CtrlK
BlogDocsLog inGet started
Tessl Logo

tessl/maven-org-deeplearning4j--deeplearning4j-modelimport

Keras model import functionality for DeepLearning4J

—

Pending
Overview
Eval results
Files

separate-files-import.mddocs/

Separate Files Import

Import models where configuration (JSON) and weights (HDF5) are stored in separate files. This is a common workflow in Keras where models are saved using model.to_json() and model.save_weights().

Functional API Separate Files Import

Import Keras Functional API models from separate JSON configuration and HDF5 weights files.

public static ComputationGraph importKerasModelAndWeights(String modelJsonFilename, String weightsHdf5Filename)
    throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException;

public static ComputationGraph importKerasModelAndWeights(String modelJsonFilename, String weightsHdf5Filename, boolean enforceTrainingConfig)
    throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException;

Parameters

  • modelJsonFilename (String): Path to JSON file containing model configuration
  • weightsHdf5Filename (String): Path to HDF5 file containing model weights
  • enforceTrainingConfig (boolean): Whether to enforce training-related configurations

Returns

  • ComputationGraph: DeepLearning4J computation graph with imported configuration and weights

Exceptions

  • IOException: File I/O errors when reading JSON or HDF5 files
  • InvalidKerasConfigurationException: Malformed or invalid Keras model configuration
  • UnsupportedKerasConfigurationException: Keras features not supported by DeepLearning4J

Usage Examples

import org.deeplearning4j.nn.modelimport.keras.KerasModelImport;
import org.deeplearning4j.nn.graph.ComputationGraph;

// Import from separate files with default enforcement
ComputationGraph model = KerasModelImport.importKerasModelAndWeights(
    "model_architecture.json",
    "model_weights.h5"
);

// Import with relaxed training config enforcement
ComputationGraph model = KerasModelImport.importKerasModelAndWeights(
    "model_architecture.json", 
    "model_weights.h5",
    false
);

// Use the imported model
INDArray input = Nd4j.randn(1, 224, 224, 3);
INDArray output = model.outputSingle(input);

Sequential Separate Files Import

Import Keras Sequential models from separate JSON configuration and HDF5 weights files.

public static MultiLayerNetwork importKerasSequentialModelAndWeights(String modelJsonFilename, String weightsHdf5Filename)
    throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException;

public static MultiLayerNetwork importKerasSequentialModelAndWeights(String modelJsonFilename, String weightsHdf5Filename, boolean enforceTrainingConfig)
    throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException;

Parameters

  • modelJsonFilename (String): Path to JSON file containing Sequential model configuration
  • weightsHdf5Filename (String): Path to HDF5 file containing model weights
  • enforceTrainingConfig (boolean): Whether to enforce training-related configurations

Returns

  • MultiLayerNetwork: DeepLearning4J multi-layer network with imported configuration and weights

Exceptions

  • IOException: File I/O errors when reading JSON or HDF5 files
  • InvalidKerasConfigurationException: Malformed or invalid Keras model configuration
  • UnsupportedKerasConfigurationException: Keras features not supported by DeepLearning4J

Usage Examples

import org.deeplearning4j.nn.modelimport.keras.KerasModelImport;
import org.deeplearning4j.nn.multilayer.MultiLayerNetwork;

// Import Sequential model from separate files
MultiLayerNetwork model = KerasModelImport.importKerasSequentialModelAndWeights(
    "sequential_architecture.json",
    "sequential_weights.h5"
);

// Use the imported model
INDArray input = Nd4j.randn(1, 784); // Example for MNIST
INDArray output = model.output(input);

Creating Separate Files in Keras

This workflow corresponds to the following Python/Keras code:

Functional API Model

import keras
import json

# Create or load your model
model = keras.models.Model(inputs=inputs, outputs=outputs)

# Save architecture to JSON
model_json = model.to_json()
with open('model_architecture.json', 'w') as json_file:
    json_file.write(model_json)

# Save weights to HDF5
model.save_weights('model_weights.h5')

Sequential Model

import keras
import json

# Create or load your Sequential model
model = keras.models.Sequential([...])

# Save architecture to JSON
model_json = model.to_json()
with open('sequential_architecture.json', 'w') as json_file:
    json_file.write(model_json)

# Save weights to HDF5
model.save_weights('sequential_weights.h5')

Advantages of Separate Files

Version Control Friendly

  • JSON configuration files are text-based and can be version controlled effectively
  • Binary weight files can be stored separately using Git LFS or other solutions
  • Architecture changes are easily tracked in version control

Flexibility

  • Same architecture can be used with different weight files
  • Easy to share model architectures without weights
  • Supports different weight formats or sources

Memory Efficiency

  • Load configuration first to validate compatibility
  • Load weights only when needed
  • Useful for model analysis without loading large weight files

File Format Requirements

JSON Configuration File

Must contain valid Keras model configuration in JSON format:

{
  "class_name": "Model" | "Sequential",
  "config": {
    // Model-specific configuration
  }
}

HDF5 Weights File

Must contain model weights in HDF5 format with:

  • Layer-specific weight groups
  • Proper weight naming conventions matching layer names
  • Compatible data types and shapes

Common Use Cases

Model Deployment Pipeline

// Step 1: Validate architecture
ComputationGraphConfiguration config = KerasModelImport.importKerasModelConfiguration("model.json");

// Step 2: Load full model with weights
ComputationGraph model = KerasModelImport.importKerasModelAndWeights("model.json", "weights.h5");

// Step 3: Deploy for inference
INDArray prediction = model.outputSingle(inputData);

Model Versioning

// Load same architecture with different weight versions
String architecture = "resnet50_architecture.json";

// Load v1 weights
ComputationGraph modelV1 = KerasModelImport.importKerasModelAndWeights(architecture, "weights_v1.h5");

// Load v2 weights  
ComputationGraph modelV2 = KerasModelImport.importKerasModelAndWeights(architecture, "weights_v2.h5");

// Compare performance or ensemble predictions

Transfer Learning Setup

// Import pre-trained model architecture
ComputationGraphConfiguration baseConfig = KerasModelImport.importKerasModelConfiguration("base_model.json");

// Create model without pre-trained weights for custom training
ComputationGraph customModel = new ComputationGraph(baseConfig);
customModel.init();

// Or load pre-trained weights as starting point
ComputationGraph pretrainedModel = KerasModelImport.importKerasModelAndWeights("base_model.json", "pretrained_weights.h5");

Error Handling

File Path Validation

import java.nio.file.Paths;
import java.nio.file.Files;

public static boolean validateFiles(String jsonPath, String weightsPath) {
    return Files.exists(Paths.get(jsonPath)) && Files.exists(Paths.get(weightsPath));
}

// Usage
if (validateFiles("model.json", "weights.h5")) {
    ComputationGraph model = KerasModelImport.importKerasModelAndWeights("model.json", "weights.h5");
} else {
    System.err.println("One or both files do not exist");
}

Weight Compatibility Check

try {
    // Try with strict enforcement first
    ComputationGraph model = KerasModelImport.importKerasModelAndWeights("model.json", "weights.h5", true);
} catch (UnsupportedKerasConfigurationException e) {
    System.out.println("Warning: " + e.getMessage());
    // Fall back to relaxed enforcement
    ComputationGraph model = KerasModelImport.importKerasModelAndWeights("model.json", "weights.h5", false);
}

Install with Tessl CLI

npx tessl i tessl/maven-org-deeplearning4j--deeplearning4j-modelimport

docs

builder-pattern.md

configuration-import.md

index.md

layer-support.md

model-import.md

pretrained-models.md

separate-files-import.md

tile.json