType.FLOAT_NORMALIZED,
    "category": FeatureType.STRING_CATEGORICAL
}

# Advanced configuration with PassthroughFeature class
features = {
    "embedding_vector": PassthroughFeature(
        name="embedding_vector",
        dtype=tf.float32  # Specify the data type
    ),
    "raw_text_embedding": PassthroughFeature(
        name="raw_text_embedding",
        dtype=tf.float32
    ),
    "age": FeatureType.FLOAT_NORMALIZED,
    "category": FeatureType.STRING_CATEGORICAL
}

# Create your preprocessor
preprocessor = PreprocessingModel(
    path_data="customer_data.csv",
    features_specs=features
)

📊 How Passthrough Features Work

Passthrough features are included in model inputs without any transformations, maintaining their original values throughout the pipeline.

➕

Added to Inputs

Included in model inputs like other features

🔄

Type Casting

Cast to specified dtype for compatibility

⚡

No Transformation

Pass through without normalization or encoding

🎯

Feature Selection

Optional feature selection if enabled

🔧 Configuration Options

Parameter	Type	Description
name	str	The name of the feature
feature_type	FeatureType	Set to FeatureType.PASSTHROUGH by default
dtype	tf.DType	The data type of the feature (default: tf.float32)

🎯 Example: Using Pre-computed Embeddings

import pandas as pd
from kdp import PreprocessingModel, FeatureType
from kdp.features import PassthroughFeature, NumericalFeature
import tensorflow as tf

# Define features
features = {
    # Regular features
    "age": NumericalFeature(
        name="age",
        feature_type=FeatureType.FLOAT_NORMALIZED
    ),
    "category": FeatureType.STRING_CATEGORICAL,

    # Passthrough features for pre-computed embeddings
    "product_embedding": PassthroughFeature(
        name="product_embedding",
        dtype=tf.float32
    )
}

# Create your preprocessor
preprocessor = PreprocessingModel(
    path_data="data.csv",
    features_specs=features
)

# Build the model
model = preprocessor.build_preprocessor()

⚠️ Things to Consider

🔍

Data Type Compatibility

Ensure the data type of your passthrough feature is compatible with the overall model

📊

Dimensionality

Make sure the feature dimensions fit your model architecture

🧹

Data Quality

Since no preprocessing is applied, ensure your data is clean and ready for use

⚡

Performance Impact

Using raw data may affect model performance; test both approaches

🚀 Best Practices

📝

Document Your Decision

Make it clear why certain features are passed through

🔍

Test Both Approaches

Compare passthrough vs preprocessed features for performance

🎯

Feature Importance

Use feature selection to see if passthrough features contribute meaningfully

📈

Monitor Gradients

Watch for gradient issues since passthrough features may have different scales

← Cross Features Optimization →