renjunxiang/Text_Generate

Implements dual-framework LSTM architectures (Keras and TensorFlow) with 2-layer RNNs trained on classical Chinese poetry and novels, using sequence-to-sequence modeling where fixed-length input windows predict subsequent characters. Addresses practical generation challenges through probabilistic sampling from top-n predictions and punctuation-aware decoding to enforce natural line breaks in poetry, while optimizing for sparse categorical cross-entropy to reduce memory overhead on large vocabularies.

128 stars. No commits in the last 6 months.