Bounce Classifier Model Files
==============================

These files are used by the @postalsys/bounce-classifier npm package
to classify SMTP bounce messages into 16 categories.

To use these model files, copy them into the model/ directory of the
bounce-classifier package, replacing the existing files.


Files required by bounce-classifier
------------------------------------

  vocab.json              Token vocabulary (word-to-index mapping).
                          Used to tokenize input text before inference.

  labels.json             Maps numeric model output indices to label
                          names (e.g. "user_unknown", "spam_blocked").

  group1-shard1of1.bin    Binary model weights (embedding matrix,
                          dense layer kernels and biases). This is
                          the core of the neural network.

  model.json              Keras model topology (layer structure,
                          activation functions, weight shapes). Used
                          to interpret the binary weights file.

  config.json             Model metadata (vocabulary size, embedding
                          dimensions, max input length, validation
                          accuracy from training).


Files NOT required by bounce-classifier
----------------------------------------

  keras_model.h5          Full Keras/TensorFlow model in HDF5 format.
                          Only needed if you want to continue training
                          or convert to other formats. Not loaded by
                          the classifier at runtime.


Usage
-----

  npm install @postalsys/bounce-classifier

  import { classify, initialize } from "@postalsys/bounce-classifier";
  await initialize();
  const result = await classify("550 5.1.1 User unknown");
  console.log(result.label);      // "user_unknown"
  console.log(result.action);     // "remove"
  console.log(result.confidence); // 0.95

More information: https://github.com/postalsys/bounce-classifier
