A simple text classifier model in Tensorflow with code quality and transfer learning.
This model uses transfer learning for text encoding, you should download (or reference to an s3 or gs endpoint) an encoder, like universal-sentence-encoder
The startup command is train-classifier.
--dataset-path: Path to CSV source dataset--epochs: number of training epochs--encoder-uri: Uri or path for encoder saved_model folder--layer-sizes: comma separated list of integers, each integer is the number of neurons in that label--encoder-features: Number of features resulting from encoding (in universal-sentence-encoder, this number is 512)--number-of-classes: Number of distinct labels--learning-rate: Optimizer learning rate--save-path: Output path where model, chekpoints and metrics will be saved--feature-column: Name of csv column with feature texts--label-column: Name of csv column with integer numeric labels
You can see more about this model in this youtube video