danielegrattarola/twitter-sentiment-cnn

Implements a single-layer CNN with parallel 1D convolutions across configurable filter sizes, word embeddings, and dropout regularization, with TensorBoard integration for visualizing the computation graph. Includes automated dataset preprocessing (CSV parsing, vocabulary building), session checkpointing for model persistence, and protobuf export capabilities. Provides fine-grained control via command-line flags for architecture parameters (embedding size, filter counts, batch size) and training dynamics (epochs, validation frequency), enabling experimentation across different configurations.

158 stars. No commits in the last 6 months.