RNN Models

/**
 * LSTM designed for text generation. Can be trained on a corpus of text.
 * Architecture follows Keras LSTM text generation implementation.
 * Includes Walt Whitman pre-trained weights for generating text.
 */
class TextGenerationLSTM extends ZooModel {
    /**
     * Creates TextGenerationLSTM with basic configuration
     * @param numLabels Total number of unique characters in vocabulary
     * @param seed Random seed for reproducibility
     * @param iterations Number of training iterations
     */
    TextGenerationLSTM(int numLabels, long seed, int iterations);
    
    /**
     * Creates TextGenerationLSTM with workspace mode configuration
     * @param numLabels Total number of unique characters in vocabulary
     * @param seed Random seed for reproducibility
     * @param iterations Number of training iterations
     * @param workspaceMode Memory workspace configuration
     */
    TextGenerationLSTM(int numLabels, long seed, int iterations, WorkspaceMode workspaceMode);
    
    /**
     * Returns the LSTM network configuration
     * @return MultiLayerConfiguration for the text generation LSTM
     */
    MultiLayerConfiguration conf();
}

// Create LSTM for text generation with 47 unique characters
int vocabularySize = 47; // Total unique characters in your text corpus
TextGenerationLSTM lstm = new TextGenerationLSTM(vocabularySize, 42, 1);

// Initialize the model
MultiLayerNetwork model = (MultiLayerNetwork) lstm.init();

// Get model metadata
ModelMetaData metadata = lstm.metaData();
int[][] inputShape = metadata.getInputShape(); // [40, 47] (sequence length, vocab size)
ZooType type = metadata.getZooType(); // ZooType.RNN

// Custom configuration with different parameters
TextGenerationLSTM customLSTM = new TextGenerationLSTM(
    100, // vocabulary size for larger corpus
    123, // custom seed
    10,  // more iterations
    WorkspaceMode.SINGLE
);

MultiLayerConfiguration config = customLSTM.conf();

// 1. Prepare your text data
String corpus = "Your training text here...";
Map<Character, Integer> charToIndex = createCharacterIndex(corpus);
int vocabSize = charToIndex.size();

// 2. Create and train the model
TextGenerationLSTM lstm = new TextGenerationLSTM(vocabSize, 42, 100);
MultiLayerNetwork model = (MultiLayerNetwork) lstm.init();

// 3. Train on your text corpus (data preparation not shown)
// model.fit(trainingDataIterator);

// 4. Generate text (sampling logic not shown in API)
// String generatedText = generateText(model, seedText, charToIndex);

TextGenerationLSTM lstm = new TextGenerationLSTM(47, 42, 1);
ModelMetaData metadata = lstm.metaData();

// Input shape: [sequenceLength, vocabularySize]
int[][] inputShape = metadata.getInputShape(); // [40, 47]
int sequenceLength = inputShape[0][0]; // 40 characters per sequence
int vocabSize = inputShape[0][1]; // 47 unique characters

// Output: vocabulary-sized probability distribution
int numOutputs = metadata.getNumOutputs(); // 1 (single output per timestep)

// For memory-constrained environments
TextGenerationLSTM lstm = new TextGenerationLSTM(
    vocabSize, 
    42, 
    1, 
    WorkspaceMode.SEPARATE // Better for memory usage
);

// For performance-optimized training
TextGenerationLSTM lstm = new TextGenerationLSTM(
    vocabSize,
    42, 
    1,
    WorkspaceMode.SINGLE // Better for speed
);